feat: add timeline API with schedule-aware enrollment aggregation

Implements POST /api/timeline endpoint that aggregates enrollment by subject over 15-minute slots, filtering courses by their actual meeting times. Includes ISR-style schedule cache with hourly background refresh using stale-while-revalidate pattern, database indexes for efficient queries, and frontend refactor to dynamically discover subjects from API.
fix: instructor/course mismatching, build order-independent map for association
2026-01-31 08:23:35 -06:00 · 2026-01-30 10:56:11 -06:00 · 2026-01-30 09:53:03 -06:00 · 2026-01-30 04:08:16 -06:00 · 2026-01-30 04:07:53 -06:00 · 2026-01-30 03:46:48 -06:00
154 changed files with 19725 additions and 2331 deletions
@@ -0,0 +1,2 @@
+[env]
+TS_RS_EXPORT_DIR = { value = "web/src/lib/bindings/", relative = true }
@@ -1,5 +1,6 @@
 .env
 /target
-/go/
-.cargo/config.toml
-src/scraper/README.md
+
+# ts-rs bindings
+web/src/lib/bindings/**/*.ts
+!web/src/lib/bindings/index.ts
@@ -26,6 +26,21 @@ dependencies = [
 "memchr",
 ]

+[[package]]
+name = "alloc-no-stdlib"
+version = "2.0.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "cc7bb162ec39d46ab1ca8c77bf72e890535becd1751bb45f64c597edb4c8c6b3"
+
+[[package]]
+name = "alloc-stdlib"
+version = "0.2.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "94fb8275041c72129eb51b7d0322c29b8387a0386127718b096429201a5d6ece"
+dependencies = [
+ "alloc-no-stdlib",
+]
+
 [[package]]
 name = "allocator-api2"
 version = "0.2.21"
@@ -106,6 +121,19 @@ dependencies = [
 "serde",
 ]

+[[package]]
+name = "async-compression"
+version = "0.4.33"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "93c1f86859c1af3d514fa19e8323147ff10ea98684e6c7b307912509f50e67b2"
+dependencies = [
+ "compression-codecs",
+ "compression-core",
+ "futures-core",
+ "pin-project-lite",
+ "tokio",
+]
+
 [[package]]
 name = "async-trait"
 version = "0.1.89"
@@ -149,11 +177,12 @@ checksum = "c08606f8c3cbf4ce6ec8e28fb0014a2c086708fe954eaa885384a6165172e7e8"

 [[package]]
 name = "axum"
-version = "0.8.4"
+version = "0.8.8"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "021e862c184ae977658b36c4500f7feac3221ca5da43e3f25bd04ab6c79a29b5"
+checksum = "8b52af3cb4058c895d37317bb27508dccc8e5f2d39454016b297bf4a400597b8"
 dependencies = [
 "axum-core",
+ "base64 0.22.1",
 "bytes",
 "form_urlencoded",
 "futures-util",
@@ -168,13 +197,14 @@ dependencies = [
 "mime",
 "percent-encoding",
 "pin-project-lite",
- "rustversion",
- "serde",
+ "serde_core",
 "serde_json",
 "serde_path_to_error",
 "serde_urlencoded",
+ "sha1",
 "sync_wrapper 1.0.2",
 "tokio",
+ "tokio-tungstenite 0.28.0",
 "tower",
 "tower-layer",
 "tower-service",
@@ -183,9 +213,9 @@ dependencies = [

 [[package]]
 name = "axum-core"
-version = "0.5.2"
+version = "0.5.6"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "68464cd0412f486726fb3373129ef5d2993f90c34bc2bc1c1e9943b2f4fc7ca6"
+checksum = "08c78f31d7b1291f7ee735c1c6780ccde7785daae9a9206026862dab7d8792d1"
 dependencies = [
 "bytes",
 "futures-core",
@@ -194,13 +224,37 @@ dependencies = [
 "http-body-util",
 "mime",
 "pin-project-lite",
- "rustversion",
 "sync_wrapper 1.0.2",
 "tower-layer",
 "tower-service",
 "tracing",
 ]

+[[package]]
+name = "axum-extra"
+version = "0.12.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "fef252edff26ddba56bbcdf2ee3307b8129acb86f5749b68990c168a6fcc9c76"
+dependencies = [
+ "axum",
+ "axum-core",
+ "bytes",
+ "form_urlencoded",
+ "futures-core",
+ "futures-util",
+ "http 1.3.1",
+ "http-body 1.0.1",
+ "http-body-util",
+ "mime",
+ "pin-project-lite",
+ "serde_core",
+ "serde_html_form",
+ "serde_path_to_error",
+ "tower-layer",
+ "tower-service",
+ "tracing",
+]
+
 [[package]]
 name = "backtrace"
 version = "0.3.75"
@@ -218,13 +272,15 @@ dependencies = [

 [[package]]
 name = "banner"
-version = "0.3.4"
+version = "0.5.0"
 dependencies = [
 "anyhow",
 "async-trait",
 "axum",
+ "axum-extra",
 "bitflags 2.9.4",
 "chrono",
+ "chrono-tz",
 "clap",
 "compile-time",
 "cookie",
@@ -235,6 +291,7 @@ dependencies = [
 "fundu",
 "futures",
 "governor",
+ "html-escape",
 "http 1.3.1",
 "mime_guess",
 "num-format",
@@ -257,7 +314,9 @@ dependencies = [
 "tower-http",
 "tracing",
 "tracing-subscriber",
+ "ts-rs",
 "url",
+ "urlencoding",
 "yansi",
 ]

@@ -303,6 +362,27 @@ dependencies = [
 "generic-array",
 ]

+[[package]]
+name = "brotli"
+version = "8.0.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "4bd8b9603c7aa97359dbd97ecf258968c95f3adddd6db2f7e7a5bef101c84560"
+dependencies = [
+ "alloc-no-stdlib",
+ "alloc-stdlib",
+ "brotli-decompressor",
+]
+
+[[package]]
+name = "brotli-decompressor"
+version = "5.0.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "874bb8112abecc98cbd6d81ea4fa7e94fb9449648c93cc89aa40c81c24d7de03"
+dependencies = [
+ "alloc-no-stdlib",
+ "alloc-stdlib",
+]
+
 [[package]]
 name = "bstr"
 version = "1.12.0"
@@ -380,6 +460,8 @@ version = "1.2.34"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "42bc4aea80032b7bf409b0bc7ccad88853858911b7713a8062fdc0623867bedc"
 dependencies = [
+ "jobserver",
+ "libc",
 "shlex",
 ]

@@ -403,6 +485,16 @@ dependencies = [
 "windows-link 0.2.0",
 ]

+[[package]]
+name = "chrono-tz"
+version = "0.10.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "a6139a8597ed92cf816dfb33f5dd6cf0bb93a6adc938f11039f371bc5bcd26c3"
+dependencies = [
+ "chrono",
+ "phf",
+]
+
 [[package]]
 name = "clap"
 version = "4.5.47"
@@ -474,6 +566,26 @@ dependencies = [
 "time",
 ]

+[[package]]
+name = "compression-codecs"
+version = "0.4.32"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "680dc087785c5230f8e8843e2e57ac7c1c90488b6a91b88caa265410568f441b"
+dependencies = [
+ "brotli",
+ "compression-core",
+ "flate2",
+ "memchr",
+ "zstd",
+ "zstd-safe",
+]
+
+[[package]]
+name = "compression-core"
+version = "0.4.31"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "75984efb6ed102a0d42db99afb6c1948f0380d1d91808d5529916e6c08b49d8d"
+
 [[package]]
 name = "concurrent-queue"
 version = "2.5.0"
@@ -1227,6 +1339,15 @@ dependencies = [
 "windows-sys 0.59.0",
 ]

+[[package]]
+name = "html-escape"
+version = "0.2.13"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "6d1ad449764d627e22bfd7cd5e8868264fc9236e07c752972b4080cd351cb476"
+dependencies = [
+ "utf8-width",
+]
+
 [[package]]
 name = "http"
 version = "0.2.12"
@@ -1606,6 +1727,16 @@ version = "1.0.15"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "4a5f13b858c8d314ee3e8f639011f7ccefe71f97f96e50151fb991f267928e2c"

+[[package]]
+name = "jobserver"
+version = "0.1.34"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "9afb3de4395d6b3e67a780b6de64b51c978ecf11cb9a462c66be7d4ca9039d33"
+dependencies = [
+ "getrandom 0.3.3",
+ "libc",
+]
+
 [[package]]
 name = "js-sys"
 version = "0.3.77"
@@ -2007,6 +2138,24 @@ version = "2.3.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "9b4f627cb1b25917193a259e49bdad08f671f8d9708acfd5fe0a8c1455d87220"

+[[package]]
+name = "phf"
+version = "0.12.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "913273894cec178f401a31ec4b656318d95473527be05c0752cc41cdc32be8b7"
+dependencies = [
+ "phf_shared",
+]
+
+[[package]]
+name = "phf_shared"
+version = "0.12.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "06005508882fb681fd97892ecff4b7fd0fee13ef1aa569f8695dae7ab9099981"
+dependencies = [
+ "siphasher",
+]
+
 [[package]]
 name = "pin-project-lite"
 version = "0.2.16"
@@ -2677,9 +2826,19 @@ dependencies = [

 [[package]]
 name = "serde"
-version = "1.0.219"
+version = "1.0.228"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "5f0e2c6ed6606019b4e29e69dbaba95b11854410e5347d525002456dbbb786b6"
+checksum = "9a8e94ea7f378bd32cbbd37198a4a91436180c5bb472411e48b5ec2e2124ae9e"
+dependencies = [
+ "serde_core",
+ "serde_derive",
+]
+
+[[package]]
+name = "serde_core"
+version = "1.0.228"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "41d385c7d4ca58e59fc732af25c3983b67ac852c1a25000afe1175de458b67ad"
 dependencies = [
 "serde_derive",
 ]
@@ -2695,15 +2854,28 @@ dependencies = [

 [[package]]
 name = "serde_derive"
-version = "1.0.219"
+version = "1.0.228"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "5b0276cf7f2c73365f7157c8123c21cd9a50fbbd844757af28ca1f5925fc2a00"
+checksum = "d540f220d3187173da220f885ab66608367b6574e925011a9353e4badda91d79"
 dependencies = [
 "proc-macro2",
 "quote",
 "syn 2.0.106",
 ]

+[[package]]
+name = "serde_html_form"
+version = "0.2.8"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "b2f2d7ff8a2140333718bb329f5c40fc5f0865b84c426183ce14c97d2ab8154f"
+dependencies = [
+ "form_urlencoded",
+ "indexmap",
+ "itoa",
+ "ryu",
+ "serde_core",
+]
+
 [[package]]
 name = "serde_json"
 version = "1.0.143"
@@ -2776,7 +2948,7 @@ dependencies = [
 "static_assertions",
 "time",
 "tokio",
- "tokio-tungstenite",
+ "tokio-tungstenite 0.21.0",
 "tracing",
 "typemap_rev",
 "typesize",
@@ -2840,6 +3012,12 @@ dependencies = [
 "rand_core 0.6.4",
 ]

+[[package]]
+name = "siphasher"
+version = "1.0.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "b2aa850e253778c88a04c3d7323b043aeda9d3e30d5971937c1855769763678e"
+
 [[package]]
 name = "skeptic"
 version = "0.13.7"
@@ -3256,6 +3434,15 @@ dependencies = [
 "windows-sys 0.60.2",
 ]

+[[package]]
+name = "termcolor"
+version = "1.4.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "06794f8f6c5c898b3275aebefa6b8a1cb24cd2c6c79397ab15774837a0bc5755"
+dependencies = [
+ "winapi-util",
+]
+
 [[package]]
 name = "thiserror"
 version = "1.0.69"
@@ -3455,10 +3642,22 @@ dependencies = [
 "rustls-pki-types",
 "tokio",
 "tokio-rustls 0.25.0",
- "tungstenite",
+ "tungstenite 0.21.0",
 "webpki-roots 0.26.11",
 ]

+[[package]]
+name = "tokio-tungstenite"
+version = "0.28.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "d25a406cddcc431a75d3d9afc6a7c0f7428d4891dd973e4d54c56b46127bf857"
+dependencies = [
+ "futures-util",
+ "log",
+ "tokio",
+ "tungstenite 0.28.0",
+]
+
 [[package]]
 name = "tokio-util"
 version = "0.7.16"
@@ -3535,14 +3734,17 @@ version = "0.6.6"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "adc82fd73de2a9722ac5da747f12383d2bfdb93591ee6c58486e0097890f05f2"
 dependencies = [
+ "async-compression",
 "bitflags 2.9.4",
 "bytes",
+ "futures-core",
 "futures-util",
 "http 1.3.1",
 "http-body 1.0.1",
 "iri-string",
 "pin-project-lite",
 "tokio",
+ "tokio-util",
 "tower",
 "tower-layer",
 "tower-service",
@@ -3648,6 +3850,30 @@ version = "0.2.5"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "e421abadd41a4225275504ea4d6566923418b7f05506fbc9c0fe86ba7396114b"

+[[package]]
+name = "ts-rs"
+version = "11.1.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "4994acea2522cd2b3b85c1d9529a55991e3ad5e25cdcd3de9d505972c4379424"
+dependencies = [
+ "chrono",
+ "serde_json",
+ "thiserror 2.0.16",
+ "ts-rs-macros",
+]
+
+[[package]]
+name = "ts-rs-macros"
+version = "11.1.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "ee6ff59666c9cbaec3533964505d39154dc4e0a56151fdea30a09ed0301f62e2"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "syn 2.0.106",
+ "termcolor",
+]
+
 [[package]]
 name = "tungstenite"
 version = "0.21.0"
@@ -3669,6 +3895,23 @@ dependencies = [
 "utf-8",
 ]

+[[package]]
+name = "tungstenite"
+version = "0.28.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "8628dcc84e5a09eb3d8423d6cb682965dea9133204e8fb3efee74c2a0c259442"
+dependencies = [
+ "bytes",
+ "data-encoding",
+ "http 1.3.1",
+ "httparse",
+ "log",
+ "rand 0.9.2",
+ "sha1",
+ "thiserror 2.0.16",
+ "utf-8",
+]
+
 [[package]]
 name = "typemap_rev"
 version = "0.3.0"
@@ -3770,12 +4013,24 @@ dependencies = [
 "serde",
 ]

+[[package]]
+name = "urlencoding"
+version = "2.1.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "daf8dba3b7eb870caf1ddeed7bc9d2a049f3cfdfae7cb521b087cc33ae4c49da"
+
 [[package]]
 name = "utf-8"
 version = "0.7.6"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "09cc8ee72d2a9becf2f2febe0205bbed8fc6615b7cb429ad062dc7b7ddd036a9"

+[[package]]
+name = "utf8-width"
+version = "0.1.8"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "1292c0d970b54115d14f2492fe0170adf21d68a1de108eebc51c1df4f346a091"
+
 [[package]]
 name = "utf8_iter"
 version = "1.0.4"
@@ -4459,3 +4714,31 @@ dependencies = [
 "quote",
 "syn 2.0.106",
 ]
+
+[[package]]
+name = "zstd"
+version = "0.13.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "e91ee311a569c327171651566e07972200e76fcfe2242a4fa446149a3881c08a"
+dependencies = [
+ "zstd-safe",
+]
+
+[[package]]
+name = "zstd-safe"
+version = "7.2.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "8f49c4d5f0abb602a93fb8736af2a4f4dd9512e36f7f570d66e65ff867ed3b9d"
+dependencies = [
+ "zstd-sys",
+]
+
+[[package]]
+name = "zstd-sys"
+version = "2.0.16+zstd.1.5.7"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "91e19ebc2adc8f83e43039e79776e3fda8ca919132d68a1fed6a5faca2683748"
+dependencies = [
+ "cc",
+ "pkg-config",
+]
@@ -1,6 +1,6 @@
 [package]
 name = "banner"
-version = "0.3.4"
+version = "0.5.0"
 edition = "2024"
 default-run = "banner"

@@ -11,7 +11,7 @@ embed-assets = ["dep:rust-embed", "dep:mime_guess"]
 [dependencies]
 anyhow = "1.0.99"
 async-trait = "0.1"
-axum = "0.8.4"
+axum = { version = "0.8.4", features = ["ws"] }
 bitflags = { version = "2.9.4", features = ["serde"] }
 chrono = { version = "0.4.42", features = ["serde"] }
 compile-time = "0.2.0"
@@ -48,13 +48,18 @@ url = "2.5"
 governor = "0.10.1"
 serde_path_to_error = "0.1.17"
 num-format = "0.4.4"
-tower-http = { version = "0.6.0", features = ["cors", "trace", "timeout"] }
+tower-http = { version = "0.6.0", features = ["cors", "trace", "timeout", "compression-full"] }
 rust-embed = { version = "8.0", features = ["include-exclude"], optional = true }
 mime_guess = { version = "2.0", optional = true }
 clap = { version = "4.5", features = ["derive"] }
 rapidhash = "4.1.0"
 yansi = "1.0.1"
 extension-traits = "2"
+ts-rs = { version = "11.1.0", features = ["chrono-impl", "serde-compat", "serde-json-impl", "no-serde-warnings"] }
+html-escape = "0.2.13"
+axum-extra = { version = "0.12.5", features = ["query"] }
+urlencoding = "2.1.3"
+chrono-tz = "0.10.4"

 [dev-dependencies]

@@ -7,6 +7,9 @@ FROM oven/bun:1 AS frontend-builder

 WORKDIR /app

+# Install zstd for pre-compression
+RUN apt-get update && apt-get install -y --no-install-recommends zstd && rm -rf /var/lib/apt/lists/*
+
 # Copy backend Cargo.toml for build-time version retrieval
 COPY ./Cargo.toml ./

@@ -19,8 +22,8 @@ RUN bun install --frozen-lockfile
 # Copy frontend source code
 COPY ./web ./

-# Build frontend
-RUN bun run build
+# Build frontend, then pre-compress static assets (gzip, brotli, zstd)
+RUN bun run build && bun run scripts/compress-assets.ts

 # --- Chef Base Stage ---
 FROM lukemathwalker/cargo-chef:latest-rust-${RUST_VERSION} AS chef
@@ -112,5 +115,5 @@ HEALTHCHECK --interval=30s --timeout=3s --start-period=5s --retries=3 \
 ENV HOSTS=0.0.0.0,[::]

 # Implicitly uses PORT environment variable
-# temporary: running without 'scraper' service
-CMD ["sh", "-c", "exec ./banner --services web,bot"]
+# Runs all services: web, bot, and scraper
+CMD ["sh", "-c", "exec ./banner"]
@@ -1,48 +1,292 @@
 set dotenv-load
-default_services := "bot,web,scraper"

 default:
    just --list

-# Run all checks (format, clippy, tests, lint)
-check:
-    cargo fmt --all -- --check
-    cargo clippy --all-features -- --deny warnings
-    cargo nextest run
-    bun run --cwd web typecheck
-    bun run --cwd web lint
-    bun run --cwd web test --run
+# Run all checks in parallel. Pass -f/--fix to auto-format and fix first.
+[script("bun")]
+check *flags:
+    const args = "{{flags}}".split(/\s+/).filter(Boolean);
+    let fix = false;
+    for (const arg of args) {
+      if (arg === "-f" || arg === "--fix") fix = true;
+      else { console.error(`Unknown flag: ${arg}`); process.exit(1); }
+    }

-# Run all tests (Rust + frontend)
-test: test-rust test-web
+    const run = (cmd) => {
+      const proc = Bun.spawnSync(cmd, { stdio: ["inherit", "inherit", "inherit"] });
+      if (proc.exitCode !== 0) process.exit(proc.exitCode);
+    };

-# Run only Rust tests
-test-rust *ARGS:
-    cargo nextest run {{ARGS}}
+    if (fix) {
+      console.log("\x1b[1;36m→ Fixing...\x1b[0m");
+      run(["cargo", "fmt", "--all"]);
+      run(["bun", "run", "--cwd", "web", "format"]);
+      run(["cargo", "clippy", "--all-features", "--fix", "--allow-dirty", "--allow-staged",
+           "--", "--deny", "warnings"]);
+      console.log("\x1b[1;36m→ Verifying...\x1b[0m");
+    }

-# Run only frontend tests
-test-web:
-    bun run --cwd web test --run
+    const checks = [
+      { name: "rustfmt",      cmd: ["cargo", "fmt", "--all", "--", "--check"] },
+      { name: "clippy",       cmd: ["cargo", "clippy", "--all-features", "--", "--deny", "warnings"] },
+      { name: "rust-test",    cmd: ["cargo", "nextest", "run", "-E", "not test(export_bindings)"] },
+      { name: "svelte-check", cmd: ["bun", "run", "--cwd", "web", "check"] },
+      { name: "biome",        cmd: ["bun", "run", "--cwd", "web", "format:check"] },
+      { name: "web-test",     cmd: ["bun", "run", "--cwd", "web", "test"] },
+      // { name: "sqlx-prepare", cmd: ["cargo", "sqlx", "prepare", "--check"] },
+    ];

-# Quick check: clippy + tests only (skips formatting)
-check-quick:
-    cargo clippy --all-features -- --deny warnings
-    cargo nextest run
-    bun run --cwd web typecheck
+    const isTTY = process.stderr.isTTY;
+    const start = Date.now();
+    const remaining = new Set(checks.map(c => c.name));

-# Run the Banner API search demo (hits live UTSA API, ~20s)
-search *ARGS:
-    cargo run -q --bin search -- {{ARGS}}
+    const promises = checks.map(async (check) => {
+      const proc = Bun.spawn(check.cmd, {
+        env: { ...process.env, FORCE_COLOR: "1" },
+        stdout: "pipe", stderr: "pipe",
+      });
+      const [stdout, stderr] = await Promise.all([
+        new Response(proc.stdout).text(),
+        new Response(proc.stderr).text(),
+      ]);
+      await proc.exited;
+      return { ...check, stdout, stderr, exitCode: proc.exitCode,
+               elapsed: ((Date.now() - start) / 1000).toFixed(1) };
+    });
+
+    const interval = isTTY ? setInterval(() => {
+      const elapsed = ((Date.now() - start) / 1000).toFixed(1);
+      process.stderr.write(`\r\x1b[K${elapsed}s [${Array.from(remaining).join(", ")}]`);
+    }, 100) : null;
+
+    let anyFailed = false;
+    for (const promise of promises) {
+      const r = await promise;
+      remaining.delete(r.name);
+      if (isTTY) process.stderr.write(`\r\x1b[K`);
+      if (r.exitCode !== 0) {
+        anyFailed = true;
+        process.stdout.write(`\x1b[31m✗ ${r.name}\x1b[0m (${r.elapsed}s)\n`);
+        if (r.stdout) process.stdout.write(r.stdout);
+        if (r.stderr) process.stderr.write(r.stderr);
+      } else {
+        process.stdout.write(`\x1b[32m✓ ${r.name}\x1b[0m (${r.elapsed}s)\n`);
+      }
+    }
+
+    if (interval) clearInterval(interval);
+    if (isTTY) process.stderr.write(`\r\x1b[K`);
+    process.exit(anyFailed ? 1 : 0);

 # Format all Rust and TypeScript code
 format:
    cargo fmt --all
    bun run --cwd web format

-# Check formatting without modifying (CI-friendly)
-format-check:
-    cargo fmt --all -- --check
-    bun run --cwd web format:check
+# Run tests. Usage: just test [rust|web|<nextest filter args>]
+[script("bun")]
+test *args:
+    const input = "{{args}}".trim();
+    const run = (cmd) => {
+      const proc = Bun.spawnSync(cmd, { stdio: ["inherit", "inherit", "inherit"] });
+      if (proc.exitCode !== 0) process.exit(proc.exitCode);
+    };
+    if (input === "web") {
+      run(["bun", "run", "--cwd", "web", "test"]);
+    } else if (input === "rust") {
+      run(["cargo", "nextest", "run", "-E", "not test(export_bindings)"]);
+    } else if (input === "") {
+      run(["cargo", "nextest", "run", "-E", "not test(export_bindings)"]);
+      run(["bun", "run", "--cwd", "web", "test"]);
+    } else {
+      run(["cargo", "nextest", "run", ...input.split(/\s+/)]);
+    }
+
+# Generate TypeScript bindings from Rust types (ts-rs)
+bindings:
+    cargo test export_bindings
+
+# Run the Banner API search demo (hits live UTSA API, ~20s)
+search *ARGS:
+    cargo run -q --bin search -- {{ARGS}}
+
+# Pass args to binary after --: just dev -n -- --some-flag
+# Dev server. Flags: -f(rontend) -b(ackend) -W(no-watch) -n(o-build) -r(elease) -e(mbed) --tracing <fmt>
+[script("bun")]
+dev *flags:
+    const argv = "{{flags}}".split(/\s+/).filter(Boolean);
+
+    let frontendOnly = false, backendOnly = false;
+    let noWatch = false, noBuild = false, release = false, embed = false;
+    let tracing = "pretty";
+    const passthrough = [];
+
+    let i = 0;
+    let seenDashDash = false;
+    while (i < argv.length) {
+      const arg = argv[i];
+      if (seenDashDash) { passthrough.push(arg); i++; continue; }
+      if (arg === "--") { seenDashDash = true; i++; continue; }
+      if (arg.startsWith("--")) {
+        if (arg === "--frontend-only") frontendOnly = true;
+        else if (arg === "--backend-only") backendOnly = true;
+        else if (arg === "--no-watch") noWatch = true;
+        else if (arg === "--no-build") noBuild = true;
+        else if (arg === "--release") release = true;
+        else if (arg === "--embed") embed = true;
+        else if (arg === "--tracing") { tracing = argv[++i] || "pretty"; }
+        else { console.error(`Unknown flag: ${arg}`); process.exit(1); }
+      } else if (arg.startsWith("-") && arg.length > 1) {
+        for (const c of arg.slice(1)) {
+          if (c === "f") frontendOnly = true;
+          else if (c === "b") backendOnly = true;
+          else if (c === "W") noWatch = true;
+          else if (c === "n") noBuild = true;
+          else if (c === "r") release = true;
+          else if (c === "e") embed = true;
+          else { console.error(`Unknown flag: -${c}`); process.exit(1); }
+        }
+      } else { console.error(`Unknown argument: ${arg}`); process.exit(1); }
+      i++;
+    }
+
+    // -e implies -b (no point running Vite if assets are embedded)
+    if (embed) backendOnly = true;
+    // -n implies -W (no build means no watch)
+    if (noBuild) noWatch = true;
+
+    // Validate conflicting flags
+    if (frontendOnly && backendOnly) {
+      console.error("Cannot use -f and -b together (or -e implies -b)");
+      process.exit(1);
+    }
+
+    const runFrontend = !backendOnly;
+    const runBackend = !frontendOnly;
+    const profile = release ? "release" : "dev";
+    const profileDir = release ? "release" : "debug";
+
+    const procs = [];
+    const cleanup = async () => {
+      for (const p of procs) p.kill();
+      await Promise.all(procs.map(p => p.exited));
+    };
+    process.on("SIGINT", async () => { await cleanup(); process.exit(0); });
+    process.on("SIGTERM", async () => { await cleanup(); process.exit(0); });
+
+    // Build frontend first when embedding assets (backend will bake them in)
+    if (embed && !noBuild) {
+      console.log(`\x1b[1;36m→ Building frontend (for embedding)...\x1b[0m`);
+      const fb = Bun.spawnSync(["bun", "run", "--cwd", "web", "build"], {
+        stdio: ["inherit", "inherit", "inherit"],
+      });
+      if (fb.exitCode !== 0) process.exit(fb.exitCode);
+    }
+
+    // Frontend: Vite dev server
+    if (runFrontend) {
+      const proc = Bun.spawn(["bun", "run", "--cwd", "web", "dev"], {
+        stdio: ["inherit", "inherit", "inherit"],
+      });
+      procs.push(proc);
+    }
+
+    // Backend
+    if (runBackend) {
+      const backendArgs = [`--tracing`, tracing, ...passthrough];
+      const bin = `target/${profileDir}/banner`;
+
+      if (noWatch) {
+        // Build first unless -n (skip build)
+        if (!noBuild) {
+          console.log(`\x1b[1;36m→ Building backend (${profile})...\x1b[0m`);
+          const cargoArgs = ["cargo", "build", "--bin", "banner"];
+          if (!embed) cargoArgs.push("--no-default-features");
+          if (release) cargoArgs.push("--release");
+          const build = Bun.spawnSync(cargoArgs, { stdio: ["inherit", "inherit", "inherit"] });
+          if (build.exitCode !== 0) { cleanup(); process.exit(build.exitCode); }
+        }
+
+        // Run the binary directly (no watch)
+        const { existsSync } = await import("fs");
+        if (!existsSync(bin)) {
+          console.error(`Binary not found: ${bin}`);
+          console.error(`Run 'just build${release ? "" : " -d"}' first, or remove -n to use bacon.`);
+          cleanup();
+          process.exit(1);
+        }
+        console.log(`\x1b[1;36m→ Running ${bin} (no watch)\x1b[0m`);
+        const proc = Bun.spawn([bin, ...backendArgs], {
+          stdio: ["inherit", "inherit", "inherit"],
+        });
+        procs.push(proc);
+      } else {
+        // Bacon watch mode
+        const baconArgs = ["bacon", "--headless", "run", "--"];
+        if (!embed) baconArgs.push("--no-default-features");
+        if (release) baconArgs.push("--profile", "release");
+        baconArgs.push("--", ...backendArgs);
+        const proc = Bun.spawn(baconArgs, {
+          stdio: ["inherit", "inherit", "inherit"],
+        });
+        procs.push(proc);
+      }
+    }
+
+    // Wait for any process to exit, then kill the rest
+    const results = procs.map((p, i) => p.exited.then(code => ({ i, code })));
+    const first = await Promise.race(results);
+    cleanup();
+    process.exit(first.code);
+
+# Production build. Flags: -d(ebug) -f(rontend-only) -b(ackend-only)
+[script("bun")]
+build *flags:
+    const argv = "{{flags}}".split(/\s+/).filter(Boolean);
+
+    let debug = false, frontendOnly = false, backendOnly = false;
+    for (const arg of argv) {
+      if (arg.startsWith("--")) {
+        if (arg === "--debug") debug = true;
+        else if (arg === "--frontend-only") frontendOnly = true;
+        else if (arg === "--backend-only") backendOnly = true;
+        else { console.error(`Unknown flag: ${arg}`); process.exit(1); }
+      } else if (arg.startsWith("-") && arg.length > 1) {
+        for (const c of arg.slice(1)) {
+          if (c === "d") debug = true;
+          else if (c === "f") frontendOnly = true;
+          else if (c === "b") backendOnly = true;
+          else { console.error(`Unknown flag: -${c}`); process.exit(1); }
+        }
+      } else { console.error(`Unknown argument: ${arg}`); process.exit(1); }
+    }
+
+    if (frontendOnly && backendOnly) {
+      console.error("Cannot use -f and -b together");
+      process.exit(1);
+    }
+
+    const run = (cmd) => {
+      const proc = Bun.spawnSync(cmd, { stdio: ["inherit", "inherit", "inherit"] });
+      if (proc.exitCode !== 0) process.exit(proc.exitCode);
+    };
+
+    const buildFrontend = !backendOnly;
+    const buildBackend = !frontendOnly;
+    const profile = debug ? "debug" : "release";
+
+    if (buildFrontend) {
+      console.log("\x1b[1;36m→ Building frontend...\x1b[0m");
+      run(["bun", "run", "--cwd", "web", "build"]);
+    }
+
+    if (buildBackend) {
+      console.log(`\x1b[1;36m→ Building backend (${profile})...\x1b[0m`);
+      const cmd = ["cargo", "build", "--bin", "banner"];
+      if (!debug) cmd.push("--release");
+      run(cmd);
+    }

 # Start PostgreSQL in Docker and update .env with connection string
 # Commands: start (default), reset, rm
@@ -112,86 +356,6 @@ db cmd="start":
      await updateEnv();
    }

-# Auto-reloading frontend server
-frontend:
-    bun run --cwd web dev
-
-# Production build of frontend
-build-frontend:
-    bun run --cwd web build
-
-# Auto-reloading backend server (with embedded assets)
-backend *ARGS:
-    bacon --headless run -- -- {{ARGS}}
-
-# Auto-reloading backend server (no embedded assets, for dev proxy mode)
-backend-dev *ARGS:
-    bacon --headless run -- --no-default-features -- {{ARGS}}
-
-# Production build
-build:
-    bun run --cwd web build
-    cargo build --release --bin banner
-
-# Run auto-reloading development build with release characteristics
-dev-build *ARGS='--services web --tracing pretty': build-frontend
-    bacon --headless run -- --profile dev-release -- {{ARGS}}
-
-# Auto-reloading development build: Vite frontend + backend (no embedded assets, proxies to Vite)
-[parallel]
-dev *ARGS='--services web,bot': frontend (backend-dev ARGS)
-
-# Smoke test: start web server, hit API endpoints, verify responses
-[script("bash")]
-test-smoke port="18080":
-    set -euo pipefail
-    PORT={{port}}
-
-    cleanup() { kill "$SERVER_PID" 2>/dev/null; wait "$SERVER_PID" 2>/dev/null; }
-
-    # Start server in background
-    PORT=$PORT cargo run -q --no-default-features -- --services web --tracing json &
-    SERVER_PID=$!
-    trap cleanup EXIT
-
-    # Wait for server to be ready (up to 15s)
-    for i in $(seq 1 30); do
-        if curl -sf "http://localhost:$PORT/api/health" >/dev/null 2>&1; then break; fi
-        if ! kill -0 "$SERVER_PID" 2>/dev/null; then echo "FAIL: server exited early"; exit 1; fi
-        sleep 0.5
-    done
-
-    PASS=0; FAIL=0
-    check() {
-        local label="$1" url="$2" expected="$3"
-        body=$(curl -sf "$url") || { echo "FAIL: $label - request failed"; FAIL=$((FAIL+1)); return; }
-        if echo "$body" | grep -q "$expected"; then
-            echo "PASS: $label"
-            PASS=$((PASS+1))
-        else
-            echo "FAIL: $label - expected '$expected' in: $body"
-            FAIL=$((FAIL+1))
-        fi
-    }
-
-    check "GET /api/health" "http://localhost:$PORT/api/health" '"status":"healthy"'
-    check "GET /api/status" "http://localhost:$PORT/api/status" '"version"'
-    check "GET /api/metrics" "http://localhost:$PORT/api/metrics" '"banner_api"'
-
-    # Test 404
-    STATUS=$(curl -s -o /dev/null -w "%{http_code}" "http://localhost:$PORT/api/nonexistent")
-    if [ "$STATUS" = "404" ]; then
-        echo "PASS: 404 on unknown route"
-        PASS=$((PASS+1))
-    else
-        echo "FAIL: expected 404, got $STATUS"
-        FAIL=$((FAIL+1))
-    fi
-
-    echo ""
-    echo "Results: $PASS passed, $FAIL failed"
-    [ "$FAIL" -eq 0 ]
-
 alias b := bun
 bun *ARGS:
 	cd web && bun {{ ARGS }}
@@ -29,8 +29,7 @@ The application consists of three modular services that can be run independently
 bun install --cwd web  # Install frontend dependencies
 cargo build  # Build the backend

-just dev # Runs auto-reloading dev build
-just dev --services bot,web # Runs auto-reloading dev build, running only the bot and web services
+just dev # Runs auto-reloading dev build with all services
 just dev-build # Development build with release characteristics (frontend is embedded, non-auto-reloading)

 just build # Production build that embeds assets
@@ -4,10 +4,33 @@

 The Banner project is built as a multi-service application with the following components:

- **Discord Bot Service**: Handles Discord interactions and commands
- **Web Service**: Serves the React frontend and provides API endpoints
- **Scraper Service**: Background data collection and synchronization
- **Database Layer**: PostgreSQL for persistent storage
+- **Discord Bot Service**: Handles Discord interactions and commands (Serenity/Poise)
+- **Web Service**: Axum HTTP server serving the SvelteKit frontend and REST API endpoints
+- **Scraper Service**: Background data collection and synchronization with job queue
+- **Database Layer**: PostgreSQL 17 for persistent storage (SQLx with compile-time verification)
+- **RateMyProfessors Client**: GraphQL-based bulk sync of professor ratings
+
+### Frontend Stack
+
+- **SvelteKit** with Svelte 5 runes (`$state`, `$derived`, `$effect`)
+- **Tailwind CSS v4** via `@tailwindcss/vite`
+- **bits-ui** for headless UI primitives (comboboxes, tooltips, dropdowns)
+- **TanStack Table** for interactive data tables with sorting and column control
+- **OverlayScrollbars** for styled, theme-aware scrollable areas
+- **ts-rs** generates TypeScript type bindings from Rust structs
+
+### API Endpoints
+
+| Endpoint | Description |
+|---|---|
+| `GET /api/health` | Health check |
+| `GET /api/status` | Service status, version, and commit hash |
+| `GET /api/metrics` | Basic metrics |
+| `GET /api/courses/search` | Paginated course search with filters (term, subject, query, open-only, sort) |
+| `GET /api/courses/:term/:crn` | Single course detail with instructors and RMP ratings |
+| `GET /api/terms` | Available terms from reference cache |
+| `GET /api/subjects?term=` | Subjects for a term, ordered by enrollment |
+| `GET /api/reference/:category` | Reference data lookups (campuses, instructional methods, etc.) |

 ## Technical Analysis

@@ -0,0 +1,61 @@
+# Changelog
+
+All notable changes to this project will be documented in this file.
+
+The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/).
+
+## [Unreleased]
+
+## [0.5.0] - 2026-01-29
+
+### Added
+
+- Multi-select subject filtering with searchable comboboxes.
+- Smart instructor name abbreviation for compact table display.
+- Delivery mode indicators and tooltips in location column.
+- Page selector dropdown with animated pagination controls.
+- FLIP animations for smooth table row transitions during pagination.
+- Time tooltip with detailed meeting schedule and day abbreviations.
+- Reusable SimpleTooltip component for consistent UI hints.
+
+### Changed
+
+- Consolidated query logic and eliminated N+1 instructor loads via batch fetching.
+- Consolidated menu snippets and strengthened component type safety.
+- Enhanced table scrolling with OverlayScrollbars and theme-aware styling.
+- Eliminated initial theme flash on page load.
+
+## [0.4.0] - 2026-01-28
+
+### Added
+
+- Web-based course search UI with interactive data table, multi-column sorting, and column visibility controls.
+- TypeScript type bindings generated from Rust types via ts-rs.
+- RateMyProfessors integration: bulk professor sync via GraphQL and inline rating display in search results.
+- Course detail expansion panel with enrollment, meeting times, and instructor info.
+- OverlayScrollbars integration for styled, theme-aware scrollable areas.
+- Pagination component for navigating large search result sets.
+- Footer component with version display.
+- API endpoints: `/api/courses/search`, `/api/courses/:term/:crn`, `/api/terms`, `/api/subjects`, `/api/reference/:category`.
+- Frontend API client with typed request/response handling and test coverage.
+- Course formatting utilities with comprehensive unit tests.
+
+## [0.3.4] - 2026-01
+
+### Added
+
+- Live service status tracking on web dashboard with auto-refresh and health indicators.
+- DB operation extraction for improved testability.
+- Unit test suite foundation covering core functionality.
+- Docker support for PostgreSQL development environment.
+- ICS calendar export with comprehensive holiday exclusion coverage.
+- Google Calendar link generation with recurrence rules and meeting details.
+- Job queue with priority-based scheduling for background scraping.
+- Rate limiting with burst allowance for Banner API requests.
+- Session management and caching for Banner API interactions.
+- Discord bot commands: search, terms, ics, gcal.
+- Intelligent scraping system with priority queues and retry tracking.
+
+### Changed
+
+- Type consolidation and dead code removal across the codebase.
@@ -1,58 +0,0 @@
-# Features
-
-## Current Features
-
-### Discord Bot Commands
-
- **search** - Search for courses with various filters (title, course code, keywords)
- **terms** - List available terms or search for a specific term
- **time** - Get meeting times for a specific course (CRN)
- **ics** - Generate ICS calendar file for a course with holiday exclusions
- **gcal** - Generate Google Calendar link for a course
-
-### Data Pipeline
-
- Intelligent scraping system with priority queues
- Rate limiting and burst handling
- Background data synchronization
-
-## Feature Wishlist
-
-### Commands
-
- ICS Download (get a ICS download of your classes with location & timing perfectly - set for every class you're in)
- Classes Now (find classes happening)
- Autocomplete
-  - Class Title
-  - Course Number
-  - Term/Part of Term
-  - Professor
-  - Attribute
- Component Pagination
- RateMyProfessor Integration (Linked/Embedded)
- Smart term selection (i.e. Summer 2024 will be selected automatically when opened)
- Rate Limiting (bursting with global/user limits)
- DMs Integration (allow usage of the bot in DMs)
- Class Change Notifications (get notified when details about a class change)
- Multi-term Querying (currently the backend for searching is kinda weird)
- Full Autocomplete for Every Search Option
- Metrics, Log Query, Privileged Error Feedback
- Search for Classes
-  - Major, Professor, Location, Name, Time of Day
- Subscribe to Classes
-  - Availability (seat, pre-seat)
-  - Waitlist Movement
-  - Detail Changes (meta, time, location, seats, professor)
-    - `time` Start, End, Days of Week
-    - `seats` Any change in seat/waitlist data
-    - `meta`
- Lookup via Course Reference Number (CRN)
- Smart Time of Day Handling
-  - "2 PM" -> Start within 2:00 PM to 2:59 PM
-  - "2-3 PM" -> Start within 2:00 PM to 3:59 PM
-  - "ends by 2 PM" -> Ends within 12:00 AM to 2:00 PM
-  - "after 2 PM" -> Start within 2:01 PM to 11:59 PM
-  - "before 2 PM" -> Ends within 12:00 AM to 1:59 PM
- Get By Section Command
-  - CS 4393 001 =>
-  - Will require SQL to be able to search for a class by its section number
@@ -4,7 +4,8 @@ This folder contains detailed documentation for the Banner project. This file ac

 ## Files

- [`FEATURES.md`](FEATURES.md) - Current features, implemented functionality, and future roadmap
+- [`CHANGELOG.md`](CHANGELOG.md) - Notable changes by version
+- [`ROADMAP.md`](ROADMAP.md) - Planned features and priorities
 - [`BANNER.md`](BANNER.md) - General API documentation on the Banner system
 - [`ARCHITECTURE.md`](ARCHITECTURE.md) - Technical implementation details, system design, and analysis

@@ -0,0 +1,35 @@
+# Roadmap
+
+## Now
+
+- **Notification and subscription system** - Subscribe to courses and get alerts on seat availability, waitlist movement, and detail changes (time, location, professor, seats). DB schema exists.
+- **Professor name search filter** - Filter search results by instructor. Backend code exists but is commented out.
+- **Autocomplete for search fields** - Typeahead for course titles, course numbers, professors, and terms.
+- **Test coverage expansion** - Broaden coverage with session/rate-limiter tests and more DB integration tests.
+
+## Soon
+
+- **Smart time-of-day search parsing** - Support natural queries like "2 PM", "2-3 PM", "ends by 2 PM", "after 2 PM", "before 2 PM" mapped to time ranges.
+- **Section-based lookup** - Search by full section identifier, e.g. "CS 4393 001".
+- **Search result pagination** - Paginated embeds for large result sets in Discord.
+- **Multi-term querying** - Query across multiple terms in a single search instead of one at a time.
+- **Historical analytics** - Track seat availability over time and visualize fill-rate trends per course or professor.
+- **Schedule builder** - Visual weekly schedule tool for assembling a conflict-free course lineup.
+- **Professor stats** - Aggregate data views: average class size, typical waitlist length, schedule patterns across semesters.
+
+## Eventually
+
+- **Degree audit helper** - Map available courses to degree requirements and suggest what to take next.
+- **Dynamic scraper scheduling** - Adjust scrape intervals based on change frequency and course count (e.g. 2 hours per 500 courses, shorter intervals when changes are detected).
+- **DM support** - Allow the Discord bot to respond in direct messages, not just guild channels.
+- **"Classes Now" command** - Find classes currently in session based on the current day and time.
+- **CRN direct lookup** - Look up a course by its CRN without going through search.
+- **Metrics dashboard** - Surface scraper and service metrics visually on the web dashboard.
+- **Privileged error feedback** - Detailed error information surfaced to bot admins when commands fail.
+
+## Done
+
+- **Web course search UI** - Browser-based course search with interactive data table, sorting, pagination, and column controls. *(0.4.0)*
+- **RateMyProfessor integration** - Bulk professor sync via GraphQL with inline ratings in search results. *(0.4.0)*
+- **Subject/major search filter** - Multi-select subject filtering with searchable comboboxes. *(0.5.0)*
+- **Test coverage expansion** - Unit tests for course formatting, API client, query builder, CLI args, and config parsing. *(0.3.4–0.4.0)*
@@ -0,0 +1,83 @@
+-- ============================================================
+-- Expand courses table with rich Banner API fields
+-- ============================================================
+
+-- Section identifiers
+ALTER TABLE courses ADD COLUMN sequence_number VARCHAR;
+ALTER TABLE courses ADD COLUMN part_of_term VARCHAR;
+
+-- Schedule & delivery (store codes, descriptions come from reference_data)
+ALTER TABLE courses ADD COLUMN instructional_method VARCHAR;
+ALTER TABLE courses ADD COLUMN campus VARCHAR;
+
+-- Credit hours
+ALTER TABLE courses ADD COLUMN credit_hours INTEGER;
+ALTER TABLE courses ADD COLUMN credit_hour_low INTEGER;
+ALTER TABLE courses ADD COLUMN credit_hour_high INTEGER;
+
+-- Cross-listing
+ALTER TABLE courses ADD COLUMN cross_list VARCHAR;
+ALTER TABLE courses ADD COLUMN cross_list_capacity INTEGER;
+ALTER TABLE courses ADD COLUMN cross_list_count INTEGER;
+
+-- Section linking
+ALTER TABLE courses ADD COLUMN link_identifier VARCHAR;
+ALTER TABLE courses ADD COLUMN is_section_linked BOOLEAN;
+
+-- JSONB columns for 1-to-many data
+ALTER TABLE courses ADD COLUMN meeting_times JSONB NOT NULL DEFAULT '[]'::jsonb;
+ALTER TABLE courses ADD COLUMN attributes JSONB NOT NULL DEFAULT '[]'::jsonb;
+
+-- ============================================================
+-- Full-text search support
+-- ============================================================
+
+-- Generated tsvector for word-based search on title
+ALTER TABLE courses ADD COLUMN title_search tsvector
+    GENERATED ALWAYS AS (to_tsvector('simple', coalesce(title, ''))) STORED;
+
+CREATE INDEX idx_courses_title_search ON courses USING GIN (title_search);
+
+-- Trigram index for substring/ILIKE search on title
+CREATE EXTENSION IF NOT EXISTS pg_trgm;
+CREATE INDEX idx_courses_title_trgm ON courses USING GIN (title gin_trgm_ops);
+
+-- ============================================================
+-- New filter indexes
+-- ============================================================
+
+CREATE INDEX idx_courses_instructional_method ON courses(instructional_method);
+CREATE INDEX idx_courses_campus ON courses(campus);
+
+-- Composite for "open CS courses in Fall 2024" pattern
+CREATE INDEX idx_courses_term_subject_avail ON courses(term_code, subject, max_enrollment, enrollment);
+
+-- ============================================================
+-- Instructors table (normalized, deduplicated)
+-- ============================================================
+
+CREATE TABLE instructors (
+    banner_id VARCHAR PRIMARY KEY,
+    display_name VARCHAR NOT NULL,
+    email VARCHAR
+);
+
+CREATE TABLE course_instructors (
+    course_id INTEGER NOT NULL REFERENCES courses(id) ON DELETE CASCADE,
+    instructor_id VARCHAR NOT NULL REFERENCES instructors(banner_id) ON DELETE CASCADE,
+    is_primary BOOLEAN NOT NULL DEFAULT false,
+    PRIMARY KEY (course_id, instructor_id)
+);
+
+CREATE INDEX idx_course_instructors_instructor ON course_instructors(instructor_id);
+
+-- ============================================================
+-- Reference data table (all code→description lookups)
+-- ============================================================
+
+CREATE TABLE reference_data (
+    category VARCHAR NOT NULL,
+    code VARCHAR NOT NULL,
+    description VARCHAR NOT NULL,
+    PRIMARY KEY (category, code)
+);
@@ -0,0 +1,17 @@
+-- RMP professor data (bulk synced from RateMyProfessors)
+CREATE TABLE rmp_professors (
+    legacy_id INTEGER PRIMARY KEY,
+    graphql_id VARCHAR NOT NULL,
+    first_name VARCHAR NOT NULL,
+    last_name VARCHAR NOT NULL,
+    department VARCHAR,
+    avg_rating REAL,
+    avg_difficulty REAL,
+    num_ratings INTEGER NOT NULL DEFAULT 0,
+    would_take_again_pct REAL,
+    last_synced_at TIMESTAMPTZ NOT NULL DEFAULT NOW()
+);
+
+-- Link Banner instructors to RMP professors
+ALTER TABLE instructors ADD COLUMN rmp_legacy_id INTEGER REFERENCES rmp_professors(legacy_id);
+ALTER TABLE instructors ADD COLUMN rmp_match_status VARCHAR NOT NULL DEFAULT 'pending';
@@ -0,0 +1,7 @@
+-- Add queued_at column to track when a job last entered the "ready to pick up" state.
+-- For fresh jobs this equals execute_at; for retried jobs it is updated to NOW().
+ALTER TABLE scrape_jobs
+    ADD COLUMN queued_at TIMESTAMPTZ NOT NULL DEFAULT NOW();
+
+-- Backfill existing rows: set queued_at = execute_at (best approximation)
+UPDATE scrape_jobs SET queued_at = execute_at;
@@ -0,0 +1,19 @@
+CREATE TABLE users (
+    discord_id BIGINT PRIMARY KEY,
+    discord_username TEXT NOT NULL,
+    discord_avatar_hash TEXT,
+    is_admin BOOLEAN NOT NULL DEFAULT false,
+    created_at TIMESTAMPTZ NOT NULL DEFAULT now(),
+    updated_at TIMESTAMPTZ NOT NULL DEFAULT now()
+);
+
+CREATE TABLE user_sessions (
+    id TEXT PRIMARY KEY,
+    user_id BIGINT NOT NULL REFERENCES users(discord_id) ON DELETE CASCADE,
+    created_at TIMESTAMPTZ NOT NULL DEFAULT now(),
+    expires_at TIMESTAMPTZ NOT NULL,
+    last_active_at TIMESTAMPTZ NOT NULL DEFAULT now()
+);
+
+CREATE INDEX idx_user_sessions_user_id ON user_sessions(user_id);
+CREATE INDEX idx_user_sessions_expires_at ON user_sessions(expires_at);
@@ -0,0 +1,80 @@
+-- Collapse instructors from per-banner-id rows to per-person rows (deduped by lowercased email).
+-- All existing RMP matches are deliberately dropped; the new auto-matcher will re-score from scratch.
+
+-- 1. Create the new instructors table (1 row per person, keyed by email)
+CREATE TABLE instructors_new (
+    id SERIAL PRIMARY KEY,
+    display_name VARCHAR NOT NULL,
+    email VARCHAR NOT NULL,
+    rmp_professor_id INTEGER UNIQUE REFERENCES rmp_professors(legacy_id),
+    rmp_match_status VARCHAR NOT NULL DEFAULT 'unmatched',
+    CONSTRAINT instructors_email_unique UNIQUE (email)
+);
+
+-- 2. Populate from existing data, deduplicating by lowercased email.
+--    For each email, pick the display_name from the row with the highest banner_id
+--    (deterministic tiebreaker). All rmp fields start fresh (NULL / 'unmatched').
+INSERT INTO instructors_new (display_name, email)
+SELECT DISTINCT ON (LOWER(email))
+    display_name,
+    LOWER(email)
+FROM instructors
+ORDER BY LOWER(email), banner_id DESC;
+
+-- 3. Create the new course_instructors table with integer FK and banner_id column
+CREATE TABLE course_instructors_new (
+    course_id INTEGER NOT NULL REFERENCES courses(id) ON DELETE CASCADE,
+    instructor_id INTEGER NOT NULL REFERENCES instructors_new(id) ON DELETE CASCADE,
+    banner_id VARCHAR NOT NULL,
+    is_primary BOOLEAN NOT NULL DEFAULT false,
+    PRIMARY KEY (course_id, instructor_id)
+);
+
+-- 4. Populate from old data, mapping old banner_id → new instructor id via lowercased email.
+--    Use DISTINCT ON to handle cases where multiple old banner_ids for the same person
+--    taught the same course (would cause duplicate (course_id, instructor_id) pairs).
+INSERT INTO course_instructors_new (course_id, instructor_id, banner_id, is_primary)
+SELECT DISTINCT ON (ci.course_id, inew.id)
+    ci.course_id,
+    inew.id,
+    ci.instructor_id,  -- old banner_id
+    ci.is_primary
+FROM course_instructors ci
+JOIN instructors iold ON iold.banner_id = ci.instructor_id
+JOIN instructors_new inew ON inew.email = LOWER(iold.email)
+ORDER BY ci.course_id, inew.id, ci.is_primary DESC;
+
+-- 5. Drop old tables (course_instructors first due to FK dependency)
+DROP TABLE course_instructors;
+DROP TABLE instructors;
+
+-- 6. Rename new tables into place
+ALTER TABLE instructors_new RENAME TO instructors;
+ALTER TABLE course_instructors_new RENAME TO course_instructors;
+
+-- 7. Rename constraints to match the final table names
+ALTER TABLE instructors RENAME CONSTRAINT instructors_new_pkey TO instructors_pkey;
+ALTER TABLE instructors RENAME CONSTRAINT instructors_new_rmp_professor_id_key TO instructors_rmp_professor_id_key;
+ALTER TABLE course_instructors RENAME CONSTRAINT course_instructors_new_pkey TO course_instructors_pkey;
+
+-- 8. Recreate indexes
+CREATE INDEX idx_course_instructors_instructor ON course_instructors (instructor_id);
+CREATE INDEX idx_instructors_rmp_status ON instructors (rmp_match_status);
+CREATE INDEX idx_instructors_email ON instructors (email);
+
+-- 9. Create rmp_match_candidates table
+CREATE TABLE rmp_match_candidates (
+    id SERIAL PRIMARY KEY,
+    instructor_id INTEGER NOT NULL REFERENCES instructors(id) ON DELETE CASCADE,
+    rmp_legacy_id INTEGER NOT NULL REFERENCES rmp_professors(legacy_id),
+    score REAL NOT NULL,
+    score_breakdown JSONB NOT NULL DEFAULT '{}',
+    status VARCHAR NOT NULL DEFAULT 'pending',
+    created_at TIMESTAMPTZ NOT NULL DEFAULT NOW(),
+    resolved_at TIMESTAMPTZ,
+    resolved_by BIGINT REFERENCES users(discord_id),
+    CONSTRAINT uq_candidate_pair UNIQUE (instructor_id, rmp_legacy_id)
+);
+
+CREATE INDEX idx_match_candidates_instructor ON rmp_match_candidates (instructor_id);
+CREATE INDEX idx_match_candidates_status ON rmp_match_candidates (status);
@@ -0,0 +1,24 @@
+-- Multi-RMP profile support: allow many RMP profiles per instructor.
+-- Each RMP profile still links to at most one instructor (rmp_legacy_id UNIQUE).
+
+-- 1. Create junction table
+CREATE TABLE instructor_rmp_links (
+    id SERIAL PRIMARY KEY,
+    instructor_id INTEGER NOT NULL REFERENCES instructors(id) ON DELETE CASCADE,
+    rmp_legacy_id INTEGER NOT NULL UNIQUE REFERENCES rmp_professors(legacy_id),
+    created_at TIMESTAMPTZ NOT NULL DEFAULT NOW(),
+    created_by BIGINT REFERENCES users(discord_id),
+    source VARCHAR NOT NULL DEFAULT 'manual'  -- 'auto' | 'manual'
+);
+
+CREATE INDEX idx_instructor_rmp_links_instructor ON instructor_rmp_links (instructor_id);
+
+-- 2. Migrate existing matches
+INSERT INTO instructor_rmp_links (instructor_id, rmp_legacy_id, source)
+SELECT id, rmp_professor_id,
+       CASE rmp_match_status WHEN 'auto' THEN 'auto' ELSE 'manual' END
+FROM instructors
+WHERE rmp_professor_id IS NOT NULL;
+
+-- 3. Drop old column (and its unique constraint)
+ALTER TABLE instructors DROP COLUMN rmp_professor_id;
@@ -0,0 +1,31 @@
+-- Scrape job results log: one row per completed (or failed) job for effectiveness tracking.
+CREATE TABLE scrape_job_results (
+    id              BIGSERIAL PRIMARY KEY,
+    target_type     target_type NOT NULL,
+    payload         JSONB NOT NULL,
+    priority        scrape_priority NOT NULL,
+
+    -- Timing
+    queued_at       TIMESTAMPTZ NOT NULL,
+    started_at      TIMESTAMPTZ NOT NULL,
+    completed_at    TIMESTAMPTZ NOT NULL DEFAULT NOW(),
+    duration_ms     INT NOT NULL,
+
+    -- Outcome
+    success         BOOLEAN NOT NULL,
+    error_message   TEXT,
+    retry_count     INT NOT NULL DEFAULT 0,
+
+    -- Effectiveness (NULL when success = false)
+    courses_fetched     INT,
+    courses_changed     INT,
+    courses_unchanged   INT,
+    audits_generated    INT,
+    metrics_generated   INT
+);
+
+CREATE INDEX idx_scrape_job_results_target_time
+    ON scrape_job_results (target_type, completed_at);
+
+CREATE INDEX idx_scrape_job_results_completed
+    ON scrape_job_results (completed_at);
@@ -0,0 +1,13 @@
+-- Indexes for the timeline aggregation endpoint.
+-- The query buckets course_metrics by 15-minute intervals, joins to courses
+-- for subject, and aggregates enrollment. These indexes support efficient
+-- time-range scans and the join.
+
+-- Primary access pattern: scan course_metrics by timestamp range
+CREATE INDEX IF NOT EXISTS idx_course_metrics_timestamp
+    ON course_metrics (timestamp);
+
+-- Composite index for the DISTINCT ON (bucket, course_id) ordered by timestamp DESC
+-- to efficiently pick the latest metric per course per bucket.
+CREATE INDEX IF NOT EXISTS idx_course_metrics_course_timestamp
+    ON course_metrics (course_id, timestamp DESC);
@@ -6,13 +6,14 @@ use crate::services::bot::BotService;
 use crate::services::manager::ServiceManager;
 use crate::services::web::WebService;
 use crate::state::AppState;
+use crate::web::auth::AuthConfig;
+use anyhow::Context;
 use figment::value::UncasedStr;
 use figment::{Figment, providers::Env};
 use sqlx::postgres::PgPoolOptions;
 use std::process::ExitCode;
 use std::sync::Arc;
 use std::time::Duration;
-use anyhow::Context;
 use tracing::{error, info};

 /// Main application struct containing all necessary components
@@ -79,6 +80,24 @@ impl App {
        let banner_api_arc = Arc::new(banner_api);
        let app_state = AppState::new(banner_api_arc.clone(), db_pool.clone());

+        // Load reference data cache from DB (may be empty on first run)
+        if let Err(e) = app_state.load_reference_cache().await {
+            info!(error = ?e, "Could not load reference cache on startup (may be empty)");
+        }
+
+        // Load schedule cache for timeline enrollment queries
+        if let Err(e) = app_state.schedule_cache.load().await {
+            info!(error = ?e, "Could not load schedule cache on startup (may be empty)");
+        }
+
+        // Seed the initial admin user if configured
+        if let Some(admin_id) = config.admin_discord_id {
+            let user = crate::data::users::ensure_seed_admin(&db_pool, admin_id as i64)
+                .await
+                .context("Failed to seed admin user")?;
+            info!(discord_id = admin_id, username = %user.discord_username, "Seed admin ensured");
+        }
+
        Ok(App {
            config,
            db_pool,
@@ -92,7 +111,16 @@ impl App {
    pub fn setup_services(&mut self, services: &[ServiceName]) -> Result<(), anyhow::Error> {
        // Register enabled services with the manager
        if services.contains(&ServiceName::Web) {
-            let web_service = Box::new(WebService::new(self.config.port, self.app_state.clone()));
+            let auth_config = AuthConfig {
+                client_id: self.config.discord_client_id.clone(),
+                client_secret: self.config.discord_client_secret.clone(),
+                redirect_base: self.config.discord_redirect_uri.clone(),
+            };
+            let web_service = Box::new(WebService::new(
+                self.config.port,
+                self.app_state.clone(),
+                auth_config,
+            ));
            self.service_manager
                .register_service(ServiceName::Web.as_str(), web_service);
        }
@@ -101,7 +129,9 @@ impl App {
            let scraper_service = Box::new(ScraperService::new(
                self.db_pool.clone(),
                self.banner_api.clone(),
+                self.app_state.reference_cache.clone(),
                self.app_state.service_statuses.clone(),
+                self.app_state.scrape_job_tx.clone(),
            ));
            self.service_manager
                .register_service(ServiceName::Scraper.as_str(), scraper_service);
@@ -40,9 +40,9 @@ impl BannerApi {
                .cookie_store(false)
                .user_agent(user_agent())
                .tcp_keepalive(Some(std::time::Duration::from_secs(60 * 5)))
-                .read_timeout(std::time::Duration::from_secs(10))
-                .connect_timeout(std::time::Duration::from_secs(10))
-                .timeout(std::time::Duration::from_secs(30))
+                .read_timeout(std::time::Duration::from_secs(20))
+                .connect_timeout(std::time::Duration::from_secs(15))
+                .timeout(std::time::Duration::from_secs(40))
                .build()
                .context("Failed to create HTTP client")?,
        )
@@ -228,6 +228,29 @@ impl BannerApi {
            .await
    }

+    /// Retrieves campus codes and descriptions.
+    pub async fn get_campuses(&self, term: &str) -> Result<Vec<Pair>> {
+        self.get_list_endpoint("get_campus", "", term, 1, 500).await
+    }
+
+    /// Retrieves instructional method codes and descriptions.
+    pub async fn get_instructional_methods(&self, term: &str) -> Result<Vec<Pair>> {
+        self.get_list_endpoint("get_instructionalMethod", "", term, 1, 500)
+            .await
+    }
+
+    /// Retrieves part-of-term codes and descriptions.
+    pub async fn get_parts_of_term(&self, term: &str) -> Result<Vec<Pair>> {
+        self.get_list_endpoint("get_partOfTerm", "", term, 1, 500)
+            .await
+    }
+
+    /// Retrieves section attribute codes and descriptions.
+    pub async fn get_attributes(&self, term: &str) -> Result<Vec<Pair>> {
+        self.get_list_endpoint("get_attribute", "", term, 1, 500)
+            .await
+    }
+
    /// Retrieves meeting time information for a course.
    pub async fn get_course_meeting_time(
        &self,
@@ -1,4 +1,4 @@
-use bitflags::{bitflags, Flags};
+use bitflags::{Flags, bitflags};
 use chrono::{DateTime, NaiveDate, NaiveTime, Timelike, Utc, Weekday};
 use extension_traits::extension;
 use serde::{Deserialize, Deserializer, Serialize};
@@ -320,10 +320,11 @@ pub enum MeetingType {
    Unknown(String),
 }

-impl MeetingType {
-    /// Parse from the meeting type string
-    pub fn from_string(s: &str) -> Self {
-        match s {
+impl std::str::FromStr for MeetingType {
+    type Err = std::convert::Infallible;
+
+    fn from_str(s: &str) -> std::result::Result<Self, Self::Err> {
+        Ok(match s {
            "HB" | "H2" | "H1" => MeetingType::HybridBlended,
            "OS" => MeetingType::OnlineSynchronous,
            "OA" => MeetingType::OnlineAsynchronous,
@@ -331,9 +332,11 @@ impl MeetingType {
            "ID" => MeetingType::IndependentStudy,
            "FF" => MeetingType::FaceToFace,
            other => MeetingType::Unknown(other.to_string()),
-        }
+        })
    }
+}

+impl MeetingType {
    /// Get description for the meeting type
    pub fn description(&self) -> &'static str {
        match self {
@@ -424,7 +427,7 @@ impl MeetingScheduleInfo {
                        end: now,
                    }
                });
-        let meeting_type = MeetingType::from_string(&meeting_time.meeting_type);
+        let meeting_type: MeetingType = meeting_time.meeting_type.parse().unwrap();
        let location = MeetingLocation::from_meeting_time(meeting_time);
        let duration_weeks = date_range.weeks_duration();

@@ -10,8 +10,9 @@ pub struct Range {
    pub high: i32,
 }

-/// Builder for constructing Banner API search queries
+/// Builder for constructing Banner API search queries.
 #[derive(Debug, Clone, Default)]
+#[allow(dead_code)]
 pub struct SearchQuery {
    subject: Option<String>,
    title: Option<String>,
@@ -32,6 +33,7 @@ pub struct SearchQuery {
    course_number_range: Option<Range>,
 }

+#[allow(dead_code)]
 impl SearchQuery {
    /// Creates a new SearchQuery with default values
    pub fn new() -> Self {
@@ -67,7 +69,6 @@ impl SearchQuery {
    }

    /// Adds a keyword to the query
-    #[allow(dead_code)]
    pub fn keyword<S: Into<String>>(mut self, keyword: S) -> Self {
        match &mut self.keywords {
            Some(keywords) => keywords.push(keyword.into()),
@@ -77,63 +78,54 @@ impl SearchQuery {
    }

    /// Sets whether to search for open courses only
-    #[allow(dead_code)]
    pub fn open_only(mut self, open_only: bool) -> Self {
        self.open_only = Some(open_only);
        self
    }

    /// Sets the term part for the query
-    #[allow(dead_code)]
    pub fn term_part(mut self, term_part: Vec<String>) -> Self {
        self.term_part = Some(term_part);
        self
    }

    /// Sets the campuses for the query
-    #[allow(dead_code)]
    pub fn campus(mut self, campus: Vec<String>) -> Self {
        self.campus = Some(campus);
        self
    }

    /// Sets the instructional methods for the query
-    #[allow(dead_code)]
    pub fn instructional_method(mut self, instructional_method: Vec<String>) -> Self {
        self.instructional_method = Some(instructional_method);
        self
    }

    /// Sets the attributes for the query
-    #[allow(dead_code)]
    pub fn attributes(mut self, attributes: Vec<String>) -> Self {
        self.attributes = Some(attributes);
        self
    }

    /// Sets the instructors for the query
-    #[allow(dead_code)]
    pub fn instructor(mut self, instructor: Vec<u64>) -> Self {
        self.instructor = Some(instructor);
        self
    }

    /// Sets the start time for the query
-    #[allow(dead_code)]
    pub fn start_time(mut self, start_time: Duration) -> Self {
        self.start_time = Some(start_time);
        self
    }

    /// Sets the end time for the query
-    #[allow(dead_code)]
    pub fn end_time(mut self, end_time: Duration) -> Self {
        self.end_time = Some(end_time);
        self
    }

    /// Sets the credit range for the query
-    #[allow(dead_code)]
    pub fn credits(mut self, low: i32, high: i32) -> Self {
        self.min_credits = Some(low);
        self.max_credits = Some(high);
@@ -141,14 +133,12 @@ impl SearchQuery {
    }

    /// Sets the minimum credits for the query
-    #[allow(dead_code)]
    pub fn min_credits(mut self, value: i32) -> Self {
        self.min_credits = Some(value);
        self
    }

    /// Sets the maximum credits for the query
-    #[allow(dead_code)]
    pub fn max_credits(mut self, value: i32) -> Self {
        self.max_credits = Some(value);
        self
@@ -161,7 +151,6 @@ impl SearchQuery {
    }

    /// Sets the offset for pagination
-    #[allow(dead_code)]
    pub fn offset(mut self, offset: i32) -> Self {
        self.offset = offset;
        self
@@ -452,7 +452,11 @@ impl SessionPool {
        self.select_term(&term.to_string(), &unique_session_id, &cookie_header)
            .await?;

-        Ok(BannerSession::new(&unique_session_id, jsessionid, ssb_cookie))
+        Ok(BannerSession::new(
+            &unique_session_id,
+            jsessionid,
+            ssb_cookie,
+        ))
    }

    /// Retrieves a list of terms from the Banner API.
@@ -18,7 +18,9 @@ fn nth_weekday_of_month(year: i32, month: u32, weekday: Weekday, n: u32) -> Opti

 /// Compute a consecutive range of dates starting from `start` for `count` days.
 fn date_range(start: NaiveDate, count: i64) -> Vec<NaiveDate> {
-    (0..count).filter_map(|i| start.checked_add_signed(Duration::days(i))).collect()
+    (0..count)
+        .filter_map(|i| start.checked_add_signed(Duration::days(i)))
+        .collect()
 }

 /// Compute university holidays for a given year.
@@ -0,0 +1,462 @@
+//! Shared calendar generation logic for ICS files and Google Calendar URLs.
+//!
+//! Used by both the Discord bot commands and the web API endpoints.
+
+use crate::data::models::DbMeetingTime;
+use chrono::{Datelike, Duration, NaiveDate, NaiveTime, Weekday};
+
+/// Course metadata needed for calendar generation (shared interface between bot and web).
+pub struct CalendarCourse {
+    pub crn: String,
+    pub subject: String,
+    pub course_number: String,
+    pub title: String,
+    pub sequence_number: Option<String>,
+    pub primary_instructor: Option<String>,
+}
+
+impl CalendarCourse {
+    /// Display title like "CS 1083 - Introduction to Computer Science"
+    pub fn display_title(&self) -> String {
+        format!("{} {} - {}", self.subject, self.course_number, self.title)
+    }
+
+    /// Filename-safe identifier: "CS_1083_001"
+    pub fn filename_stem(&self) -> String {
+        format!(
+            "{}_{}{}",
+            self.subject.replace(' ', "_"),
+            self.course_number,
+            self.sequence_number
+                .as_deref()
+                .map(|s| format!("_{s}"))
+                .unwrap_or_default()
+        )
+    }
+}
+
+// ---------------------------------------------------------------------------
+// Date parsing helpers
+// ---------------------------------------------------------------------------
+
+/// Parse a date string in either MM/DD/YYYY or YYYY-MM-DD format.
+fn parse_date(s: &str) -> Option<NaiveDate> {
+    NaiveDate::parse_from_str(s, "%m/%d/%Y")
+        .or_else(|_| NaiveDate::parse_from_str(s, "%Y-%m-%d"))
+        .ok()
+}
+
+/// Parse an HHMM time string into `NaiveTime`.
+fn parse_hhmm(s: &str) -> Option<NaiveTime> {
+    if s.len() != 4 {
+        return None;
+    }
+    let hours = s[..2].parse::<u32>().ok()?;
+    let minutes = s[2..].parse::<u32>().ok()?;
+    NaiveTime::from_hms_opt(hours, minutes, 0)
+}
+
+/// Active weekdays for a meeting time.
+fn active_weekdays(mt: &DbMeetingTime) -> Vec<Weekday> {
+    let mapping: [(bool, Weekday); 7] = [
+        (mt.monday, Weekday::Mon),
+        (mt.tuesday, Weekday::Tue),
+        (mt.wednesday, Weekday::Wed),
+        (mt.thursday, Weekday::Thu),
+        (mt.friday, Weekday::Fri),
+        (mt.saturday, Weekday::Sat),
+        (mt.sunday, Weekday::Sun),
+    ];
+    mapping
+        .iter()
+        .filter(|(active, _)| *active)
+        .map(|(_, day)| *day)
+        .collect()
+}
+
+/// ICS two-letter day code for RRULE BYDAY.
+fn ics_day_code(day: Weekday) -> &'static str {
+    match day {
+        Weekday::Mon => "MO",
+        Weekday::Tue => "TU",
+        Weekday::Wed => "WE",
+        Weekday::Thu => "TH",
+        Weekday::Fri => "FR",
+        Weekday::Sat => "SA",
+        Weekday::Sun => "SU",
+    }
+}
+
+/// Location string from a `DbMeetingTime`.
+fn location_string(mt: &DbMeetingTime) -> String {
+    let building = mt
+        .building_description
+        .as_deref()
+        .or(mt.building.as_deref())
+        .unwrap_or("");
+    let room = mt.room.as_deref().unwrap_or("");
+    let combined = format!("{building} {room}").trim().to_string();
+    if combined.is_empty() {
+        "Online".to_string()
+    } else {
+        combined
+    }
+}
+
+/// Days display string (e.g. "MWF", "TTh").
+fn days_display(mt: &DbMeetingTime) -> String {
+    let weekdays = active_weekdays(mt);
+    if weekdays.is_empty() {
+        return "TBA".to_string();
+    }
+    weekdays
+        .iter()
+        .map(|d| ics_day_code(*d))
+        .collect::<Vec<_>>()
+        .join("")
+}
+
+/// Escape text for ICS property values.
+fn escape_ics(text: &str) -> String {
+    text.replace('\\', "\\\\")
+        .replace(';', "\\;")
+        .replace(',', "\\,")
+        .replace('\n', "\\n")
+        .replace('\r', "")
+}
+
+// ---------------------------------------------------------------------------
+// University holidays (ported from bot/commands/ics.rs)
+// ---------------------------------------------------------------------------
+
+/// Find the nth occurrence of a weekday in a given month/year (1-based).
+fn nth_weekday_of_month(year: i32, month: u32, weekday: Weekday, n: u32) -> Option<NaiveDate> {
+    let first = NaiveDate::from_ymd_opt(year, month, 1)?;
+    let days_ahead = (weekday.num_days_from_monday() as i64
+        - first.weekday().num_days_from_monday() as i64)
+        .rem_euclid(7) as u32;
+    let day = 1 + days_ahead + 7 * (n - 1);
+    NaiveDate::from_ymd_opt(year, month, day)
+}
+
+/// Compute a consecutive range of dates starting from `start` for `count` days.
+fn date_range(start: NaiveDate, count: i64) -> Vec<NaiveDate> {
+    (0..count)
+        .filter_map(|i| start.checked_add_signed(Duration::days(i)))
+        .collect()
+}
+
+/// Compute university holidays for a given year.
+fn compute_holidays_for_year(year: i32) -> Vec<(&'static str, Vec<NaiveDate>)> {
+    let mut holidays = Vec::new();
+
+    // Labor Day: 1st Monday of September
+    if let Some(d) = nth_weekday_of_month(year, 9, Weekday::Mon, 1) {
+        holidays.push(("Labor Day", vec![d]));
+    }
+
+    // Fall Break: Mon-Tue of Columbus Day week
+    if let Some(mon) = nth_weekday_of_month(year, 10, Weekday::Mon, 2) {
+        holidays.push(("Fall Break", date_range(mon, 2)));
+    }
+
+    // Day before Thanksgiving
+    if let Some(thu) = nth_weekday_of_month(year, 11, Weekday::Thu, 4)
+        && let Some(wed) = thu.checked_sub_signed(Duration::days(1))
+    {
+        holidays.push(("Day Before Thanksgiving", vec![wed]));
+    }
+
+    // Thanksgiving: 4th Thursday + Friday
+    if let Some(thu) = nth_weekday_of_month(year, 11, Weekday::Thu, 4) {
+        holidays.push(("Thanksgiving", date_range(thu, 2)));
+    }
+
+    // Winter Holiday: Dec 23-31
+    if let Some(start) = NaiveDate::from_ymd_opt(year, 12, 23) {
+        holidays.push(("Winter Holiday", date_range(start, 9)));
+    }
+
+    // New Year's Day
+    if let Some(d) = NaiveDate::from_ymd_opt(year, 1, 1) {
+        holidays.push(("New Year's Day", vec![d]));
+    }
+
+    // MLK Day: 3rd Monday of January
+    if let Some(d) = nth_weekday_of_month(year, 1, Weekday::Mon, 3) {
+        holidays.push(("MLK Day", vec![d]));
+    }
+
+    // Spring Break: full week starting 2nd Monday of March
+    if let Some(mon) = nth_weekday_of_month(year, 3, Weekday::Mon, 2) {
+        holidays.push(("Spring Break", date_range(mon, 6)));
+    }
+
+    holidays
+}
+
+/// Get holiday dates within a date range that fall on specific weekdays.
+fn holiday_exceptions(start: NaiveDate, end: NaiveDate, weekdays: &[Weekday]) -> Vec<NaiveDate> {
+    let start_year = start.year();
+    let end_year = end.year();
+
+    (start_year..=end_year)
+        .flat_map(compute_holidays_for_year)
+        .flat_map(|(_, dates)| dates)
+        .filter(|&date| date >= start && date <= end && weekdays.contains(&date.weekday()))
+        .collect()
+}
+
+/// Names of excluded holidays (for user-facing messages).
+fn excluded_holiday_names(
+    start: NaiveDate,
+    end: NaiveDate,
+    exceptions: &[NaiveDate],
+) -> Vec<String> {
+    let start_year = start.year();
+    let end_year = end.year();
+    let all_holidays: Vec<_> = (start_year..=end_year)
+        .flat_map(compute_holidays_for_year)
+        .collect();
+
+    let mut names = Vec::new();
+    for (holiday_name, holiday_dates) in &all_holidays {
+        for &exc in exceptions {
+            if holiday_dates.contains(&exc) {
+                names.push(format!("{} ({})", holiday_name, exc.format("%a, %b %d")));
+            }
+        }
+    }
+    names.sort();
+    names.dedup();
+    names
+}
+
+// ---------------------------------------------------------------------------
+// ICS generation
+// ---------------------------------------------------------------------------
+
+/// Result from ICS generation, including the file content and excluded holiday names.
+pub struct IcsResult {
+    pub content: String,
+    pub filename: String,
+    /// Holiday dates excluded via EXDATE rules, for user-facing messages.
+    #[allow(dead_code)]
+    pub excluded_holidays: Vec<String>,
+}
+
+/// Generate an ICS calendar file for a course.
+pub fn generate_ics(
+    course: &CalendarCourse,
+    meeting_times: &[DbMeetingTime],
+) -> Result<IcsResult, anyhow::Error> {
+    let mut ics = String::new();
+    let mut all_excluded = Vec::new();
+
+    // Header
+    ics.push_str("BEGIN:VCALENDAR\r\n");
+    ics.push_str("VERSION:2.0\r\n");
+    ics.push_str("PRODID:-//Banner Bot//Course Calendar//EN\r\n");
+    ics.push_str("CALSCALE:GREGORIAN\r\n");
+    ics.push_str("METHOD:PUBLISH\r\n");
+    ics.push_str(&format!(
+        "X-WR-CALNAME:{}\r\n",
+        escape_ics(&course.display_title())
+    ));
+
+    for (index, mt) in meeting_times.iter().enumerate() {
+        let (event, holidays) = generate_ics_event(course, mt, index)?;
+        ics.push_str(&event);
+        all_excluded.extend(holidays);
+    }
+
+    ics.push_str("END:VCALENDAR\r\n");
+
+    Ok(IcsResult {
+        content: ics,
+        filename: format!("{}.ics", course.filename_stem()),
+        excluded_holidays: all_excluded,
+    })
+}
+
+/// Generate a single VEVENT for one meeting time.
+fn generate_ics_event(
+    course: &CalendarCourse,
+    mt: &DbMeetingTime,
+    index: usize,
+) -> Result<(String, Vec<String>), anyhow::Error> {
+    let start_date = parse_date(&mt.start_date)
+        .ok_or_else(|| anyhow::anyhow!("Invalid start_date: {}", mt.start_date))?;
+    let end_date = parse_date(&mt.end_date)
+        .ok_or_else(|| anyhow::anyhow!("Invalid end_date: {}", mt.end_date))?;
+
+    let start_time = mt.begin_time.as_deref().and_then(parse_hhmm);
+    let end_time = mt.end_time.as_deref().and_then(parse_hhmm);
+
+    // DTSTART/DTEND: first occurrence with time, or all-day on start_date
+    let (dtstart, dtend) = match (start_time, end_time) {
+        (Some(st), Some(et)) => {
+            let s = start_date.and_time(st).and_utc();
+            let e = start_date.and_time(et).and_utc();
+            (
+                s.format("%Y%m%dT%H%M%SZ").to_string(),
+                e.format("%Y%m%dT%H%M%SZ").to_string(),
+            )
+        }
+        _ => {
+            let s = start_date.and_hms_opt(0, 0, 0).unwrap().and_utc();
+            let e = start_date.and_hms_opt(0, 0, 0).unwrap().and_utc();
+            (
+                s.format("%Y%m%dT%H%M%SZ").to_string(),
+                e.format("%Y%m%dT%H%M%SZ").to_string(),
+            )
+        }
+    };
+
+    let event_title = if index > 0 {
+        format!("{} (Meeting {})", course.display_title(), index + 1)
+    } else {
+        course.display_title()
+    };
+
+    let instructor = course.primary_instructor.as_deref().unwrap_or("Staff");
+
+    let description = format!(
+        "CRN: {}\\nInstructor: {}\\nDays: {}\\nMeeting Type: {}",
+        course.crn,
+        instructor,
+        days_display(mt),
+        mt.meeting_type,
+    );
+
+    let location = location_string(mt);
+
+    let uid = format!(
+        "{}-{}-{}@banner-bot.local",
+        course.crn,
+        index,
+        start_date
+            .and_hms_opt(0, 0, 0)
+            .unwrap()
+            .and_utc()
+            .timestamp()
+    );
+
+    let mut event = String::new();
+    event.push_str("BEGIN:VEVENT\r\n");
+    event.push_str(&format!("UID:{uid}\r\n"));
+    event.push_str(&format!("DTSTART:{dtstart}\r\n"));
+    event.push_str(&format!("DTEND:{dtend}\r\n"));
+    event.push_str(&format!("SUMMARY:{}\r\n", escape_ics(&event_title)));
+    event.push_str(&format!("DESCRIPTION:{}\r\n", escape_ics(&description)));
+    event.push_str(&format!("LOCATION:{}\r\n", escape_ics(&location)));
+
+    let weekdays = active_weekdays(mt);
+    let mut holiday_names = Vec::new();
+
+    if let (false, Some(st)) = (weekdays.is_empty(), start_time) {
+        let by_day: Vec<&str> = weekdays.iter().map(|d| ics_day_code(*d)).collect();
+        let until = end_date.format("%Y%m%dT000000Z").to_string();
+
+        event.push_str(&format!(
+            "RRULE:FREQ=WEEKLY;BYDAY={};UNTIL={}\r\n",
+            by_day.join(","),
+            until,
+        ));
+
+        // Holiday exceptions
+        let exceptions = holiday_exceptions(start_date, end_date, &weekdays);
+        if !exceptions.is_empty() {
+            let start_utc = start_date.and_time(st).and_utc();
+            let exdates: Vec<String> = exceptions
+                .iter()
+                .map(|&d| {
+                    d.and_time(start_utc.time())
+                        .and_utc()
+                        .format("%Y%m%dT%H%M%SZ")
+                        .to_string()
+                })
+                .collect();
+            event.push_str(&format!("EXDATE:{}\r\n", exdates.join(",")));
+        }
+
+        holiday_names = excluded_holiday_names(start_date, end_date, &exceptions);
+    }
+
+    event.push_str("END:VEVENT\r\n");
+    Ok((event, holiday_names))
+}
+
+// ---------------------------------------------------------------------------
+// Google Calendar URL generation
+// ---------------------------------------------------------------------------
+
+/// Generate a Google Calendar "add event" URL for a single meeting time.
+pub fn generate_gcal_url(
+    course: &CalendarCourse,
+    mt: &DbMeetingTime,
+) -> Result<String, anyhow::Error> {
+    let start_date = parse_date(&mt.start_date)
+        .ok_or_else(|| anyhow::anyhow!("Invalid start_date: {}", mt.start_date))?;
+    let end_date = parse_date(&mt.end_date)
+        .ok_or_else(|| anyhow::anyhow!("Invalid end_date: {}", mt.end_date))?;
+
+    let start_time = mt.begin_time.as_deref().and_then(parse_hhmm);
+    let end_time = mt.end_time.as_deref().and_then(parse_hhmm);
+
+    let dates_text = match (start_time, end_time) {
+        (Some(st), Some(et)) => {
+            let s = start_date.and_time(st);
+            let e = start_date.and_time(et);
+            format!(
+                "{}/{}",
+                s.format("%Y%m%dT%H%M%S"),
+                e.format("%Y%m%dT%H%M%S")
+            )
+        }
+        _ => {
+            let s = start_date.format("%Y%m%d").to_string();
+            format!("{s}/{s}")
+        }
+    };
+
+    let instructor = course.primary_instructor.as_deref().unwrap_or("Staff");
+
+    let details = format!(
+        "CRN: {}\nInstructor: {}\nDays: {}",
+        course.crn,
+        instructor,
+        days_display(mt),
+    );
+
+    let location = location_string(mt);
+
+    let weekdays = active_weekdays(mt);
+    let recur = if !weekdays.is_empty() && start_time.is_some() {
+        let by_day: Vec<&str> = weekdays.iter().map(|d| ics_day_code(*d)).collect();
+        let until = end_date.format("%Y%m%dT000000Z").to_string();
+        format!(
+            "RRULE:FREQ=WEEKLY;BYDAY={};UNTIL={}",
+            by_day.join(","),
+            until
+        )
+    } else {
+        String::new()
+    };
+
+    let course_text = course.display_title();
+
+    let params: Vec<(&str, &str)> = vec![
+        ("action", "TEMPLATE"),
+        ("text", &course_text),
+        ("dates", &dates_text),
+        ("details", &details),
+        ("location", &location),
+        ("trp", "true"),
+        ("ctz", "America/Chicago"),
+        ("recur", &recur),
+    ];
+
+    let url = url::Url::parse_with_params("https://calendar.google.com/calendar/render", &params)?;
+    Ok(url.to_string())
+}
@@ -2,34 +2,16 @@ use clap::Parser;

 /// Banner Discord Bot - Course availability monitoring
 ///
-/// This application runs multiple services that can be controlled via CLI arguments:
+/// This application runs all services:
 /// - bot: Discord bot for course monitoring commands
 /// - web: HTTP server for web interface and API
 /// - scraper: Background service for scraping course data
-///
-/// Use --services to specify which services to run, or --disable-services to exclude specific services.
 #[derive(Parser, Debug)]
 #[command(author, version, about, long_about = None)]
 pub struct Args {
    /// Log formatter to use
    #[arg(long, value_enum, default_value_t = default_tracing_format())]
    pub tracing: TracingFormat,
-
-    /// Services to run (comma-separated). Default: all services
-    ///
-    /// Examples:
-    ///   --services bot,web    # Run only bot and web services
-    ///   --services scraper    # Run only the scraper service
-    #[arg(long, value_delimiter = ',', conflicts_with = "disable_services")]
-    pub services: Option<Vec<ServiceName>>,
-
-    /// Services to disable (comma-separated)
-    ///
-    /// Examples:
-    ///   --disable-services bot        # Run web and scraper only
-    ///   --disable-services bot,web    # Run only the scraper service
-    #[arg(long, value_delimiter = ',', conflicts_with = "services")]
-    pub disable_services: Option<Vec<ServiceName>>,
 }

 #[derive(clap::ValueEnum, Clone, Debug)]
@@ -66,34 +48,6 @@ impl ServiceName {
    }
 }

-/// Determine which services should be enabled based on CLI arguments
-pub fn determine_enabled_services(args: &Args) -> Result<Vec<ServiceName>, anyhow::Error> {
-    match (&args.services, &args.disable_services) {
-        (Some(services), None) => {
-            // User specified which services to run
-            Ok(services.clone())
-        }
-        (None, Some(disabled)) => {
-            // User specified which services to disable
-            let enabled: Vec<ServiceName> = ServiceName::all()
-                .into_iter()
-                .filter(|s| !disabled.contains(s))
-                .collect();
-            Ok(enabled)
-        }
-        (None, None) => {
-            // Default: run all services
-            Ok(ServiceName::all())
-        }
-        (Some(_), Some(_)) => {
-            // This should be prevented by clap's conflicts_with, but just in case
-            Err(anyhow::anyhow!(
-                "Cannot specify both --services and --disable-services"
-            ))
-        }
-    }
-}
-
 #[cfg(debug_assertions)]
 const DEFAULT_TRACING_FORMAT: TracingFormat = TracingFormat::Pretty;
 #[cfg(not(debug_assertions))]
@@ -107,64 +61,6 @@ fn default_tracing_format() -> TracingFormat {
 mod tests {
    use super::*;

-    fn args_with_services(
-        services: Option<Vec<ServiceName>>,
-        disable: Option<Vec<ServiceName>>,
-    ) -> Args {
-        Args {
-            tracing: TracingFormat::Pretty,
-            services,
-            disable_services: disable,
-        }
-    }
-
-    #[test]
-    fn test_default_enables_all_services() {
-        let result = determine_enabled_services(&args_with_services(None, None)).unwrap();
-        assert_eq!(result.len(), 3);
-    }
-
-    #[test]
-    fn test_explicit_services_only_those() {
-        let result =
-            determine_enabled_services(&args_with_services(Some(vec![ServiceName::Web]), None))
-                .unwrap();
-        assert_eq!(result.len(), 1);
-        assert_eq!(result[0].as_str(), "web");
-    }
-
-    #[test]
-    fn test_disable_bot_leaves_web_and_scraper() {
-        let result =
-            determine_enabled_services(&args_with_services(None, Some(vec![ServiceName::Bot])))
-                .unwrap();
-        assert_eq!(result.len(), 2);
-        assert!(result.iter().all(|s| s.as_str() != "bot"));
-    }
-
-    #[test]
-    fn test_disable_all_leaves_empty() {
-        let result = determine_enabled_services(&args_with_services(
-            None,
-            Some(vec![
-                ServiceName::Bot,
-                ServiceName::Web,
-                ServiceName::Scraper,
-            ]),
-        ))
-        .unwrap();
-        assert!(result.is_empty());
-    }
-
-    #[test]
-    fn test_both_specified_returns_error() {
-        let result = determine_enabled_services(&args_with_services(
-            Some(vec![ServiceName::Web]),
-            Some(vec![ServiceName::Bot]),
-        ));
-        assert!(result.is_err());
-    }
-
    #[test]
    fn test_service_name_as_str() {
        assert_eq!(ServiceName::Bot.as_str(), "bot");
@@ -47,6 +47,19 @@ pub struct Config {
    /// Rate limiting configuration for Banner API requests
    #[serde(default = "default_rate_limiting")]
    pub rate_limiting: RateLimitingConfig,
+
+    /// Discord OAuth2 client ID for web authentication
+    #[serde(deserialize_with = "deserialize_string_or_uint")]
+    pub discord_client_id: String,
+    /// Discord OAuth2 client secret for web authentication
+    pub discord_client_secret: String,
+    /// Optional base URL override for OAuth2 redirect (e.g. "https://banner.xevion.dev").
+    /// When unset, the redirect URI is derived from the incoming request's Origin/Host.
+    #[serde(default)]
+    pub discord_redirect_uri: Option<String>,
+    /// Discord user ID to seed as initial admin on startup (optional)
+    #[serde(default)]
+    pub admin_discord_id: Option<u64>,
 }

 /// Default log level of "info"
@@ -216,6 +229,43 @@ where
    deserializer.deserialize_any(DurationVisitor)
 }

+/// Deserializes a value that may arrive as either a string or unsigned integer.
+///
+/// Figment's env provider infers types from raw values, so numeric-looking strings
+/// like Discord client IDs get parsed as integers. This accepts both forms.
+fn deserialize_string_or_uint<'de, D>(deserializer: D) -> Result<String, D::Error>
+where
+    D: Deserializer<'de>,
+{
+    use serde::de::Visitor;
+
+    struct StringOrUintVisitor;
+
+    impl<'de> Visitor<'de> for StringOrUintVisitor {
+        type Value = String;
+
+        fn expecting(&self, formatter: &mut std::fmt::Formatter) -> std::fmt::Result {
+            formatter.write_str("a string or unsigned integer")
+        }
+
+        fn visit_str<E>(self, value: &str) -> Result<Self::Value, E>
+        where
+            E: serde::de::Error,
+        {
+            Ok(value.to_owned())
+        }
+
+        fn visit_u64<E>(self, value: u64) -> Result<Self::Value, E>
+        where
+            E: serde::de::Error,
+        {
+            Ok(value.to_string())
+        }
+    }
+
+    deserializer.deserialize_any(StringOrUintVisitor)
+}
+
 #[cfg(test)]
 mod tests {
    use super::*;
@@ -1,98 +1,593 @@
 //! Batch database operations for improved performance.

 use crate::banner::Course;
+use crate::data::models::{DbMeetingTime, UpsertCounts};
 use crate::error::Result;
+use sqlx::PgConnection;
 use sqlx::PgPool;
+use std::collections::{HashMap, HashSet};
 use std::time::Instant;
 use tracing::info;

+/// Convert a Banner API course's meeting times to the DB JSONB shape.
+fn to_db_meeting_times(course: &Course) -> serde_json::Value {
+    let meetings: Vec<DbMeetingTime> = course
+        .meetings_faculty
+        .iter()
+        .map(|mf| {
+            let mt = &mf.meeting_time;
+            DbMeetingTime {
+                begin_time: mt.begin_time.clone(),
+                end_time: mt.end_time.clone(),
+                start_date: mt.start_date.clone(),
+                end_date: mt.end_date.clone(),
+                monday: mt.monday,
+                tuesday: mt.tuesday,
+                wednesday: mt.wednesday,
+                thursday: mt.thursday,
+                friday: mt.friday,
+                saturday: mt.saturday,
+                sunday: mt.sunday,
+                building: mt.building.clone(),
+                building_description: mt.building_description.clone(),
+                room: mt.room.clone(),
+                campus: mt.campus.clone(),
+                meeting_type: mt.meeting_type.clone(),
+                meeting_schedule_type: mt.meeting_schedule_type.clone(),
+            }
+        })
+        .collect();
+    serde_json::to_value(meetings).unwrap_or_default()
+}
+
+/// Convert a Banner API course's section attributes to a JSONB array of code strings.
+fn to_db_attributes(course: &Course) -> serde_json::Value {
+    let codes: Vec<&str> = course
+        .section_attributes
+        .iter()
+        .map(|a| a.code.as_str())
+        .collect();
+    serde_json::to_value(codes).unwrap_or_default()
+}
+
+/// Extract the campus code from the first meeting time (Banner doesn't put it on the course directly).
+fn extract_campus_code(course: &Course) -> Option<String> {
+    course
+        .meetings_faculty
+        .first()
+        .and_then(|mf| mf.meeting_time.campus.clone())
+}
+
+// ---------------------------------------------------------------------------
+// Task 1: UpsertDiffRow — captures pre- and post-upsert state for diffing
+// ---------------------------------------------------------------------------
+
+/// Row returned by the CTE-based upsert query, carrying both old and new values
+/// for every auditable field. `old_id` is `None` for fresh inserts.
+#[derive(sqlx::FromRow, Debug)]
+struct UpsertDiffRow {
+    id: i32,
+    old_id: Option<i32>,
+    crn: String,
+    term_code: String,
+
+    // enrollment fields
+    old_enrollment: Option<i32>,
+    new_enrollment: i32,
+    old_max_enrollment: Option<i32>,
+    new_max_enrollment: i32,
+    old_wait_count: Option<i32>,
+    new_wait_count: i32,
+    old_wait_capacity: Option<i32>,
+    new_wait_capacity: i32,
+
+    // text fields (non-nullable in DB)
+    old_subject: Option<String>,
+    new_subject: String,
+    old_course_number: Option<String>,
+    new_course_number: String,
+    old_title: Option<String>,
+    new_title: String,
+
+    // nullable text fields
+    old_sequence_number: Option<String>,
+    new_sequence_number: Option<String>,
+    old_part_of_term: Option<String>,
+    new_part_of_term: Option<String>,
+    old_instructional_method: Option<String>,
+    new_instructional_method: Option<String>,
+    old_campus: Option<String>,
+    new_campus: Option<String>,
+
+    // nullable int fields
+    old_credit_hours: Option<i32>,
+    new_credit_hours: Option<i32>,
+    old_credit_hour_low: Option<i32>,
+    new_credit_hour_low: Option<i32>,
+    old_credit_hour_high: Option<i32>,
+    new_credit_hour_high: Option<i32>,
+
+    // cross-list fields
+    old_cross_list: Option<String>,
+    new_cross_list: Option<String>,
+    old_cross_list_capacity: Option<i32>,
+    new_cross_list_capacity: Option<i32>,
+    old_cross_list_count: Option<i32>,
+    new_cross_list_count: Option<i32>,
+
+    // link fields
+    old_link_identifier: Option<String>,
+    new_link_identifier: Option<String>,
+    old_is_section_linked: Option<bool>,
+    new_is_section_linked: Option<bool>,
+
+    // JSONB fields
+    old_meeting_times: Option<serde_json::Value>,
+    new_meeting_times: serde_json::Value,
+    old_attributes: Option<serde_json::Value>,
+    new_attributes: serde_json::Value,
+}
+
+// ---------------------------------------------------------------------------
+// Task 3: Entry types and diff logic
+// ---------------------------------------------------------------------------
+
+struct AuditEntry {
+    course_id: i32,
+    field_changed: &'static str,
+    old_value: String,
+    new_value: String,
+}
+
+struct MetricEntry {
+    course_id: i32,
+    enrollment: i32,
+    wait_count: i32,
+    seats_available: i32,
+}
+
+/// Compare old vs new for a single field, pushing an `AuditEntry` when they differ.
+///
+/// Three variants:
+/// - `diff_field!(audits, row, field_name, old_field, new_field)` — `Option<T>` old vs `T` new
+/// - `diff_field!(opt audits, row, field_name, old_field, new_field)` — `Option<T>` old vs `Option<T>` new
+/// - `diff_field!(json audits, row, field_name, old_field, new_field)` — `Option<Value>` old vs `Value` new
+///
+/// All variants skip when `old_id` is None (fresh insert).
+macro_rules! diff_field {
+    // Standard: Option<T> old vs T new (non-nullable columns)
+    ($audits:ident, $row:ident, $field:expr, $old:ident, $new:ident) => {
+        if $row.old_id.is_some() {
+            let old_str = $row
+                .$old
+                .as_ref()
+                .map(|v| v.to_string())
+                .unwrap_or_default();
+            let new_str = $row.$new.to_string();
+            if old_str != new_str {
+                $audits.push(AuditEntry {
+                    course_id: $row.id,
+                    field_changed: $field,
+                    old_value: old_str,
+                    new_value: new_str,
+                });
+            }
+        }
+    };
+    // Nullable: Option<T> old vs Option<T> new
+    (opt $audits:ident, $row:ident, $field:expr, $old:ident, $new:ident) => {
+        if $row.old_id.is_some() {
+            let old_str = $row
+                .$old
+                .as_ref()
+                .map(|v| v.to_string())
+                .unwrap_or_default();
+            let new_str = $row
+                .$new
+                .as_ref()
+                .map(|v| v.to_string())
+                .unwrap_or_default();
+            if old_str != new_str {
+                $audits.push(AuditEntry {
+                    course_id: $row.id,
+                    field_changed: $field,
+                    old_value: old_str,
+                    new_value: new_str,
+                });
+            }
+        }
+    };
+    // JSONB: Option<Value> old vs Value new
+    (json $audits:ident, $row:ident, $field:expr, $old:ident, $new:ident) => {
+        if $row.old_id.is_some() {
+            let old_val = $row
+                .$old
+                .as_ref()
+                .cloned()
+                .unwrap_or(serde_json::Value::Null);
+            let new_val = &$row.$new;
+            if old_val != *new_val {
+                $audits.push(AuditEntry {
+                    course_id: $row.id,
+                    field_changed: $field,
+                    old_value: old_val.to_string(),
+                    new_value: new_val.to_string(),
+                });
+            }
+        }
+    };
+}
+
+/// Compute audit entries (field-level diffs) and metric entries from upsert diff rows.
+fn compute_diffs(rows: &[UpsertDiffRow]) -> (Vec<AuditEntry>, Vec<MetricEntry>) {
+    let mut audits = Vec::new();
+    let mut metrics = Vec::new();
+
+    for row in rows {
+        // Non-nullable fields
+        diff_field!(audits, row, "enrollment", old_enrollment, new_enrollment);
+        diff_field!(
+            audits,
+            row,
+            "max_enrollment",
+            old_max_enrollment,
+            new_max_enrollment
+        );
+        diff_field!(audits, row, "wait_count", old_wait_count, new_wait_count);
+        diff_field!(
+            audits,
+            row,
+            "wait_capacity",
+            old_wait_capacity,
+            new_wait_capacity
+        );
+        diff_field!(audits, row, "subject", old_subject, new_subject);
+        diff_field!(
+            audits,
+            row,
+            "course_number",
+            old_course_number,
+            new_course_number
+        );
+        diff_field!(audits, row, "title", old_title, new_title);
+
+        // Nullable text fields
+        diff_field!(opt audits, row, "sequence_number", old_sequence_number, new_sequence_number);
+        diff_field!(opt audits, row, "part_of_term", old_part_of_term, new_part_of_term);
+        diff_field!(opt audits, row, "instructional_method", old_instructional_method, new_instructional_method);
+        diff_field!(opt audits, row, "campus", old_campus, new_campus);
+
+        // Nullable int fields
+        diff_field!(opt audits, row, "credit_hours", old_credit_hours, new_credit_hours);
+        diff_field!(opt audits, row, "credit_hour_low", old_credit_hour_low, new_credit_hour_low);
+        diff_field!(opt audits, row, "credit_hour_high", old_credit_hour_high, new_credit_hour_high);
+
+        // Cross-list fields
+        diff_field!(opt audits, row, "cross_list", old_cross_list, new_cross_list);
+        diff_field!(opt audits, row, "cross_list_capacity", old_cross_list_capacity, new_cross_list_capacity);
+        diff_field!(opt audits, row, "cross_list_count", old_cross_list_count, new_cross_list_count);
+
+        // Link fields
+        diff_field!(opt audits, row, "link_identifier", old_link_identifier, new_link_identifier);
+        diff_field!(opt audits, row, "is_section_linked", old_is_section_linked, new_is_section_linked);
+
+        // JSONB fields
+        diff_field!(json audits, row, "meeting_times", old_meeting_times, new_meeting_times);
+        diff_field!(json audits, row, "attributes", old_attributes, new_attributes);
+
+        // Emit a metric entry when enrollment/wait_count/max_enrollment changed
+        // Skip fresh inserts (no old data to compare against)
+        let enrollment_changed = row.old_id.is_some()
+            && (row.old_enrollment != Some(row.new_enrollment)
+                || row.old_wait_count != Some(row.new_wait_count)
+                || row.old_max_enrollment != Some(row.new_max_enrollment));
+
+        if enrollment_changed {
+            metrics.push(MetricEntry {
+                course_id: row.id,
+                enrollment: row.new_enrollment,
+                wait_count: row.new_wait_count,
+                seats_available: row.new_max_enrollment - row.new_enrollment,
+            });
+        }
+    }
+
+    (audits, metrics)
+}
+
+// ---------------------------------------------------------------------------
+// Task 4: Batch insert functions for audits and metrics
+// ---------------------------------------------------------------------------
+
+async fn insert_audits(audits: &[AuditEntry], conn: &mut PgConnection) -> Result<()> {
+    if audits.is_empty() {
+        return Ok(());
+    }
+
+    let course_ids: Vec<i32> = audits.iter().map(|a| a.course_id).collect();
+    let fields: Vec<&str> = audits.iter().map(|a| a.field_changed).collect();
+    let old_values: Vec<&str> = audits.iter().map(|a| a.old_value.as_str()).collect();
+    let new_values: Vec<&str> = audits.iter().map(|a| a.new_value.as_str()).collect();
+
+    sqlx::query(
+        r#"
+        INSERT INTO course_audits (course_id, timestamp, field_changed, old_value, new_value)
+        SELECT v.course_id, NOW(), v.field_changed, v.old_value, v.new_value
+        FROM UNNEST($1::int4[], $2::text[], $3::text[], $4::text[])
+            AS v(course_id, field_changed, old_value, new_value)
+        "#,
+    )
+    .bind(&course_ids)
+    .bind(&fields)
+    .bind(&old_values)
+    .bind(&new_values)
+    .execute(&mut *conn)
+    .await
+    .map_err(|e| anyhow::anyhow!("Failed to batch insert course_audits: {}", e))?;
+
+    Ok(())
+}
+
+async fn insert_metrics(metrics: &[MetricEntry], conn: &mut PgConnection) -> Result<()> {
+    if metrics.is_empty() {
+        return Ok(());
+    }
+
+    let course_ids: Vec<i32> = metrics.iter().map(|m| m.course_id).collect();
+    let enrollments: Vec<i32> = metrics.iter().map(|m| m.enrollment).collect();
+    let wait_counts: Vec<i32> = metrics.iter().map(|m| m.wait_count).collect();
+    let seats_available: Vec<i32> = metrics.iter().map(|m| m.seats_available).collect();
+
+    sqlx::query(
+        r#"
+        INSERT INTO course_metrics (course_id, timestamp, enrollment, wait_count, seats_available)
+        SELECT v.course_id, NOW(), v.enrollment, v.wait_count, v.seats_available
+        FROM UNNEST($1::int4[], $2::int4[], $3::int4[], $4::int4[])
+            AS v(course_id, enrollment, wait_count, seats_available)
+        "#,
+    )
+    .bind(&course_ids)
+    .bind(&enrollments)
+    .bind(&wait_counts)
+    .bind(&seats_available)
+    .execute(&mut *conn)
+    .await
+    .map_err(|e| anyhow::anyhow!("Failed to batch insert course_metrics: {}", e))?;
+
+    Ok(())
+}
+
+// ---------------------------------------------------------------------------
+// Core upsert functions (updated to use &mut PgConnection)
+// ---------------------------------------------------------------------------
+
 /// Batch upsert courses in a single database query.
 ///
-/// This function performs a bulk INSERT...ON CONFLICT DO UPDATE for all courses
-/// in a single round-trip to the database, significantly reducing overhead compared
-/// to individual inserts.
+/// Performs a bulk INSERT...ON CONFLICT DO UPDATE for all courses, including
+/// new fields (meeting times, attributes, instructor data). Captures pre-update
+/// state for audit/metric tracking, all within a single transaction.
 ///
 /// # Performance
-/// - Reduces N database round-trips to 1
+/// - Reduces N database round-trips to 5 (old-data CTE + upsert, audits, metrics, instructors, junction)
 /// - Typical usage: 50-200 courses per batch
-/// - PostgreSQL parameter limit: 65,535 (we use ~10 per course)
-///
-/// # Arguments
-/// * `courses` - Slice of Course structs from the Banner API
-/// * `db_pool` - PostgreSQL connection pool
-///
-/// # Returns
-/// * `Ok(())` on success
-/// * `Err(_)` if the database operation fails
-///
-/// # Example
-/// ```no_run
-/// use banner::data::batch::batch_upsert_courses;
-/// use banner::banner::Course;
-/// use sqlx::PgPool;
-///
-/// async fn example(courses: &[Course], pool: &PgPool) -> anyhow::Result<()> {
-///     batch_upsert_courses(courses, pool).await?;
-///     Ok(())
-/// }
-/// ```
-pub async fn batch_upsert_courses(courses: &[Course], db_pool: &PgPool) -> Result<()> {
-    // Early return for empty batches
+pub async fn batch_upsert_courses(courses: &[Course], db_pool: &PgPool) -> Result<UpsertCounts> {
    if courses.is_empty() {
        info!("No courses to upsert, skipping batch operation");
-        return Ok(());
+        return Ok(UpsertCounts::default());
    }

    let start = Instant::now();
    let course_count = courses.len();

-    // Extract course fields into vectors for UNNEST
+    let mut tx = db_pool.begin().await?;
+
+    // Step 1: Upsert courses with CTE, returning diff rows
+    let diff_rows = upsert_courses(courses, &mut tx).await?;
+
+    // Step 2: Build (crn, term_code) → course_id map for instructor linking.
+    // RETURNING order from INSERT ... ON CONFLICT is not guaranteed to match
+    // the input array order, so we must key by (crn, term_code) rather than
+    // relying on positional correspondence.
+    let crn_term_to_id: HashMap<(&str, &str), i32> = diff_rows
+        .iter()
+        .map(|r| ((r.crn.as_str(), r.term_code.as_str()), r.id))
+        .collect();
+
+    // Step 3: Compute audit/metric diffs
+    let (audits, metrics) = compute_diffs(&diff_rows);
+
+    // Count courses that had at least one field change (existing rows only)
+    let changed_ids: HashSet<i32> = audits.iter().map(|a| a.course_id).collect();
+    let existing_count = diff_rows.iter().filter(|r| r.old_id.is_some()).count() as i32;
+    let courses_changed = changed_ids.len() as i32;
+
+    let counts = UpsertCounts {
+        courses_fetched: course_count as i32,
+        courses_changed,
+        courses_unchanged: existing_count - courses_changed,
+        audits_generated: audits.len() as i32,
+        metrics_generated: metrics.len() as i32,
+    };
+
+    // Step 4: Insert audits and metrics
+    insert_audits(&audits, &mut tx).await?;
+    insert_metrics(&metrics, &mut tx).await?;
+
+    // Step 5: Upsert instructors (returns email -> id map)
+    let email_to_id = upsert_instructors(courses, &mut tx).await?;
+
+    // Step 6: Link courses to instructors via junction table
+    upsert_course_instructors(courses, &crn_term_to_id, &email_to_id, &mut tx).await?;
+
+    tx.commit().await?;
+
+    let duration = start.elapsed();
+    info!(
+        courses_count = course_count,
+        courses_changed = counts.courses_changed,
+        courses_unchanged = counts.courses_unchanged,
+        audit_entries = counts.audits_generated,
+        metric_entries = counts.metrics_generated,
+        duration_ms = duration.as_millis(),
+        "Batch upserted courses with instructors, audits, and metrics"
+    );
+
+    Ok(counts)
+}
+
+// ---------------------------------------------------------------------------
+// Task 2: CTE-based upsert returning old+new values
+// ---------------------------------------------------------------------------
+
+/// Upsert all courses and return diff rows with old and new values for auditing.
+async fn upsert_courses(courses: &[Course], conn: &mut PgConnection) -> Result<Vec<UpsertDiffRow>> {
    let crns: Vec<&str> = courses
        .iter()
        .map(|c| c.course_reference_number.as_str())
        .collect();
-
    let subjects: Vec<&str> = courses.iter().map(|c| c.subject.as_str()).collect();
-
    let course_numbers: Vec<&str> = courses.iter().map(|c| c.course_number.as_str()).collect();
-
    let titles: Vec<&str> = courses.iter().map(|c| c.course_title.as_str()).collect();
-
    let term_codes: Vec<&str> = courses.iter().map(|c| c.term.as_str()).collect();
-
    let enrollments: Vec<i32> = courses.iter().map(|c| c.enrollment).collect();
-
    let max_enrollments: Vec<i32> = courses.iter().map(|c| c.maximum_enrollment).collect();
-
    let wait_counts: Vec<i32> = courses.iter().map(|c| c.wait_count).collect();
-
    let wait_capacities: Vec<i32> = courses.iter().map(|c| c.wait_capacity).collect();

-    // Perform batch upsert using UNNEST for efficient bulk insertion
-    let result = sqlx::query(
+    // New scalar fields
+    let sequence_numbers: Vec<Option<&str>> = courses
+        .iter()
+        .map(|c| Some(c.sequence_number.as_str()))
+        .collect();
+    let parts_of_term: Vec<Option<&str>> = courses
+        .iter()
+        .map(|c| Some(c.part_of_term.as_str()))
+        .collect();
+    let instructional_methods: Vec<Option<&str>> = courses
+        .iter()
+        .map(|c| Some(c.instructional_method.as_str()))
+        .collect();
+    let campuses: Vec<Option<String>> = courses.iter().map(extract_campus_code).collect();
+    let credit_hours: Vec<Option<i32>> = courses.iter().map(|c| c.credit_hours).collect();
+    let credit_hour_lows: Vec<Option<i32>> = courses.iter().map(|c| c.credit_hour_low).collect();
+    let credit_hour_highs: Vec<Option<i32>> = courses.iter().map(|c| c.credit_hour_high).collect();
+    let cross_lists: Vec<Option<&str>> = courses.iter().map(|c| c.cross_list.as_deref()).collect();
+    let cross_list_capacities: Vec<Option<i32>> =
+        courses.iter().map(|c| c.cross_list_capacity).collect();
+    let cross_list_counts: Vec<Option<i32>> = courses.iter().map(|c| c.cross_list_count).collect();
+    let link_identifiers: Vec<Option<&str>> = courses
+        .iter()
+        .map(|c| c.link_identifier.as_deref())
+        .collect();
+    let is_section_linkeds: Vec<Option<bool>> =
+        courses.iter().map(|c| Some(c.is_section_linked)).collect();
+
+    // JSONB fields
+    let meeting_times_json: Vec<serde_json::Value> =
+        courses.iter().map(to_db_meeting_times).collect();
+    let attributes_json: Vec<serde_json::Value> = courses.iter().map(to_db_attributes).collect();
+
+    let rows = sqlx::query_as::<_, UpsertDiffRow>(
        r#"
-        INSERT INTO courses (
-            crn, subject, course_number, title, term_code,
-            enrollment, max_enrollment, wait_count, wait_capacity, last_scraped_at
+        WITH old_data AS (
+            SELECT id, enrollment, max_enrollment, wait_count, wait_capacity,
+                   subject, course_number, title,
+                   sequence_number, part_of_term, instructional_method, campus,
+                   credit_hours, credit_hour_low, credit_hour_high,
+                   cross_list, cross_list_capacity, cross_list_count,
+                   link_identifier, is_section_linked,
+                   meeting_times, attributes,
+                   crn, term_code
+            FROM courses
+            WHERE (crn, term_code) IN (SELECT * FROM UNNEST($1::text[], $5::text[]))
+        ),
+        upserted AS (
+            INSERT INTO courses (
+                crn, subject, course_number, title, term_code,
+                enrollment, max_enrollment, wait_count, wait_capacity, last_scraped_at,
+                sequence_number, part_of_term, instructional_method, campus,
+                credit_hours, credit_hour_low, credit_hour_high,
+                cross_list, cross_list_capacity, cross_list_count,
+                link_identifier, is_section_linked,
+                meeting_times, attributes
+            )
+            SELECT
+                v.crn, v.subject, v.course_number, v.title, v.term_code,
+                v.enrollment, v.max_enrollment, v.wait_count, v.wait_capacity, NOW(),
+                v.sequence_number, v.part_of_term, v.instructional_method, v.campus,
+                v.credit_hours, v.credit_hour_low, v.credit_hour_high,
+                v.cross_list, v.cross_list_capacity, v.cross_list_count,
+                v.link_identifier, v.is_section_linked,
+                v.meeting_times, v.attributes
+            FROM UNNEST(
+                $1::text[], $2::text[], $3::text[], $4::text[], $5::text[],
+                $6::int4[], $7::int4[], $8::int4[], $9::int4[],
+                $10::text[], $11::text[], $12::text[], $13::text[],
+                $14::int4[], $15::int4[], $16::int4[],
+                $17::text[], $18::int4[], $19::int4[],
+                $20::text[], $21::bool[],
+                $22::jsonb[], $23::jsonb[]
+            ) AS v(
+                crn, subject, course_number, title, term_code,
+                enrollment, max_enrollment, wait_count, wait_capacity,
+                sequence_number, part_of_term, instructional_method, campus,
+                credit_hours, credit_hour_low, credit_hour_high,
+                cross_list, cross_list_capacity, cross_list_count,
+                link_identifier, is_section_linked,
+                meeting_times, attributes
+            )
+            ON CONFLICT (crn, term_code)
+            DO UPDATE SET
+                subject = EXCLUDED.subject,
+                course_number = EXCLUDED.course_number,
+                title = EXCLUDED.title,
+                enrollment = EXCLUDED.enrollment,
+                max_enrollment = EXCLUDED.max_enrollment,
+                wait_count = EXCLUDED.wait_count,
+                wait_capacity = EXCLUDED.wait_capacity,
+                last_scraped_at = EXCLUDED.last_scraped_at,
+                sequence_number = EXCLUDED.sequence_number,
+                part_of_term = EXCLUDED.part_of_term,
+                instructional_method = EXCLUDED.instructional_method,
+                campus = EXCLUDED.campus,
+                credit_hours = EXCLUDED.credit_hours,
+                credit_hour_low = EXCLUDED.credit_hour_low,
+                credit_hour_high = EXCLUDED.credit_hour_high,
+                cross_list = EXCLUDED.cross_list,
+                cross_list_capacity = EXCLUDED.cross_list_capacity,
+                cross_list_count = EXCLUDED.cross_list_count,
+                link_identifier = EXCLUDED.link_identifier,
+                is_section_linked = EXCLUDED.is_section_linked,
+                meeting_times = EXCLUDED.meeting_times,
+                attributes = EXCLUDED.attributes
+            RETURNING *
        )
-        SELECT * FROM UNNEST(
-            $1::text[], $2::text[], $3::text[], $4::text[], $5::text[],
-            $6::int4[], $7::int4[], $8::int4[], $9::int4[],
-            array_fill(NOW()::timestamptz, ARRAY[$10])
-        ) AS t(
-            crn, subject, course_number, title, term_code,
-            enrollment, max_enrollment, wait_count, wait_capacity, last_scraped_at
-        )
-        ON CONFLICT (crn, term_code)
-        DO UPDATE SET
-            subject = EXCLUDED.subject,
-            course_number = EXCLUDED.course_number,
-            title = EXCLUDED.title,
-            enrollment = EXCLUDED.enrollment,
-            max_enrollment = EXCLUDED.max_enrollment,
-            wait_count = EXCLUDED.wait_count,
-            wait_capacity = EXCLUDED.wait_capacity,
-            last_scraped_at = EXCLUDED.last_scraped_at
+        SELECT u.id,
+               o.id AS old_id,
+               u.crn, u.term_code,
+               o.enrollment AS old_enrollment, u.enrollment AS new_enrollment,
+               o.max_enrollment AS old_max_enrollment, u.max_enrollment AS new_max_enrollment,
+               o.wait_count AS old_wait_count, u.wait_count AS new_wait_count,
+               o.wait_capacity AS old_wait_capacity, u.wait_capacity AS new_wait_capacity,
+               o.subject AS old_subject, u.subject AS new_subject,
+               o.course_number AS old_course_number, u.course_number AS new_course_number,
+               o.title AS old_title, u.title AS new_title,
+               o.sequence_number AS old_sequence_number, u.sequence_number AS new_sequence_number,
+               o.part_of_term AS old_part_of_term, u.part_of_term AS new_part_of_term,
+               o.instructional_method AS old_instructional_method, u.instructional_method AS new_instructional_method,
+               o.campus AS old_campus, u.campus AS new_campus,
+               o.credit_hours AS old_credit_hours, u.credit_hours AS new_credit_hours,
+               o.credit_hour_low AS old_credit_hour_low, u.credit_hour_low AS new_credit_hour_low,
+               o.credit_hour_high AS old_credit_hour_high, u.credit_hour_high AS new_credit_hour_high,
+               o.cross_list AS old_cross_list, u.cross_list AS new_cross_list,
+               o.cross_list_capacity AS old_cross_list_capacity, u.cross_list_capacity AS new_cross_list_capacity,
+               o.cross_list_count AS old_cross_list_count, u.cross_list_count AS new_cross_list_count,
+               o.link_identifier AS old_link_identifier, u.link_identifier AS new_link_identifier,
+               o.is_section_linked AS old_is_section_linked, u.is_section_linked AS new_is_section_linked,
+               o.meeting_times AS old_meeting_times, u.meeting_times AS new_meeting_times,
+               o.attributes AS old_attributes, u.attributes AS new_attributes
+        FROM upserted u
+        LEFT JOIN old_data o ON u.crn = o.crn AND u.term_code = o.term_code
        "#,
    )
    .bind(&crns)
@@ -104,19 +599,150 @@ pub async fn batch_upsert_courses(courses: &[Course], db_pool: &PgPool) -> Resul
    .bind(&max_enrollments)
    .bind(&wait_counts)
    .bind(&wait_capacities)
-    .bind(course_count as i32)
-    .execute(db_pool)
+    .bind(&sequence_numbers)
+    .bind(&parts_of_term)
+    .bind(&instructional_methods)
+    .bind(&campuses)
+    .bind(&credit_hours)
+    .bind(&credit_hour_lows)
+    .bind(&credit_hour_highs)
+    .bind(&cross_lists)
+    .bind(&cross_list_capacities)
+    .bind(&cross_list_counts)
+    .bind(&link_identifiers)
+    .bind(&is_section_linkeds)
+    .bind(&meeting_times_json)
+    .bind(&attributes_json)
+    .fetch_all(&mut *conn)
    .await
    .map_err(|e| anyhow::anyhow!("Failed to batch upsert courses: {}", e))?;

-    let duration = start.elapsed();
+    Ok(rows)
+}

-    info!(
-        courses_count = course_count,
-        rows_affected = result.rows_affected(),
-        duration_ms = duration.as_millis(),
-        "Batch upserted courses"
-    );
+/// Deduplicate and upsert all instructors from the batch by email.
+/// Returns a map of lowercased_email -> instructor id for junction linking.
+async fn upsert_instructors(
+    courses: &[Course],
+    conn: &mut PgConnection,
+) -> Result<HashMap<String, i32>> {
+    let mut seen = HashSet::new();
+    let mut display_names: Vec<&str> = Vec::new();
+    let mut emails_lower: Vec<String> = Vec::new();
+    let mut skipped_no_email = 0u32;
+
+    for course in courses {
+        for faculty in &course.faculty {
+            if let Some(email) = &faculty.email_address {
+                let email_lower = email.to_lowercase();
+                if seen.insert(email_lower.clone()) {
+                    display_names.push(faculty.display_name.as_str());
+                    emails_lower.push(email_lower);
+                }
+            } else {
+                skipped_no_email += 1;
+            }
+        }
+    }
+
+    if skipped_no_email > 0 {
+        tracing::warn!(
+            count = skipped_no_email,
+            "Skipped instructors with no email address"
+        );
+    }
+
+    if display_names.is_empty() {
+        return Ok(HashMap::new());
+    }
+
+    let email_refs: Vec<&str> = emails_lower.iter().map(|s| s.as_str()).collect();
+
+    let rows: Vec<(i32, String)> = sqlx::query_as(
+        r#"
+        INSERT INTO instructors (display_name, email)
+        SELECT * FROM UNNEST($1::text[], $2::text[])
+        ON CONFLICT (email)
+        DO UPDATE SET display_name = EXCLUDED.display_name
+        RETURNING id, email
+        "#,
+    )
+    .bind(&display_names)
+    .bind(&email_refs)
+    .fetch_all(&mut *conn)
+    .await
+    .map_err(|e| anyhow::anyhow!("Failed to batch upsert instructors: {}", e))?;
+
+    Ok(rows.into_iter().map(|(id, email)| (email, id)).collect())
+}
+
+/// Link courses to their instructors via the junction table.
+async fn upsert_course_instructors(
+    courses: &[Course],
+    crn_term_to_id: &HashMap<(&str, &str), i32>,
+    email_to_id: &HashMap<String, i32>,
+    conn: &mut PgConnection,
+) -> Result<()> {
+    let mut cids = Vec::new();
+    let mut instructor_ids: Vec<i32> = Vec::new();
+    let mut banner_ids: Vec<&str> = Vec::new();
+    let mut primaries = Vec::new();
+
+    for course in courses {
+        let key = (
+            course.course_reference_number.as_str(),
+            course.term.as_str(),
+        );
+        let Some(&course_id) = crn_term_to_id.get(&key) else {
+            tracing::warn!(
+                crn = %course.course_reference_number,
+                term = %course.term,
+                "No course_id found for CRN/term pair during instructor linking"
+            );
+            continue;
+        };
+
+        for faculty in &course.faculty {
+            if let Some(email) = &faculty.email_address {
+                let email_lower = email.to_lowercase();
+                if let Some(&instructor_id) = email_to_id.get(&email_lower) {
+                    cids.push(course_id);
+                    instructor_ids.push(instructor_id);
+                    banner_ids.push(faculty.banner_id.as_str());
+                    primaries.push(faculty.primary_indicator);
+                }
+            }
+        }
+    }
+
+    if cids.is_empty() {
+        return Ok(());
+    }
+
+    // Delete existing links for these courses then re-insert.
+    // This handles instructor changes cleanly.
+    sqlx::query("DELETE FROM course_instructors WHERE course_id = ANY($1)")
+        .bind(&cids)
+        .execute(&mut *conn)
+        .await?;
+
+    sqlx::query(
+        r#"
+        INSERT INTO course_instructors (course_id, instructor_id, banner_id, is_primary)
+        SELECT * FROM UNNEST($1::int4[], $2::int4[], $3::text[], $4::bool[])
+        ON CONFLICT (course_id, instructor_id)
+        DO UPDATE SET
+            banner_id = EXCLUDED.banner_id,
+            is_primary = EXCLUDED.is_primary
+        "#,
+    )
+    .bind(&cids)
+    .bind(&instructor_ids)
+    .bind(&banner_ids)
+    .bind(&primaries)
+    .execute(&mut *conn)
+    .await
+    .map_err(|e| anyhow::anyhow!("Failed to batch upsert course_instructors: {}", e))?;

    Ok(())
 }
@@ -0,0 +1,249 @@
+//! Database query functions for courses, used by the web API.
+
+use crate::data::models::{Course, CourseInstructorDetail};
+use crate::error::Result;
+use sqlx::PgPool;
+use std::collections::HashMap;
+
+/// Column to sort search results by.
+#[derive(Debug, Clone, Copy, serde::Deserialize)]
+#[serde(rename_all = "snake_case")]
+pub enum SortColumn {
+    CourseCode,
+    Title,
+    Instructor,
+    Time,
+    Seats,
+}
+
+/// Sort direction.
+#[derive(Debug, Clone, Copy, serde::Deserialize)]
+#[serde(rename_all = "snake_case")]
+pub enum SortDirection {
+    Asc,
+    Desc,
+}
+
+/// Shared WHERE clause for course search filters.
+///
+/// Parameters $1-$8 match the bind order in `search_courses`.
+const SEARCH_WHERE: &str = r#"
+    WHERE term_code = $1
+      AND ($2::text[] IS NULL OR subject = ANY($2))
+      AND ($3::text IS NULL OR title_search @@ plainto_tsquery('simple', $3) OR title ILIKE '%' || $3 || '%')
+      AND ($4::int IS NULL OR course_number::int >= $4)
+      AND ($5::int IS NULL OR course_number::int <= $5)
+      AND ($6::bool = false OR max_enrollment > enrollment)
+      AND ($7::text IS NULL OR instructional_method = $7)
+      AND ($8::text IS NULL OR campus = $8)
+"#;
+
+/// Build a safe ORDER BY clause from typed sort parameters.
+///
+/// All column names are hardcoded string literals — no caller input is interpolated.
+fn sort_clause(column: Option<SortColumn>, direction: Option<SortDirection>) -> String {
+    let dir = match direction.unwrap_or(SortDirection::Asc) {
+        SortDirection::Asc => "ASC",
+        SortDirection::Desc => "DESC",
+    };
+
+    match column {
+        Some(SortColumn::CourseCode) => {
+            format!("subject {dir}, course_number {dir}, sequence_number {dir}")
+        }
+        Some(SortColumn::Title) => format!("title {dir}"),
+        Some(SortColumn::Instructor) => {
+            format!(
+                "(SELECT i.display_name FROM course_instructors ci \
+                 JOIN instructors i ON i.id = ci.instructor_id \
+                 WHERE ci.course_id = courses.id AND ci.is_primary = true \
+                 LIMIT 1) {dir} NULLS LAST"
+            )
+        }
+        Some(SortColumn::Time) => {
+            format!("(meeting_times->0->>'begin_time') {dir} NULLS LAST")
+        }
+        Some(SortColumn::Seats) => {
+            format!("(max_enrollment - enrollment) {dir}")
+        }
+        None => "subject ASC, course_number ASC, sequence_number ASC".to_string(),
+    }
+}
+
+/// Search courses by term with optional filters.
+///
+/// Returns `(courses, total_count)` for pagination. Uses FTS tsvector for word
+/// search and falls back to trigram ILIKE for substring matching.
+#[allow(clippy::too_many_arguments)]
+pub async fn search_courses(
+    db_pool: &PgPool,
+    term_code: &str,
+    subject: Option<&[String]>,
+    title_query: Option<&str>,
+    course_number_low: Option<i32>,
+    course_number_high: Option<i32>,
+    open_only: bool,
+    instructional_method: Option<&str>,
+    campus: Option<&str>,
+    limit: i32,
+    offset: i32,
+    sort_by: Option<SortColumn>,
+    sort_dir: Option<SortDirection>,
+) -> Result<(Vec<Course>, i64)> {
+    let order_by = sort_clause(sort_by, sort_dir);
+
+    let data_query =
+        format!("SELECT * FROM courses {SEARCH_WHERE} ORDER BY {order_by} LIMIT $9 OFFSET $10");
+    let count_query = format!("SELECT COUNT(*) FROM courses {SEARCH_WHERE}");
+
+    let courses = sqlx::query_as::<_, Course>(&data_query)
+        .bind(term_code)
+        .bind(subject)
+        .bind(title_query)
+        .bind(course_number_low)
+        .bind(course_number_high)
+        .bind(open_only)
+        .bind(instructional_method)
+        .bind(campus)
+        .bind(limit)
+        .bind(offset)
+        .fetch_all(db_pool)
+        .await?;
+
+    let total: (i64,) = sqlx::query_as(&count_query)
+        .bind(term_code)
+        .bind(subject)
+        .bind(title_query)
+        .bind(course_number_low)
+        .bind(course_number_high)
+        .bind(open_only)
+        .bind(instructional_method)
+        .bind(campus)
+        .fetch_one(db_pool)
+        .await?;
+
+    Ok((courses, total.0))
+}
+
+/// Get a single course by CRN and term.
+pub async fn get_course_by_crn(
+    db_pool: &PgPool,
+    crn: &str,
+    term_code: &str,
+) -> Result<Option<Course>> {
+    let course =
+        sqlx::query_as::<_, Course>("SELECT * FROM courses WHERE crn = $1 AND term_code = $2")
+            .bind(crn)
+            .bind(term_code)
+            .fetch_optional(db_pool)
+            .await?;
+    Ok(course)
+}
+
+/// Get instructors for a single course by course ID.
+pub async fn get_course_instructors(
+    db_pool: &PgPool,
+    course_id: i32,
+) -> Result<Vec<CourseInstructorDetail>> {
+    let rows = sqlx::query_as::<_, CourseInstructorDetail>(
+        r#"
+        SELECT i.id as instructor_id, ci.banner_id, i.display_name, i.email, ci.is_primary,
+               rmp.avg_rating, rmp.num_ratings, rmp.rmp_legacy_id,
+               ci.course_id
+        FROM course_instructors ci
+        JOIN instructors i ON i.id = ci.instructor_id
+        LEFT JOIN LATERAL (
+            SELECT rp.avg_rating, rp.num_ratings, rp.legacy_id as rmp_legacy_id
+            FROM instructor_rmp_links irl
+            JOIN rmp_professors rp ON rp.legacy_id = irl.rmp_legacy_id
+            WHERE irl.instructor_id = i.id
+            ORDER BY rp.num_ratings DESC NULLS LAST, rp.legacy_id ASC
+            LIMIT 1
+        ) rmp ON true
+        WHERE ci.course_id = $1
+        ORDER BY ci.is_primary DESC, i.display_name
+        "#,
+    )
+    .bind(course_id)
+    .fetch_all(db_pool)
+    .await?;
+    Ok(rows)
+}
+
+/// Batch-fetch instructors for multiple courses in a single query.
+///
+/// Returns a map of `course_id → Vec<CourseInstructorDetail>`.
+pub async fn get_instructors_for_courses(
+    db_pool: &PgPool,
+    course_ids: &[i32],
+) -> Result<HashMap<i32, Vec<CourseInstructorDetail>>> {
+    if course_ids.is_empty() {
+        return Ok(HashMap::new());
+    }
+
+    let rows = sqlx::query_as::<_, CourseInstructorDetail>(
+        r#"
+        SELECT i.id as instructor_id, ci.banner_id, i.display_name, i.email, ci.is_primary,
+               rmp.avg_rating, rmp.num_ratings, rmp.rmp_legacy_id,
+               ci.course_id
+        FROM course_instructors ci
+        JOIN instructors i ON i.id = ci.instructor_id
+        LEFT JOIN LATERAL (
+            SELECT rp.avg_rating, rp.num_ratings, rp.legacy_id as rmp_legacy_id
+            FROM instructor_rmp_links irl
+            JOIN rmp_professors rp ON rp.legacy_id = irl.rmp_legacy_id
+            WHERE irl.instructor_id = i.id
+            ORDER BY rp.num_ratings DESC NULLS LAST, rp.legacy_id ASC
+            LIMIT 1
+        ) rmp ON true
+        WHERE ci.course_id = ANY($1)
+        ORDER BY ci.course_id, ci.is_primary DESC, i.display_name
+        "#,
+    )
+    .bind(course_ids)
+    .fetch_all(db_pool)
+    .await?;
+
+    let mut map: HashMap<i32, Vec<CourseInstructorDetail>> = HashMap::new();
+    for row in rows {
+        // course_id is always present in the batch query
+        let cid = row.course_id.unwrap_or_default();
+        map.entry(cid).or_default().push(row);
+    }
+    Ok(map)
+}
+
+/// Get subjects for a term, sorted by total enrollment (descending).
+///
+/// Returns only subjects that have courses in the given term, with their
+/// descriptions from reference_data and enrollment totals for ranking.
+pub async fn get_subjects_by_enrollment(
+    db_pool: &PgPool,
+    term_code: &str,
+) -> Result<Vec<(String, String, i64)>> {
+    let rows: Vec<(String, String, i64)> = sqlx::query_as(
+        r#"
+        SELECT c.subject,
+               COALESCE(rd.description, c.subject),
+               COALESCE(SUM(c.enrollment), 0) as total_enrollment
+        FROM courses c
+        LEFT JOIN reference_data rd ON rd.category = 'subject' AND rd.code = c.subject
+        WHERE c.term_code = $1
+        GROUP BY c.subject, rd.description
+        ORDER BY total_enrollment DESC
+        "#,
+    )
+    .bind(term_code)
+    .fetch_all(db_pool)
+    .await?;
+    Ok(rows)
+}
+
+/// Get all distinct term codes that have courses in the DB.
+pub async fn get_available_terms(db_pool: &PgPool) -> Result<Vec<String>> {
+    let rows: Vec<(String,)> =
+        sqlx::query_as("SELECT DISTINCT term_code FROM courses ORDER BY term_code DESC")
+            .fetch_all(db_pool)
+            .await?;
+    Ok(rows.into_iter().map(|(tc,)| tc).collect())
+}
@@ -1,5 +1,11 @@
 //! Database models and schema.

 pub mod batch;
+pub mod courses;
 pub mod models;
+pub mod reference;
+pub mod rmp;
+pub mod rmp_matching;
 pub mod scrape_jobs;
+pub mod sessions;
+pub mod users;
@@ -1,7 +1,68 @@
 //! `sqlx` models for the database schema.

 use chrono::{DateTime, Utc};
+use serde::{Deserialize, Deserializer, Serialize, Serializer};
 use serde_json::Value;
+use ts_rs::TS;
+
+/// Serialize an `i64` as a string to avoid JavaScript precision loss for values exceeding 2^53.
+fn serialize_i64_as_string<S: Serializer>(value: &i64, serializer: S) -> Result<S::Ok, S::Error> {
+    serializer.serialize_str(&value.to_string())
+}
+
+/// Deserialize an `i64` from either a number or a string.
+fn deserialize_i64_from_string<'de, D: Deserializer<'de>>(
+    deserializer: D,
+) -> Result<i64, D::Error> {
+    use serde::de;
+
+    struct I64OrStringVisitor;
+
+    impl<'de> de::Visitor<'de> for I64OrStringVisitor {
+        type Value = i64;
+
+        fn expecting(&self, formatter: &mut std::fmt::Formatter) -> std::fmt::Result {
+            formatter.write_str("an integer or a string containing an integer")
+        }
+
+        fn visit_i64<E: de::Error>(self, value: i64) -> Result<i64, E> {
+            Ok(value)
+        }
+
+        fn visit_u64<E: de::Error>(self, value: u64) -> Result<i64, E> {
+            i64::try_from(value).map_err(|_| E::custom(format!("u64 {value} out of i64 range")))
+        }
+
+        fn visit_str<E: de::Error>(self, value: &str) -> Result<i64, E> {
+            value.parse().map_err(de::Error::custom)
+        }
+    }
+
+    deserializer.deserialize_any(I64OrStringVisitor)
+}
+
+/// Represents a meeting time stored as JSONB in the courses table.
+#[derive(Debug, Clone, Serialize, Deserialize, TS)]
+#[ts(export)]
+pub struct DbMeetingTime {
+    pub begin_time: Option<String>,
+    pub end_time: Option<String>,
+    pub start_date: String,
+    pub end_date: String,
+    pub monday: bool,
+    pub tuesday: bool,
+    pub wednesday: bool,
+    pub thursday: bool,
+    pub friday: bool,
+    pub saturday: bool,
+    pub sunday: bool,
+    pub building: Option<String>,
+    pub building_description: Option<String>,
+    pub room: Option<String>,
+    pub campus: Option<String>,
+    pub meeting_type: String,
+    pub meeting_schedule_type: String,
+}

 #[allow(dead_code)]
 #[derive(sqlx::FromRow, Debug, Clone)]
@@ -17,6 +78,63 @@ pub struct Course {
    pub wait_count: i32,
    pub wait_capacity: i32,
    pub last_scraped_at: DateTime<Utc>,
+    // New scalar fields
+    pub sequence_number: Option<String>,
+    pub part_of_term: Option<String>,
+    pub instructional_method: Option<String>,
+    pub campus: Option<String>,
+    pub credit_hours: Option<i32>,
+    pub credit_hour_low: Option<i32>,
+    pub credit_hour_high: Option<i32>,
+    pub cross_list: Option<String>,
+    pub cross_list_capacity: Option<i32>,
+    pub cross_list_count: Option<i32>,
+    pub link_identifier: Option<String>,
+    pub is_section_linked: Option<bool>,
+    // JSONB fields
+    pub meeting_times: Value,
+    pub attributes: Value,
+}
+
+#[allow(dead_code)]
+#[derive(sqlx::FromRow, Debug, Clone)]
+pub struct Instructor {
+    pub id: i32,
+    pub display_name: String,
+    pub email: String,
+    pub rmp_match_status: String,
+}
+
+#[allow(dead_code)]
+#[derive(sqlx::FromRow, Debug, Clone)]
+pub struct CourseInstructor {
+    pub course_id: i32,
+    pub instructor_id: i32,
+    pub banner_id: String,
+    pub is_primary: bool,
+}
+
+/// Joined instructor data for a course (from course_instructors + instructors + rmp_professors).
+#[derive(sqlx::FromRow, Debug, Clone)]
+pub struct CourseInstructorDetail {
+    pub instructor_id: i32,
+    pub banner_id: String,
+    pub display_name: String,
+    pub email: String,
+    pub is_primary: bool,
+    pub avg_rating: Option<f32>,
+    pub num_ratings: Option<i32>,
+    pub rmp_legacy_id: Option<i32>,
+    /// Present when fetched via batch query; `None` for single-course queries.
+    pub course_id: Option<i32>,
+}
+
+#[allow(dead_code)]
+#[derive(sqlx::FromRow, Debug, Clone)]
+pub struct ReferenceData {
+    pub category: String,
+    pub code: String,
+    pub description: String,
 }

 #[allow(dead_code)]
@@ -41,6 +159,16 @@ pub struct CourseAudit {
    pub new_value: String,
 }

+/// Aggregate counts returned by batch upsert, used for scrape job result logging.
+#[derive(Debug, Clone, Default)]
+pub struct UpsertCounts {
+    pub courses_fetched: i32,
+    pub courses_changed: i32,
+    pub courses_unchanged: i32,
+    pub audits_generated: i32,
+    pub metrics_generated: i32,
+}
+
 /// The priority level of a scrape job.
 #[derive(sqlx::Type, Copy, Debug, Clone)]
 #[sqlx(type_name = "scrape_priority", rename_all = "PascalCase")]
@@ -61,6 +189,20 @@ pub enum TargetType {
    SingleCrn,
 }

+/// Computed status for a scrape job, derived from existing fields.
+#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize)]
+#[serde(rename_all = "camelCase")]
+pub enum ScrapeJobStatus {
+    Processing,
+    StaleLock,
+    Exhausted,
+    Scheduled,
+    Pending,
+}
+
+/// How long a lock can be held before it is considered stale (mirrors `scrape_jobs::LOCK_EXPIRY`).
+const LOCK_EXPIRY_SECS: i64 = 10 * 60;
+
 /// Represents a queryable job from the database.
 #[allow(dead_code)]
 #[derive(sqlx::FromRow, Debug, Clone)]
@@ -76,4 +218,54 @@ pub struct ScrapeJob {
    pub retry_count: i32,
    /// Maximum number of retry attempts allowed (non-negative, enforced by CHECK constraint)
    pub max_retries: i32,
+    /// When the job last entered the "ready to pick up" state.
+    /// Set to NOW() on creation; updated to NOW() on retry.
+    pub queued_at: DateTime<Utc>,
+}
+
+impl ScrapeJob {
+    /// Compute the current status of this job from its fields.
+    pub fn status(&self) -> ScrapeJobStatus {
+        let now = Utc::now();
+        match self.locked_at {
+            Some(locked) if (now - locked).num_seconds() < LOCK_EXPIRY_SECS => {
+                ScrapeJobStatus::Processing
+            }
+            Some(_) => ScrapeJobStatus::StaleLock,
+            None if self.retry_count >= self.max_retries && self.max_retries > 0 => {
+                ScrapeJobStatus::Exhausted
+            }
+            None if self.execute_at > now => ScrapeJobStatus::Scheduled,
+            None => ScrapeJobStatus::Pending,
+        }
+    }
+}
+
+/// A user authenticated via Discord OAuth.
+#[derive(sqlx::FromRow, Debug, Clone, Serialize, Deserialize, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export)]
+pub struct User {
+    #[serde(
+        serialize_with = "serialize_i64_as_string",
+        deserialize_with = "deserialize_i64_from_string"
+    )]
+    #[ts(type = "string")]
+    pub discord_id: i64,
+    pub discord_username: String,
+    pub discord_avatar_hash: Option<String>,
+    pub is_admin: bool,
+    pub created_at: DateTime<Utc>,
+    pub updated_at: DateTime<Utc>,
+}
+
+/// A server-side session for an authenticated user.
+#[allow(dead_code)] // Fields read via sqlx::FromRow; some only used in DB queries
+#[derive(sqlx::FromRow, Debug, Clone)]
+pub struct UserSession {
+    pub id: String,
+    pub user_id: i64,
+    pub created_at: DateTime<Utc>,
+    pub expires_at: DateTime<Utc>,
+    pub last_active_at: DateTime<Utc>,
 }
@@ -0,0 +1,57 @@
+//! Database operations for the `reference_data` table (code→description lookups).
+
+use crate::data::models::ReferenceData;
+use crate::error::Result;
+use html_escape::decode_html_entities;
+use sqlx::PgPool;
+
+/// Batch upsert reference data entries.
+pub async fn batch_upsert(entries: &[ReferenceData], db_pool: &PgPool) -> Result<()> {
+    if entries.is_empty() {
+        return Ok(());
+    }
+
+    let categories: Vec<&str> = entries.iter().map(|e| e.category.as_str()).collect();
+    let codes: Vec<&str> = entries.iter().map(|e| e.code.as_str()).collect();
+    let descriptions: Vec<String> = entries
+        .iter()
+        .map(|e| decode_html_entities(&e.description).into_owned())
+        .collect();
+
+    sqlx::query(
+        r#"
+        INSERT INTO reference_data (category, code, description)
+        SELECT * FROM UNNEST($1::text[], $2::text[], $3::text[])
+        ON CONFLICT (category, code)
+        DO UPDATE SET description = EXCLUDED.description
+        "#,
+    )
+    .bind(&categories)
+    .bind(&codes)
+    .bind(&descriptions)
+    .execute(db_pool)
+    .await?;
+
+    Ok(())
+}
+
+/// Get all reference data entries for a category.
+pub async fn get_by_category(category: &str, db_pool: &PgPool) -> Result<Vec<ReferenceData>> {
+    let rows = sqlx::query_as::<_, ReferenceData>(
+        "SELECT category, code, description FROM reference_data WHERE category = $1 ORDER BY description",
+    )
+    .bind(category)
+    .fetch_all(db_pool)
+    .await?;
+    Ok(rows)
+}
+
+/// Get all reference data entries (for cache initialization).
+pub async fn get_all(db_pool: &PgPool) -> Result<Vec<ReferenceData>> {
+    let rows = sqlx::query_as::<_, ReferenceData>(
+        "SELECT category, code, description FROM reference_data ORDER BY category, description",
+    )
+    .fetch_all(db_pool)
+    .await?;
+    Ok(rows)
+}
@@ -0,0 +1,209 @@
+//! Database operations for RateMyProfessors data.
+
+use crate::error::Result;
+use crate::rmp::RmpProfessor;
+use sqlx::PgPool;
+use std::collections::HashSet;
+
+/// Bulk upsert RMP professors using the UNNEST pattern.
+///
+/// Deduplicates by `legacy_id` before inserting — the RMP API can return
+/// the same professor on multiple pages.
+pub async fn batch_upsert_rmp_professors(
+    professors: &[RmpProfessor],
+    db_pool: &PgPool,
+) -> Result<()> {
+    if professors.is_empty() {
+        return Ok(());
+    }
+
+    // Deduplicate: keep last occurrence per legacy_id (latest page wins)
+    let mut seen = HashSet::new();
+    let deduped: Vec<&RmpProfessor> = professors
+        .iter()
+        .rev()
+        .filter(|p| seen.insert(p.legacy_id))
+        .collect();
+
+    let legacy_ids: Vec<i32> = deduped.iter().map(|p| p.legacy_id).collect();
+    let graphql_ids: Vec<&str> = deduped.iter().map(|p| p.graphql_id.as_str()).collect();
+    let first_names: Vec<String> = deduped
+        .iter()
+        .map(|p| p.first_name.trim().to_string())
+        .collect();
+    let first_name_refs: Vec<&str> = first_names.iter().map(|s| s.as_str()).collect();
+    let last_names: Vec<String> = deduped
+        .iter()
+        .map(|p| p.last_name.trim().to_string())
+        .collect();
+    let last_name_refs: Vec<&str> = last_names.iter().map(|s| s.as_str()).collect();
+    let departments: Vec<Option<&str>> = deduped.iter().map(|p| p.department.as_deref()).collect();
+    let avg_ratings: Vec<Option<f32>> = deduped.iter().map(|p| p.avg_rating).collect();
+    let avg_difficulties: Vec<Option<f32>> = deduped.iter().map(|p| p.avg_difficulty).collect();
+    let num_ratings: Vec<i32> = deduped.iter().map(|p| p.num_ratings).collect();
+    let would_take_again_pcts: Vec<Option<f32>> =
+        deduped.iter().map(|p| p.would_take_again_pct).collect();
+
+    sqlx::query(
+        r#"
+        INSERT INTO rmp_professors (
+            legacy_id, graphql_id, first_name, last_name, department,
+            avg_rating, avg_difficulty, num_ratings, would_take_again_pct,
+            last_synced_at
+        )
+        SELECT
+            v.legacy_id, v.graphql_id, v.first_name, v.last_name, v.department,
+            v.avg_rating, v.avg_difficulty, v.num_ratings, v.would_take_again_pct,
+            NOW()
+        FROM UNNEST(
+            $1::int4[], $2::text[], $3::text[], $4::text[], $5::text[],
+            $6::real[], $7::real[], $8::int4[], $9::real[]
+        ) AS v(
+            legacy_id, graphql_id, first_name, last_name, department,
+            avg_rating, avg_difficulty, num_ratings, would_take_again_pct
+        )
+        ON CONFLICT (legacy_id)
+        DO UPDATE SET
+            graphql_id = EXCLUDED.graphql_id,
+            first_name = EXCLUDED.first_name,
+            last_name = EXCLUDED.last_name,
+            department = EXCLUDED.department,
+            avg_rating = EXCLUDED.avg_rating,
+            avg_difficulty = EXCLUDED.avg_difficulty,
+            num_ratings = EXCLUDED.num_ratings,
+            would_take_again_pct = EXCLUDED.would_take_again_pct,
+            last_synced_at = EXCLUDED.last_synced_at
+        "#,
+    )
+    .bind(&legacy_ids)
+    .bind(&graphql_ids)
+    .bind(&first_name_refs)
+    .bind(&last_name_refs)
+    .bind(&departments)
+    .bind(&avg_ratings)
+    .bind(&avg_difficulties)
+    .bind(&num_ratings)
+    .bind(&would_take_again_pcts)
+    .execute(db_pool)
+    .await
+    .map_err(|e| anyhow::anyhow!("Failed to batch upsert RMP professors: {}", e))?;
+
+    Ok(())
+}
+
+/// Normalize a name for matching: lowercase, trim, strip trailing periods.
+pub(crate) fn normalize(s: &str) -> String {
+    s.trim().to_lowercase().trim_end_matches('.').to_string()
+}
+
+/// Parse Banner's "Last, First Middle" display name into (last, first) tokens.
+///
+/// Returns `None` if the format is unparseable (no comma, empty parts).
+pub(crate) fn parse_display_name(display_name: &str) -> Option<(String, String)> {
+    let (last_part, first_part) = display_name.split_once(',')?;
+    let last = normalize(last_part);
+    // Take only the first token of the first-name portion to drop middle names/initials.
+    let first = normalize(first_part.split_whitespace().next()?);
+    if last.is_empty() || first.is_empty() {
+        return None;
+    }
+    Some((last, first))
+}
+
+/// Retrieve RMP rating data for an instructor by instructor id.
+///
+/// Returns `(avg_rating, num_ratings)` for the best linked RMP profile
+/// (most ratings). Returns `None` if no link exists.
+#[allow(dead_code)]
+pub async fn get_instructor_rmp_data(
+    db_pool: &PgPool,
+    instructor_id: i32,
+) -> Result<Option<(f32, i32)>> {
+    let row: Option<(f32, i32)> = sqlx::query_as(
+        r#"
+        SELECT rp.avg_rating, rp.num_ratings
+        FROM instructor_rmp_links irl
+        JOIN rmp_professors rp ON rp.legacy_id = irl.rmp_legacy_id
+        WHERE irl.instructor_id = $1
+          AND rp.avg_rating IS NOT NULL
+        ORDER BY rp.num_ratings DESC NULLS LAST
+        LIMIT 1
+        "#,
+    )
+    .bind(instructor_id)
+    .fetch_optional(db_pool)
+    .await?;
+    Ok(row)
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn parse_standard_name() {
+        assert_eq!(
+            parse_display_name("Smith, John"),
+            Some(("smith".into(), "john".into()))
+        );
+    }
+
+    #[test]
+    fn parse_name_with_middle() {
+        assert_eq!(
+            parse_display_name("Smith, John David"),
+            Some(("smith".into(), "john".into()))
+        );
+    }
+
+    #[test]
+    fn parse_name_with_middle_initial() {
+        assert_eq!(
+            parse_display_name("Garcia, Maria L."),
+            Some(("garcia".into(), "maria".into()))
+        );
+    }
+
+    #[test]
+    fn parse_name_with_suffix_in_last() {
+        // Banner may encode "Jr." as part of the last name.
+        // normalize() strips trailing periods so "Jr." becomes "jr".
+        assert_eq!(
+            parse_display_name("Smith Jr., James"),
+            Some(("smith jr".into(), "james".into()))
+        );
+    }
+
+    #[test]
+    fn parse_no_comma_returns_none() {
+        assert_eq!(parse_display_name("SingleName"), None);
+    }
+
+    #[test]
+    fn parse_empty_first_returns_none() {
+        assert_eq!(parse_display_name("Smith,"), None);
+    }
+
+    #[test]
+    fn parse_empty_last_returns_none() {
+        assert_eq!(parse_display_name(", John"), None);
+    }
+
+    #[test]
+    fn parse_extra_whitespace() {
+        assert_eq!(
+            parse_display_name("  Doe ,  Jane   Marie  "),
+            Some(("doe".into(), "jane".into()))
+        );
+    }
+
+    #[test]
+    fn normalize_trims_and_lowercases() {
+        assert_eq!(normalize("  FOO  "), "foo");
+    }
+
+    #[test]
+    fn normalize_strips_trailing_period() {
+        assert_eq!(normalize("Jr."), "jr");
+    }
+}
@@ -0,0 +1,513 @@
+//! Confidence scoring and candidate generation for RMP instructor matching.
+
+use crate::data::rmp::{normalize, parse_display_name};
+use crate::error::Result;
+use serde::{Deserialize, Serialize};
+use sqlx::PgPool;
+use std::collections::{HashMap, HashSet};
+use tracing::{debug, info};
+
+// ---------------------------------------------------------------------------
+// Scoring types
+// ---------------------------------------------------------------------------
+
+/// Breakdown of individual scoring signals.
+#[derive(Debug, Clone, Default, Serialize, Deserialize)]
+pub struct ScoreBreakdown {
+    pub department: f32,
+    pub uniqueness: f32,
+    pub volume: f32,
+}
+
+/// Result of scoring a single instructor–RMP candidate pair.
+#[derive(Debug, Clone)]
+pub struct MatchScore {
+    pub score: f32,
+    pub breakdown: ScoreBreakdown,
+}
+
+// ---------------------------------------------------------------------------
+// Thresholds
+// ---------------------------------------------------------------------------
+
+/// Minimum composite score to store a candidate row.
+const MIN_CANDIDATE_THRESHOLD: f32 = 0.40;
+
+/// Score at or above which a candidate is auto-accepted.
+const AUTO_ACCEPT_THRESHOLD: f32 = 0.85;
+
+// ---------------------------------------------------------------------------
+// Weights
+// ---------------------------------------------------------------------------
+
+const WEIGHT_DEPARTMENT: f32 = 0.50;
+const WEIGHT_UNIQUENESS: f32 = 0.30;
+const WEIGHT_VOLUME: f32 = 0.20;
+
+// ---------------------------------------------------------------------------
+// Pure scoring functions
+// ---------------------------------------------------------------------------
+
+/// Check if an instructor's subjects overlap with an RMP department.
+///
+/// Returns `1.0` for a match, `0.2` for a mismatch, `0.5` when the RMP
+/// department is unknown.
+fn department_similarity(subjects: &[String], rmp_department: Option<&str>) -> f32 {
+    let Some(dept) = rmp_department else {
+        return 0.5;
+    };
+    let dept_lower = dept.to_lowercase();
+
+    // Quick check: does any subject appear directly in the department string
+    // or vice-versa?
+    for subj in subjects {
+        let subj_lower = subj.to_lowercase();
+        if dept_lower.contains(&subj_lower) || subj_lower.contains(&dept_lower) {
+            return 1.0;
+        }
+
+        // Handle common UTSA abbreviation mappings.
+        if matches_known_abbreviation(&subj_lower, &dept_lower) {
+            return 1.0;
+        }
+    }
+
+    0.2
+}
+
+/// Expand common subject abbreviations used at UTSA and check for overlap.
+fn matches_known_abbreviation(subject: &str, department: &str) -> bool {
+    const MAPPINGS: &[(&str, &[&str])] = &[
+        ("cs", &["computer science"]),
+        ("ece", &["electrical", "computer engineering"]),
+        ("ee", &["electrical engineering", "electrical"]),
+        ("me", &["mechanical engineering", "mechanical"]),
+        ("ce", &["civil engineering", "civil"]),
+        ("bio", &["biology", "biological"]),
+        ("chem", &["chemistry"]),
+        ("phys", &["physics"]),
+        ("math", &["mathematics"]),
+        ("sta", &["statistics"]),
+        ("eng", &["english"]),
+        ("his", &["history"]),
+        ("pol", &["political science"]),
+        ("psy", &["psychology"]),
+        ("soc", &["sociology"]),
+        ("mus", &["music"]),
+        ("art", &["art"]),
+        ("phi", &["philosophy"]),
+        ("eco", &["economics"]),
+        ("acc", &["accounting"]),
+        ("fin", &["finance"]),
+        ("mgt", &["management"]),
+        ("mkt", &["marketing"]),
+        ("is", &["information systems"]),
+        ("ms", &["management science"]),
+        ("kin", &["kinesiology"]),
+        ("com", &["communication"]),
+    ];
+
+    for &(abbr, expansions) in MAPPINGS {
+        if subject == abbr {
+            return expansions
+                .iter()
+                .any(|expansion| department.contains(expansion));
+        }
+    }
+    false
+}
+
+/// Compute match confidence score (0.0–1.0) for an instructor–RMP pair.
+///
+/// Name matching is handled by the caller via pre-filtering on exact
+/// normalized `(last, first)`, so only department, uniqueness, and volume
+/// signals are scored here.
+pub fn compute_match_score(
+    instructor_subjects: &[String],
+    rmp_department: Option<&str>,
+    candidate_count: usize,
+    rmp_num_ratings: i32,
+) -> MatchScore {
+    // --- Department (0.50) ---
+    let dept_score = department_similarity(instructor_subjects, rmp_department);
+
+    // --- Uniqueness (0.30) ---
+    let uniqueness_score = match candidate_count {
+        0 | 1 => 1.0,
+        2 => 0.5,
+        _ => 0.2,
+    };
+
+    // --- Volume (0.20) ---
+    let volume_score = ((rmp_num_ratings as f32).ln_1p() / 5.0_f32.ln_1p()).clamp(0.0, 1.0);
+
+    let composite = dept_score * WEIGHT_DEPARTMENT
+        + uniqueness_score * WEIGHT_UNIQUENESS
+        + volume_score * WEIGHT_VOLUME;
+
+    MatchScore {
+        score: composite,
+        breakdown: ScoreBreakdown {
+            department: dept_score,
+            uniqueness: uniqueness_score,
+            volume: volume_score,
+        },
+    }
+}
+
+// ---------------------------------------------------------------------------
+// Candidate generation (DB)
+// ---------------------------------------------------------------------------
+
+/// Statistics returned from candidate generation.
+#[derive(Debug)]
+pub struct MatchingStats {
+    pub total_unmatched: usize,
+    pub candidates_created: usize,
+    pub auto_matched: usize,
+    pub skipped_unparseable: usize,
+    pub skipped_no_candidates: usize,
+}
+
+/// Lightweight row for building the in-memory RMP name index.
+struct RmpProfForMatching {
+    legacy_id: i32,
+    department: Option<String>,
+    num_ratings: i32,
+}
+
+/// Generate match candidates for all unmatched instructors.
+///
+/// For each unmatched instructor:
+/// 1. Parse `display_name` into (last, first).
+/// 2. Find RMP professors with matching normalized name.
+/// 3. Score each candidate.
+/// 4. Store candidates scoring above [`MIN_CANDIDATE_THRESHOLD`].
+/// 5. Auto-accept if the top candidate scores ≥ [`AUTO_ACCEPT_THRESHOLD`]
+///    and no existing rejected candidate exists for that pair.
+///
+/// Already-evaluated instructor–RMP pairs (any status) are skipped.
+pub async fn generate_candidates(db_pool: &PgPool) -> Result<MatchingStats> {
+    // 1. Load unmatched instructors
+    let instructors: Vec<(i32, String)> = sqlx::query_as(
+        "SELECT id, display_name FROM instructors WHERE rmp_match_status = 'unmatched'",
+    )
+    .fetch_all(db_pool)
+    .await?;
+
+    if instructors.is_empty() {
+        info!("No unmatched instructors to generate candidates for");
+        return Ok(MatchingStats {
+            total_unmatched: 0,
+            candidates_created: 0,
+            auto_matched: 0,
+            skipped_unparseable: 0,
+            skipped_no_candidates: 0,
+        });
+    }
+
+    let instructor_ids: Vec<i32> = instructors.iter().map(|(id, _)| *id).collect();
+    let total_unmatched = instructors.len();
+
+    // 2. Load instructor subjects
+    let subject_rows: Vec<(i32, String)> = sqlx::query_as(
+        r#"
+        SELECT DISTINCT ci.instructor_id, c.subject
+        FROM course_instructors ci
+        JOIN courses c ON c.id = ci.course_id
+        WHERE ci.instructor_id = ANY($1)
+        "#,
+    )
+    .bind(&instructor_ids)
+    .fetch_all(db_pool)
+    .await?;
+
+    let mut subject_map: HashMap<i32, Vec<String>> = HashMap::new();
+    for (iid, subject) in subject_rows {
+        subject_map.entry(iid).or_default().push(subject);
+    }
+
+    // 3. Load all RMP professors
+    let prof_rows: Vec<(i32, String, String, Option<String>, i32)> = sqlx::query_as(
+        "SELECT legacy_id, first_name, last_name, department, num_ratings FROM rmp_professors",
+    )
+    .fetch_all(db_pool)
+    .await?;
+
+    // Build name index: (normalized_last, normalized_first) -> Vec<RmpProfForMatching>
+    let mut name_index: HashMap<(String, String), Vec<RmpProfForMatching>> = HashMap::new();
+    for (legacy_id, first_name, last_name, department, num_ratings) in prof_rows {
+        let key = (normalize(&last_name), normalize(&first_name));
+        name_index.entry(key).or_default().push(RmpProfForMatching {
+            legacy_id,
+            department,
+            num_ratings,
+        });
+    }
+
+    // 4. Load existing candidate pairs (and rejected subset) in a single query
+    let candidate_rows: Vec<(i32, i32, String)> =
+        sqlx::query_as("SELECT instructor_id, rmp_legacy_id, status FROM rmp_match_candidates")
+            .fetch_all(db_pool)
+            .await?;
+
+    let mut existing_pairs: HashSet<(i32, i32)> = HashSet::with_capacity(candidate_rows.len());
+    let mut rejected_pairs: HashSet<(i32, i32)> = HashSet::new();
+    for (iid, lid, status) in candidate_rows {
+        existing_pairs.insert((iid, lid));
+        if status == "rejected" {
+            rejected_pairs.insert((iid, lid));
+        }
+    }
+
+    // 5. Score and collect candidates
+    let empty_subjects: Vec<String> = Vec::new();
+    let mut candidates: Vec<(i32, i32, f32, serde_json::Value)> = Vec::new();
+    let mut auto_accept: Vec<(i32, i32)> = Vec::new(); // (instructor_id, legacy_id)
+    let mut skipped_unparseable = 0usize;
+    let mut skipped_no_candidates = 0usize;
+
+    for (instructor_id, display_name) in &instructors {
+        let Some((norm_last, norm_first)) = parse_display_name(display_name) else {
+            skipped_unparseable += 1;
+            debug!(
+                instructor_id,
+                display_name, "Unparseable display name, skipping"
+            );
+            continue;
+        };
+
+        let subjects = subject_map.get(instructor_id).unwrap_or(&empty_subjects);
+
+        let key = (norm_last.clone(), norm_first.clone());
+        let Some(rmp_candidates) = name_index.get(&key) else {
+            skipped_no_candidates += 1;
+            continue;
+        };
+
+        let candidate_count = rmp_candidates.len();
+        let mut best: Option<(f32, i32)> = None;
+
+        for prof in rmp_candidates {
+            let pair = (*instructor_id, prof.legacy_id);
+            if existing_pairs.contains(&pair) {
+                continue;
+            }
+
+            let ms = compute_match_score(
+                subjects,
+                prof.department.as_deref(),
+                candidate_count,
+                prof.num_ratings,
+            );
+
+            if ms.score < MIN_CANDIDATE_THRESHOLD {
+                continue;
+            }
+
+            let breakdown_json =
+                serde_json::to_value(&ms.breakdown).unwrap_or_else(|_| serde_json::json!({}));
+
+            candidates.push((*instructor_id, prof.legacy_id, ms.score, breakdown_json));
+
+            match best {
+                Some((s, _)) if ms.score > s => best = Some((ms.score, prof.legacy_id)),
+                None => best = Some((ms.score, prof.legacy_id)),
+                _ => {}
+            }
+        }
+
+        // Auto-accept the top candidate if it meets the threshold and is not
+        // previously rejected.
+        if let Some((score, legacy_id)) = best
+            && score >= AUTO_ACCEPT_THRESHOLD
+            && !rejected_pairs.contains(&(*instructor_id, legacy_id))
+        {
+            auto_accept.push((*instructor_id, legacy_id));
+        }
+    }
+
+    // 6–7. Write candidates and auto-accept within a single transaction
+    let candidates_created = candidates.len();
+    let auto_matched = auto_accept.len();
+
+    let mut tx = db_pool.begin().await?;
+
+    // 6. Batch-insert candidates
+    if !candidates.is_empty() {
+        let c_instructor_ids: Vec<i32> = candidates.iter().map(|(iid, _, _, _)| *iid).collect();
+        let c_legacy_ids: Vec<i32> = candidates.iter().map(|(_, lid, _, _)| *lid).collect();
+        let c_scores: Vec<f32> = candidates.iter().map(|(_, _, s, _)| *s).collect();
+        let c_breakdowns: Vec<serde_json::Value> =
+            candidates.into_iter().map(|(_, _, _, b)| b).collect();
+
+        sqlx::query(
+            r#"
+            INSERT INTO rmp_match_candidates (instructor_id, rmp_legacy_id, score, score_breakdown)
+            SELECT v.instructor_id, v.rmp_legacy_id, v.score, v.score_breakdown
+            FROM UNNEST($1::int4[], $2::int4[], $3::real[], $4::jsonb[])
+                AS v(instructor_id, rmp_legacy_id, score, score_breakdown)
+            ON CONFLICT (instructor_id, rmp_legacy_id) DO NOTHING
+            "#,
+        )
+        .bind(&c_instructor_ids)
+        .bind(&c_legacy_ids)
+        .bind(&c_scores)
+        .bind(&c_breakdowns)
+        .execute(&mut *tx)
+        .await?;
+    }
+
+    // 7. Auto-accept top candidates
+    if !auto_accept.is_empty() {
+        let aa_instructor_ids: Vec<i32> = auto_accept.iter().map(|(iid, _)| *iid).collect();
+        let aa_legacy_ids: Vec<i32> = auto_accept.iter().map(|(_, lid)| *lid).collect();
+
+        // Mark the candidate row as accepted
+        sqlx::query(
+            r#"
+            UPDATE rmp_match_candidates mc
+            SET status = 'accepted', resolved_at = NOW()
+            FROM UNNEST($1::int4[], $2::int4[]) AS v(instructor_id, rmp_legacy_id)
+            WHERE mc.instructor_id = v.instructor_id
+              AND mc.rmp_legacy_id = v.rmp_legacy_id
+            "#,
+        )
+        .bind(&aa_instructor_ids)
+        .bind(&aa_legacy_ids)
+        .execute(&mut *tx)
+        .await?;
+
+        // Insert links into instructor_rmp_links
+        sqlx::query(
+            r#"
+            INSERT INTO instructor_rmp_links (instructor_id, rmp_legacy_id, source)
+            SELECT v.instructor_id, v.rmp_legacy_id, 'auto'
+            FROM UNNEST($1::int4[], $2::int4[]) AS v(instructor_id, rmp_legacy_id)
+            ON CONFLICT (rmp_legacy_id) DO NOTHING
+            "#,
+        )
+        .bind(&aa_instructor_ids)
+        .bind(&aa_legacy_ids)
+        .execute(&mut *tx)
+        .await?;
+
+        // Update instructor match status
+        sqlx::query(
+            r#"
+            UPDATE instructors i
+            SET rmp_match_status = 'auto'
+            FROM UNNEST($1::int4[]) AS v(instructor_id)
+            WHERE i.id = v.instructor_id
+            "#,
+        )
+        .bind(&aa_instructor_ids)
+        .execute(&mut *tx)
+        .await?;
+    }
+
+    tx.commit().await?;
+
+    let stats = MatchingStats {
+        total_unmatched,
+        candidates_created,
+        auto_matched,
+        skipped_unparseable,
+        skipped_no_candidates,
+    };
+
+    info!(
+        total_unmatched = stats.total_unmatched,
+        candidates_created = stats.candidates_created,
+        auto_matched = stats.auto_matched,
+        skipped_unparseable = stats.skipped_unparseable,
+        skipped_no_candidates = stats.skipped_no_candidates,
+        "Candidate generation complete"
+    );
+
+    Ok(stats)
+}
+
+// ---------------------------------------------------------------------------
+// Tests
+// ---------------------------------------------------------------------------
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn test_ideal_candidate_high_score() {
+        let ms = compute_match_score(
+            &["CS".to_string()],
+            Some("Computer Science"),
+            1,  // unique candidate
+            50, // decent ratings
+        );
+        // dept 1.0*0.50 + unique 1.0*0.30 + volume ~0.97*0.20 ≈ 0.99
+        assert!(ms.score >= 0.85, "Expected score >= 0.85, got {}", ms.score);
+        assert_eq!(ms.breakdown.uniqueness, 1.0);
+        assert_eq!(ms.breakdown.department, 1.0);
+    }
+
+    #[test]
+    fn test_ambiguous_candidates_lower_score() {
+        let unique = compute_match_score(&[], None, 1, 10);
+        let ambiguous = compute_match_score(&[], None, 3, 10);
+        assert!(
+            unique.score > ambiguous.score,
+            "Unique ({}) should outscore ambiguous ({})",
+            unique.score,
+            ambiguous.score
+        );
+        assert_eq!(unique.breakdown.uniqueness, 1.0);
+        assert_eq!(ambiguous.breakdown.uniqueness, 0.2);
+    }
+
+    #[test]
+    fn test_no_department_neutral() {
+        let ms = compute_match_score(&["CS".to_string()], None, 1, 10);
+        assert_eq!(ms.breakdown.department, 0.5);
+    }
+
+    #[test]
+    fn test_department_match() {
+        let ms = compute_match_score(&["CS".to_string()], Some("Computer Science"), 1, 10);
+        assert_eq!(ms.breakdown.department, 1.0);
+    }
+
+    #[test]
+    fn test_department_mismatch() {
+        let ms = compute_match_score(&["CS".to_string()], Some("History"), 1, 10);
+        assert_eq!(ms.breakdown.department, 0.2);
+    }
+
+    #[test]
+    fn test_department_match_outscores_mismatch() {
+        let matched = compute_match_score(&["CS".to_string()], Some("Computer Science"), 1, 10);
+        let mismatched = compute_match_score(&["CS".to_string()], Some("History"), 1, 10);
+        assert!(
+            matched.score > mismatched.score,
+            "Department match ({}) should outscore mismatch ({})",
+            matched.score,
+            mismatched.score
+        );
+    }
+
+    #[test]
+    fn test_volume_scaling() {
+        let zero = compute_match_score(&[], None, 1, 0);
+        let many = compute_match_score(&[], None, 1, 100);
+        assert!(
+            many.breakdown.volume > zero.breakdown.volume,
+            "100 ratings ({}) should outscore 0 ratings ({})",
+            many.breakdown.volume,
+            zero.breakdown.volume
+        );
+        assert_eq!(zero.breakdown.volume, 0.0);
+        assert!(
+            many.breakdown.volume > 0.9,
+            "100 ratings should be near max"
+        );
+    }
+}
@@ -1,15 +1,40 @@
 //! Database operations for scrape job queue management.

-use crate::data::models::{ScrapeJob, ScrapePriority, TargetType};
+use crate::data::models::{ScrapeJob, ScrapePriority, TargetType, UpsertCounts};
 use crate::error::Result;
+use chrono::{DateTime, Utc};
 use sqlx::PgPool;
 use std::collections::HashSet;

+/// Force-unlock all jobs that have a non-NULL `locked_at`.
+///
+/// Intended to be called once at startup to recover jobs left locked by
+/// a previous unclean shutdown (crash, OOM kill, etc.).
+///
+/// # Returns
+/// The number of jobs that were unlocked.
+pub async fn force_unlock_all(db_pool: &PgPool) -> Result<u64> {
+    let result = sqlx::query(
+        "UPDATE scrape_jobs SET locked_at = NULL, queued_at = NOW() WHERE locked_at IS NOT NULL",
+    )
+    .execute(db_pool)
+    .await?;
+    Ok(result.rows_affected())
+}
+
+/// How long a lock can be held before it is considered expired and reclaimable.
+///
+/// This acts as a safety net for cases where a worker dies without unlocking
+/// (OOM kill, crash, network partition). Under normal operation, the worker's
+/// own job timeout fires well before this threshold.
+const LOCK_EXPIRY: std::time::Duration = std::time::Duration::from_secs(10 * 60);
+
 /// Atomically fetch and lock the next available scrape job.
 ///
 /// Uses `FOR UPDATE SKIP LOCKED` to allow multiple workers to poll the queue
-/// concurrently without conflicts. Only jobs that are unlocked and ready to
-/// execute (based on `execute_at`) are considered.
+/// concurrently without conflicts. Considers jobs that are:
+/// - Unlocked and ready to execute, OR
+/// - Locked but past [`LOCK_EXPIRY`] (abandoned by a dead worker)
 ///
 /// # Arguments
 /// * `db_pool` - PostgreSQL connection pool
@@ -20,9 +45,16 @@ use std::collections::HashSet;
 pub async fn fetch_and_lock_job(db_pool: &PgPool) -> Result<Option<ScrapeJob>> {
    let mut tx = db_pool.begin().await?;

+    let lock_expiry_secs = LOCK_EXPIRY.as_secs() as i32;
    let job = sqlx::query_as::<_, ScrapeJob>(
-        "SELECT * FROM scrape_jobs WHERE locked_at IS NULL AND execute_at <= NOW() ORDER BY priority DESC, execute_at ASC LIMIT 1 FOR UPDATE SKIP LOCKED"
+        "SELECT * FROM scrape_jobs \
+         WHERE (locked_at IS NULL OR locked_at < NOW() - make_interval(secs => $1::double precision)) \
+         AND execute_at <= NOW() \
+         ORDER BY priority DESC, execute_at ASC \
+         LIMIT 1 \
+         FOR UPDATE SKIP LOCKED"
    )
+    .bind(lock_expiry_secs)
    .fetch_optional(&mut *tx)
    .await?;

@@ -68,10 +100,11 @@ pub async fn unlock_job(job_id: i32, db_pool: &PgPool) -> Result<()> {
    Ok(())
 }

-/// Atomically unlock a job and increment its retry count.
+/// Atomically unlock a job, increment its retry count, and reset `queued_at`.
 ///
-/// Returns whether the job still has retries remaining. This is determined
-/// atomically in the database to avoid race conditions between workers.
+/// Returns the new `queued_at` timestamp if retries remain, or `None` if
+/// the job has exhausted its retries. This is determined atomically in the
+/// database to avoid race conditions between workers.
 ///
 /// # Arguments
 /// * `job_id` - The database ID of the job
@@ -79,31 +112,31 @@ pub async fn unlock_job(job_id: i32, db_pool: &PgPool) -> Result<()> {
 /// * `db_pool` - PostgreSQL connection pool
 ///
 /// # Returns
-/// * `Ok(true)` if the job was unlocked and retries remain
-/// * `Ok(false)` if the job has exhausted its retries
+/// * `Ok(Some(queued_at))` if the job was unlocked and retries remain
+/// * `Ok(None)` if the job has exhausted its retries
 pub async fn unlock_and_increment_retry(
    job_id: i32,
    max_retries: i32,
    db_pool: &PgPool,
-) -> Result<bool> {
-    let result = sqlx::query_scalar::<_, Option<i32>>(
+) -> Result<Option<chrono::DateTime<chrono::Utc>>> {
+    let result = sqlx::query_scalar::<_, Option<chrono::DateTime<chrono::Utc>>>(
        "UPDATE scrape_jobs
-         SET locked_at = NULL, retry_count = retry_count + 1
+         SET locked_at = NULL, retry_count = retry_count + 1, queued_at = NOW()
         WHERE id = $1
-         RETURNING CASE WHEN retry_count < $2 THEN retry_count ELSE NULL END",
+         RETURNING CASE WHEN retry_count <= $2 THEN queued_at ELSE NULL END",
    )
    .bind(job_id)
    .bind(max_retries)
    .fetch_one(db_pool)
    .await?;

-    Ok(result.is_some())
+    Ok(result)
 }

-/// Find existing unlocked job payloads matching the given target type and candidates.
+/// Find existing job payloads matching the given target type and candidates.
 ///
-/// Returns a set of stringified JSON payloads that already exist in the queue,
-/// used for deduplication when scheduling new jobs.
+/// Returns a set of stringified JSON payloads that already exist in the queue
+/// (both locked and unlocked), used for deduplication when scheduling new jobs.
 ///
 /// # Arguments
 /// * `target_type` - The target type to filter by
@@ -111,7 +144,7 @@ pub async fn unlock_and_increment_retry(
 /// * `db_pool` - PostgreSQL connection pool
 ///
 /// # Returns
-/// A `HashSet` of stringified JSON payloads that already have pending jobs
+/// A `HashSet` of stringified JSON payloads that already have pending or in-progress jobs
 pub async fn find_existing_job_payloads(
    target_type: TargetType,
    candidate_payloads: &[serde_json::Value],
@@ -119,7 +152,7 @@ pub async fn find_existing_job_payloads(
 ) -> Result<HashSet<String>> {
    let existing_jobs: Vec<(serde_json::Value,)> = sqlx::query_as(
        "SELECT target_payload FROM scrape_jobs
-         WHERE target_type = $1 AND target_payload = ANY($2) AND locked_at IS NULL",
+         WHERE target_type = $1 AND target_payload = ANY($2)",
    )
    .bind(target_type)
    .bind(candidate_payloads)
@@ -134,7 +167,117 @@ pub async fn find_existing_job_payloads(
    Ok(existing_payloads)
 }

-/// Batch insert scrape jobs in a single transaction.
+/// Insert a scrape job result log entry.
+#[allow(clippy::too_many_arguments)]
+pub async fn insert_job_result(
+    target_type: TargetType,
+    payload: serde_json::Value,
+    priority: ScrapePriority,
+    queued_at: DateTime<Utc>,
+    started_at: DateTime<Utc>,
+    duration_ms: i32,
+    success: bool,
+    error_message: Option<&str>,
+    retry_count: i32,
+    counts: Option<&UpsertCounts>,
+    db_pool: &PgPool,
+) -> Result<()> {
+    sqlx::query(
+        r#"
+        INSERT INTO scrape_job_results (
+            target_type, payload, priority,
+            queued_at, started_at, duration_ms,
+            success, error_message, retry_count,
+            courses_fetched, courses_changed, courses_unchanged,
+            audits_generated, metrics_generated
+        ) VALUES ($1, $2, $3, $4, $5, $6, $7, $8, $9, $10, $11, $12, $13, $14)
+        "#,
+    )
+    .bind(target_type)
+    .bind(&payload)
+    .bind(priority)
+    .bind(queued_at)
+    .bind(started_at)
+    .bind(duration_ms)
+    .bind(success)
+    .bind(error_message)
+    .bind(retry_count)
+    .bind(counts.map(|c| c.courses_fetched))
+    .bind(counts.map(|c| c.courses_changed))
+    .bind(counts.map(|c| c.courses_unchanged))
+    .bind(counts.map(|c| c.audits_generated))
+    .bind(counts.map(|c| c.metrics_generated))
+    .execute(db_pool)
+    .await?;
+
+    Ok(())
+}
+
+/// Per-subject aggregated stats from recent scrape results.
+///
+/// Populated by [`fetch_subject_stats`] and converted into
+/// [`crate::scraper::adaptive::SubjectStats`] for interval computation.
+#[derive(sqlx::FromRow, Debug, Clone)]
+pub struct SubjectResultStats {
+    pub subject: String,
+    pub recent_runs: i64,
+    pub avg_change_ratio: f64,
+    pub consecutive_zero_changes: i64,
+    pub consecutive_empty_fetches: i64,
+    pub recent_failure_count: i64,
+    pub recent_success_count: i64,
+    pub last_completed: DateTime<Utc>,
+}
+
+/// Fetch aggregated per-subject statistics from the last 24 hours of results.
+///
+/// For each subject, examines the 20 most recent results and computes:
+/// - Average change ratio (courses_changed / courses_fetched)
+/// - Consecutive zero-change runs from the most recent result
+/// - Consecutive empty-fetch runs from the most recent result
+/// - Failure and success counts
+/// - Last completion timestamp
+pub async fn fetch_subject_stats(db_pool: &PgPool) -> Result<Vec<SubjectResultStats>> {
+    let rows = sqlx::query_as::<_, SubjectResultStats>(
+        r#"
+        WITH recent AS (
+            SELECT payload->>'subject' AS subject, success,
+                   COALESCE(courses_fetched, 0) AS courses_fetched,
+                   COALESCE(courses_changed, 0) AS courses_changed,
+                   completed_at,
+                   ROW_NUMBER() OVER (PARTITION BY payload->>'subject' ORDER BY completed_at DESC) AS rn
+            FROM scrape_job_results
+            WHERE target_type = 'Subject' AND completed_at > NOW() - INTERVAL '24 hours'
+        ),
+        filtered AS (SELECT * FROM recent WHERE rn <= 20),
+        zero_break AS (
+            SELECT subject,
+                   MIN(rn) FILTER (WHERE courses_changed > 0 AND success) AS first_nonzero_rn,
+                   MIN(rn) FILTER (WHERE courses_fetched > 0 AND success) AS first_nonempty_rn
+            FROM filtered GROUP BY subject
+        )
+        SELECT
+            f.subject::TEXT AS subject,
+            COUNT(*)::BIGINT AS recent_runs,
+            COALESCE(AVG(CASE WHEN f.success AND f.courses_fetched > 0
+                 THEN f.courses_changed::FLOAT / f.courses_fetched ELSE NULL END), 0.0)::FLOAT8 AS avg_change_ratio,
+            COALESCE(zb.first_nonzero_rn - 1, COUNT(*) FILTER (WHERE f.success AND f.courses_changed = 0))::BIGINT AS consecutive_zero_changes,
+            COALESCE(zb.first_nonempty_rn - 1, COUNT(*) FILTER (WHERE f.success AND f.courses_fetched = 0))::BIGINT AS consecutive_empty_fetches,
+            COUNT(*) FILTER (WHERE NOT f.success)::BIGINT AS recent_failure_count,
+            COUNT(*) FILTER (WHERE f.success)::BIGINT AS recent_success_count,
+            MAX(f.completed_at) AS last_completed
+        FROM filtered f
+        LEFT JOIN zero_break zb ON f.subject = zb.subject
+        GROUP BY f.subject, zb.first_nonzero_rn, zb.first_nonempty_rn
+        "#,
+    )
+    .fetch_all(db_pool)
+    .await?;
+
+    Ok(rows)
+}
+
+/// Batch insert scrape jobs using UNNEST for a single round-trip.
 ///
 /// All jobs are inserted with `execute_at` set to the current time.
 ///
@@ -144,27 +287,35 @@ pub async fn find_existing_job_payloads(
 pub async fn batch_insert_jobs(
    jobs: &[(serde_json::Value, TargetType, ScrapePriority)],
    db_pool: &PgPool,
-) -> Result<()> {
+) -> Result<Vec<ScrapeJob>> {
    if jobs.is_empty() {
-        return Ok(());
+        return Ok(Vec::new());
    }

-    let now = chrono::Utc::now();
-    let mut tx = db_pool.begin().await?;
+    let mut target_types: Vec<String> = Vec::with_capacity(jobs.len());
+    let mut payloads: Vec<serde_json::Value> = Vec::with_capacity(jobs.len());
+    let mut priorities: Vec<String> = Vec::with_capacity(jobs.len());

    for (payload, target_type, priority) in jobs {
-        sqlx::query(
-            "INSERT INTO scrape_jobs (target_type, target_payload, priority, execute_at) VALUES ($1, $2, $3, $4)"
-        )
-        .bind(target_type)
-        .bind(payload)
-        .bind(priority)
-        .bind(now)
-        .execute(&mut *tx)
-        .await?;
+        target_types.push(format!("{target_type:?}"));
+        payloads.push(payload.clone());
+        priorities.push(format!("{priority:?}"));
    }

-    tx.commit().await?;
+    let inserted = sqlx::query_as::<_, ScrapeJob>(
+        r#"
+        INSERT INTO scrape_jobs (target_type, target_payload, priority, execute_at, queued_at)
+        SELECT v.target_type::target_type, v.payload, v.priority::scrape_priority, NOW(), NOW()
+        FROM UNNEST($1::text[], $2::jsonb[], $3::text[])
+            AS v(target_type, payload, priority)
+        RETURNING *
+        "#,
+    )
+    .bind(&target_types)
+    .bind(&payloads)
+    .bind(&priorities)
+    .fetch_all(db_pool)
+    .await?;

-    Ok(())
+    Ok(inserted)
 }
@@ -0,0 +1,90 @@
+//! Database query functions for user sessions.
+
+use anyhow::Context;
+use rand::Rng;
+use sqlx::PgPool;
+
+use super::models::UserSession;
+use crate::error::Result;
+
+/// Generate a cryptographically random 32-byte hex token.
+fn generate_token() -> String {
+    let bytes: [u8; 32] = rand::rng().random();
+    bytes.iter().map(|b| format!("{b:02x}")).collect()
+}
+
+/// Create a new session for a user with the given duration.
+pub async fn create_session(
+    pool: &PgPool,
+    user_id: i64,
+    duration: std::time::Duration,
+) -> Result<UserSession> {
+    let token = generate_token();
+    let duration_secs = duration.as_secs() as i64;
+
+    sqlx::query_as::<_, UserSession>(
+        r#"
+        INSERT INTO user_sessions (id, user_id, expires_at)
+        VALUES ($1, $2, now() + make_interval(secs => $3::double precision))
+        RETURNING *
+        "#,
+    )
+    .bind(&token)
+    .bind(user_id)
+    .bind(duration_secs as f64)
+    .fetch_one(pool)
+    .await
+    .context("failed to create session")
+}
+
+/// Fetch a session by token, only if it has not expired.
+pub async fn get_session(pool: &PgPool, token: &str) -> Result<Option<UserSession>> {
+    sqlx::query_as::<_, UserSession>(
+        "SELECT * FROM user_sessions WHERE id = $1 AND expires_at > now()",
+    )
+    .bind(token)
+    .fetch_optional(pool)
+    .await
+    .context("failed to get session")
+}
+
+/// Update the last-active timestamp for a session.
+pub async fn touch_session(pool: &PgPool, token: &str) -> Result<()> {
+    sqlx::query("UPDATE user_sessions SET last_active_at = now() WHERE id = $1")
+        .bind(token)
+        .execute(pool)
+        .await
+        .context("failed to touch session")?;
+    Ok(())
+}
+
+/// Delete a session by token.
+pub async fn delete_session(pool: &PgPool, token: &str) -> Result<()> {
+    sqlx::query("DELETE FROM user_sessions WHERE id = $1")
+        .bind(token)
+        .execute(pool)
+        .await
+        .context("failed to delete session")?;
+    Ok(())
+}
+
+/// Delete all sessions for a user. Returns the number of sessions deleted.
+#[allow(dead_code)] // Available for admin user-deletion flow
+pub async fn delete_user_sessions(pool: &PgPool, user_id: i64) -> Result<u64> {
+    let result = sqlx::query("DELETE FROM user_sessions WHERE user_id = $1")
+        .bind(user_id)
+        .execute(pool)
+        .await
+        .context("failed to delete user sessions")?;
+    Ok(result.rows_affected())
+}
+
+/// Delete all expired sessions. Returns the number of sessions cleaned up.
+#[allow(dead_code)] // Called by SessionCache::cleanup_expired (not yet wired to periodic task)
+pub async fn cleanup_expired(pool: &PgPool) -> Result<u64> {
+    let result = sqlx::query("DELETE FROM user_sessions WHERE expires_at <= now()")
+        .execute(pool)
+        .await
+        .context("failed to cleanup expired sessions")?;
+    Ok(result.rows_affected())
+}
@@ -0,0 +1,86 @@
+//! Database query functions for users.
+
+use anyhow::Context;
+use sqlx::PgPool;
+
+use super::models::User;
+use crate::error::Result;
+
+/// Insert a new user or update username/avatar on conflict.
+pub async fn upsert_user(
+    pool: &PgPool,
+    discord_id: i64,
+    username: &str,
+    avatar_hash: Option<&str>,
+) -> Result<User> {
+    sqlx::query_as::<_, User>(
+        r#"
+        INSERT INTO users (discord_id, discord_username, discord_avatar_hash)
+        VALUES ($1, $2, $3)
+        ON CONFLICT (discord_id) DO UPDATE
+            SET discord_username = EXCLUDED.discord_username,
+                discord_avatar_hash = EXCLUDED.discord_avatar_hash,
+                updated_at = now()
+        RETURNING *
+        "#,
+    )
+    .bind(discord_id)
+    .bind(username)
+    .bind(avatar_hash)
+    .fetch_one(pool)
+    .await
+    .context("failed to upsert user")
+}
+
+/// Fetch a user by Discord ID.
+pub async fn get_user(pool: &PgPool, discord_id: i64) -> Result<Option<User>> {
+    sqlx::query_as::<_, User>("SELECT * FROM users WHERE discord_id = $1")
+        .bind(discord_id)
+        .fetch_optional(pool)
+        .await
+        .context("failed to get user")
+}
+
+/// List all users ordered by creation date (newest first).
+pub async fn list_users(pool: &PgPool) -> Result<Vec<User>> {
+    sqlx::query_as::<_, User>("SELECT * FROM users ORDER BY created_at DESC")
+        .fetch_all(pool)
+        .await
+        .context("failed to list users")
+}
+
+/// Set the admin flag for a user, returning the updated user if found.
+pub async fn set_admin(pool: &PgPool, discord_id: i64, is_admin: bool) -> Result<Option<User>> {
+    sqlx::query_as::<_, User>(
+        r#"
+        UPDATE users
+        SET is_admin = $2, updated_at = now()
+        WHERE discord_id = $1
+        RETURNING *
+        "#,
+    )
+    .bind(discord_id)
+    .bind(is_admin)
+    .fetch_optional(pool)
+    .await
+    .context("failed to set admin status")
+}
+
+/// Ensure a seed admin exists. Upserts with `is_admin = true` and a placeholder
+/// username that will be replaced on first OAuth login.
+pub async fn ensure_seed_admin(pool: &PgPool, discord_id: i64) -> Result<User> {
+    sqlx::query_as::<_, User>(
+        r#"
+        INSERT INTO users (discord_id, discord_username, is_admin)
+        VALUES ($1, 'seed-admin', true)
+        ON CONFLICT (discord_id) DO UPDATE
+            SET is_admin = true,
+                updated_at = now()
+        RETURNING *
+        "#,
+    )
+    .bind(discord_id)
+    .fetch_one(pool)
+    .await
+    .context("failed to ensure seed admin")
+}
@@ -1,12 +1,14 @@
 pub mod app;
 pub mod banner;
 pub mod bot;
+pub mod calendar;
 pub mod cli;
 pub mod config;
 pub mod data;
 pub mod error;
 pub mod formatter;
 pub mod logging;
+pub mod rmp;
 pub mod scraper;
 pub mod services;
 pub mod signals;
@@ -1,5 +1,5 @@
 use crate::app::App;
-use crate::cli::{Args, ServiceName, determine_enabled_services};
+use crate::cli::{Args, ServiceName};
 use crate::logging::setup_logging;
 use clap::Parser;
 use std::process::ExitCode;
@@ -8,17 +8,18 @@ use tracing::info;
 mod app;
 mod banner;
 mod bot;
+mod calendar;
 mod cli;
 mod config;
 mod data;
 mod error;
 mod formatter;
 mod logging;
+mod rmp;
 mod scraper;
 mod services;
 mod signals;
 mod state;
-#[allow(dead_code)]
 mod status;
 mod web;

@@ -29,9 +30,8 @@ async fn main() -> ExitCode {
    // Parse CLI arguments
    let args = Args::parse();

-    // Determine which services should be enabled
-    let enabled_services: Vec<ServiceName> =
-        determine_enabled_services(&args).expect("Failed to determine enabled services");
+    // Always run all services
+    let enabled_services = ServiceName::all();

    // Create and initialize the application
    let mut app = App::new().await.expect("Failed to initialize application");
@@ -0,0 +1,156 @@
+//! RateMyProfessors GraphQL client for bulk professor data sync.
+
+use anyhow::Result;
+use serde::{Deserialize, Serialize};
+use tracing::{debug, info};
+
+/// UTSA's school ID on RateMyProfessors (base64 of "School-1516").
+const UTSA_SCHOOL_ID: &str = "U2Nob29sLTE1MTY=";
+
+/// Basic auth header value (base64 of "test:test").
+const AUTH_HEADER: &str = "Basic dGVzdDp0ZXN0";
+
+/// GraphQL endpoint.
+const GRAPHQL_URL: &str = "https://www.ratemyprofessors.com/graphql";
+
+/// Page size for paginated fetches.
+const PAGE_SIZE: u32 = 100;
+
+/// A professor record from RateMyProfessors.
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct RmpProfessor {
+    pub legacy_id: i32,
+    pub graphql_id: String,
+    pub first_name: String,
+    pub last_name: String,
+    pub department: Option<String>,
+    pub avg_rating: Option<f32>,
+    pub avg_difficulty: Option<f32>,
+    pub num_ratings: i32,
+    pub would_take_again_pct: Option<f32>,
+}
+
+/// Client for fetching professor data from RateMyProfessors.
+pub struct RmpClient {
+    http: reqwest::Client,
+}
+
+impl Default for RmpClient {
+    fn default() -> Self {
+        Self::new()
+    }
+}
+
+impl RmpClient {
+    pub fn new() -> Self {
+        Self {
+            http: reqwest::Client::new(),
+        }
+    }
+
+    /// Fetch all professors for UTSA via paginated GraphQL queries.
+    pub async fn fetch_all_professors(&self) -> Result<Vec<RmpProfessor>> {
+        let mut all = Vec::new();
+        let mut cursor: Option<String> = None;
+
+        loop {
+            let after_clause = match &cursor {
+                Some(c) => format!(r#", after: "{}""#, c),
+                None => String::new(),
+            };
+
+            let query = format!(
+                r#"query {{
+  newSearch {{
+    teachers(query: {{ text: "", schoolID: "{school_id}" }}, first: {page_size}{after}) {{
+      edges {{
+        cursor
+        node {{
+          id
+          legacyId
+          firstName
+          lastName
+          department
+          avgRating
+          avgDifficulty
+          numRatings
+          wouldTakeAgainPercent
+        }}
+      }}
+      pageInfo {{
+        hasNextPage
+        endCursor
+      }}
+    }}
+  }}
+}}"#,
+                school_id = UTSA_SCHOOL_ID,
+                page_size = PAGE_SIZE,
+                after = after_clause,
+            );
+
+            let body = serde_json::json!({ "query": query });
+
+            let resp = self
+                .http
+                .post(GRAPHQL_URL)
+                .header("Authorization", AUTH_HEADER)
+                .json(&body)
+                .send()
+                .await?;
+
+            let status = resp.status();
+            if !status.is_success() {
+                let text = resp.text().await.unwrap_or_default();
+                anyhow::bail!("RMP GraphQL request failed ({status}): {text}");
+            }
+
+            let json: serde_json::Value = resp.json().await?;
+
+            let teachers = &json["data"]["newSearch"]["teachers"];
+            let edges = teachers["edges"]
+                .as_array()
+                .ok_or_else(|| anyhow::anyhow!("Missing edges in RMP response"))?;
+
+            for edge in edges {
+                let node = &edge["node"];
+                let wta = node["wouldTakeAgainPercent"]
+                    .as_f64()
+                    .map(|v| v as f32)
+                    .filter(|&v| v >= 0.0);
+
+                all.push(RmpProfessor {
+                    legacy_id: node["legacyId"]
+                        .as_i64()
+                        .ok_or_else(|| anyhow::anyhow!("Missing legacyId"))?
+                        as i32,
+                    graphql_id: node["id"]
+                        .as_str()
+                        .ok_or_else(|| anyhow::anyhow!("Missing id"))?
+                        .to_string(),
+                    first_name: node["firstName"].as_str().unwrap_or_default().to_string(),
+                    last_name: node["lastName"].as_str().unwrap_or_default().to_string(),
+                    department: node["department"].as_str().map(|s| s.to_string()),
+                    avg_rating: node["avgRating"].as_f64().map(|v| v as f32),
+                    avg_difficulty: node["avgDifficulty"].as_f64().map(|v| v as f32),
+                    num_ratings: node["numRatings"].as_i64().unwrap_or(0) as i32,
+                    would_take_again_pct: wta,
+                });
+            }
+
+            let page_info = &teachers["pageInfo"];
+            let has_next = page_info["hasNextPage"].as_bool().unwrap_or(false);
+
+            if !has_next {
+                break;
+            }
+
+            cursor = page_info["endCursor"].as_str().map(|s| s.to_string());
+
+            debug!(fetched = all.len(), "RMP pagination: fetching next page");
+        }
+
+        info!(total = all.len(), "Fetched all RMP professors");
+        Ok(all)
+    }
+}
@@ -0,0 +1,326 @@
+//! Adaptive scraping interval computation.
+//!
+//! Assigns per-subject scrape intervals based on recent change rates,
+//! consecutive zero-change runs, failure patterns, and time of day.
+
+use chrono::{DateTime, Datelike, Timelike, Utc};
+use chrono_tz::US::Central;
+use std::time::Duration;
+
+use crate::data::scrape_jobs::SubjectResultStats;
+
+const FLOOR_INTERVAL: Duration = Duration::from_secs(3 * 60);
+const MODERATE_HIGH_INTERVAL: Duration = Duration::from_secs(5 * 60);
+const MODERATE_LOW_INTERVAL: Duration = Duration::from_secs(15 * 60);
+const LOW_CHANGE_INTERVAL: Duration = Duration::from_secs(30 * 60);
+const ZERO_5_INTERVAL: Duration = Duration::from_secs(60 * 60);
+const ZERO_10_INTERVAL: Duration = Duration::from_secs(2 * 60 * 60);
+const CEILING_INTERVAL: Duration = Duration::from_secs(4 * 60 * 60);
+const COLD_START_INTERVAL: Duration = FLOOR_INTERVAL;
+const PAUSE_PROBE_INTERVAL: Duration = Duration::from_secs(6 * 60 * 60);
+const EMPTY_FETCH_PAUSE_THRESHOLD: i64 = 3;
+const FAILURE_PAUSE_THRESHOLD: i64 = 5;
+
+/// Aggregated per-subject statistics derived from recent scrape results.
+#[derive(Debug, Clone)]
+pub struct SubjectStats {
+    pub subject: String,
+    pub recent_runs: i64,
+    pub avg_change_ratio: f64,
+    pub consecutive_zero_changes: i64,
+    pub consecutive_empty_fetches: i64,
+    pub recent_failure_count: i64,
+    pub recent_success_count: i64,
+    pub last_completed: DateTime<Utc>,
+}
+
+/// Scheduling decision for a subject.
+#[derive(Debug, Clone, PartialEq, Eq)]
+pub enum SubjectSchedule {
+    /// Subject is due for scraping, with the computed interval.
+    Eligible(Duration),
+    /// Subject was scraped recently; wait for the remaining cooldown.
+    Cooldown(Duration),
+    /// Subject is paused due to repeated empty fetches or failures.
+    Paused,
+    /// Subject belongs to a past term and should not be scraped.
+    ReadOnly,
+}
+
+impl From<SubjectResultStats> for SubjectStats {
+    fn from(row: SubjectResultStats) -> Self {
+        Self {
+            subject: row.subject,
+            recent_runs: row.recent_runs,
+            avg_change_ratio: row.avg_change_ratio,
+            consecutive_zero_changes: row.consecutive_zero_changes,
+            consecutive_empty_fetches: row.consecutive_empty_fetches,
+            recent_failure_count: row.recent_failure_count,
+            recent_success_count: row.recent_success_count,
+            last_completed: row.last_completed,
+        }
+    }
+}
+
+/// Compute the base interval tier from change-rate statistics.
+pub fn compute_base_interval(stats: &SubjectStats) -> Duration {
+    if stats.recent_runs == 0 {
+        return COLD_START_INTERVAL;
+    }
+
+    // Consecutive-zero tiers take precedence when change ratio is near zero
+    if stats.avg_change_ratio < 0.001 {
+        return match stats.consecutive_zero_changes {
+            0..5 => LOW_CHANGE_INTERVAL,
+            5..10 => ZERO_5_INTERVAL,
+            10..20 => ZERO_10_INTERVAL,
+            _ => CEILING_INTERVAL,
+        };
+    }
+
+    match stats.avg_change_ratio {
+        r if r >= 0.10 => FLOOR_INTERVAL,
+        r if r >= 0.05 => MODERATE_HIGH_INTERVAL,
+        r if r >= 0.01 => MODERATE_LOW_INTERVAL,
+        _ => LOW_CHANGE_INTERVAL,
+    }
+}
+
+/// Return a time-of-day multiplier for the given UTC timestamp.
+///
+/// Peak hours (weekdays 8am-6pm CT) return 1; off-peak (weekdays 6pm-midnight CT)
+/// return 2; night (midnight-8am CT) and weekends return 4.
+pub fn time_of_day_multiplier(now: DateTime<Utc>) -> u32 {
+    let ct = now.with_timezone(&Central);
+    let weekday = ct.weekday();
+    let hour = ct.hour();
+
+    // Weekends get the slowest multiplier
+    if matches!(weekday, chrono::Weekday::Sat | chrono::Weekday::Sun) {
+        return 4;
+    }
+
+    match hour {
+        8..18 => 1,  // peak
+        18..24 => 2, // off-peak
+        _ => 4,      // night (0..8)
+    }
+}
+
+/// Evaluate whether a subject should be scraped now.
+///
+/// Combines base interval, time-of-day multiplier, pause detection (empty
+/// fetches / consecutive failures), and past-term read-only status.
+pub fn evaluate_subject(
+    stats: &SubjectStats,
+    now: DateTime<Utc>,
+    is_past_term: bool,
+) -> SubjectSchedule {
+    if is_past_term {
+        return SubjectSchedule::ReadOnly;
+    }
+
+    let elapsed = (now - stats.last_completed)
+        .to_std()
+        .unwrap_or(Duration::ZERO);
+    let probe_due = elapsed >= PAUSE_PROBE_INTERVAL;
+
+    // Pause on repeated empty fetches
+    if stats.consecutive_empty_fetches >= EMPTY_FETCH_PAUSE_THRESHOLD {
+        return if probe_due {
+            SubjectSchedule::Eligible(PAUSE_PROBE_INTERVAL)
+        } else {
+            SubjectSchedule::Paused
+        };
+    }
+
+    // Pause on all-failures
+    if stats.recent_success_count == 0 && stats.recent_failure_count >= FAILURE_PAUSE_THRESHOLD {
+        return if probe_due {
+            SubjectSchedule::Eligible(PAUSE_PROBE_INTERVAL)
+        } else {
+            SubjectSchedule::Paused
+        };
+    }
+
+    let base = compute_base_interval(stats);
+    let multiplier = time_of_day_multiplier(now);
+    let effective = base * multiplier;
+
+    if elapsed >= effective {
+        SubjectSchedule::Eligible(effective)
+    } else {
+        let remaining = effective - elapsed;
+        SubjectSchedule::Cooldown(remaining)
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use chrono::TimeZone;
+
+    /// Create a default `SubjectStats` for testing. Callers mutate fields as needed.
+    fn make_stats(subject: &str) -> SubjectStats {
+        SubjectStats {
+            subject: subject.to_string(),
+            recent_runs: 10,
+            avg_change_ratio: 0.0,
+            consecutive_zero_changes: 0,
+            consecutive_empty_fetches: 0,
+            recent_failure_count: 0,
+            recent_success_count: 10,
+            last_completed: Utc::now() - chrono::Duration::hours(1),
+        }
+    }
+
+    // -- compute_base_interval tests --
+
+    #[test]
+    fn test_cold_start_returns_floor() {
+        let mut stats = make_stats("CS");
+        stats.recent_runs = 0;
+        assert_eq!(compute_base_interval(&stats), COLD_START_INTERVAL);
+    }
+
+    #[test]
+    fn test_high_change_rate() {
+        let mut stats = make_stats("CS");
+        stats.avg_change_ratio = 0.15;
+        assert_eq!(compute_base_interval(&stats), FLOOR_INTERVAL);
+    }
+
+    #[test]
+    fn test_moderate_high_change() {
+        let mut stats = make_stats("CS");
+        stats.avg_change_ratio = 0.07;
+        assert_eq!(compute_base_interval(&stats), MODERATE_HIGH_INTERVAL);
+    }
+
+    #[test]
+    fn test_moderate_low_change() {
+        let mut stats = make_stats("CS");
+        stats.avg_change_ratio = 0.03;
+        assert_eq!(compute_base_interval(&stats), MODERATE_LOW_INTERVAL);
+    }
+
+    #[test]
+    fn test_low_change() {
+        let mut stats = make_stats("CS");
+        stats.avg_change_ratio = 0.005;
+        assert_eq!(compute_base_interval(&stats), LOW_CHANGE_INTERVAL);
+    }
+
+    #[test]
+    fn test_zero_5_consecutive() {
+        let mut stats = make_stats("CS");
+        stats.avg_change_ratio = 0.0;
+        stats.consecutive_zero_changes = 5;
+        assert_eq!(compute_base_interval(&stats), ZERO_5_INTERVAL);
+    }
+
+    #[test]
+    fn test_zero_10_consecutive() {
+        let mut stats = make_stats("CS");
+        stats.avg_change_ratio = 0.0;
+        stats.consecutive_zero_changes = 10;
+        assert_eq!(compute_base_interval(&stats), ZERO_10_INTERVAL);
+    }
+
+    #[test]
+    fn test_zero_20_consecutive() {
+        let mut stats = make_stats("CS");
+        stats.avg_change_ratio = 0.0;
+        stats.consecutive_zero_changes = 20;
+        assert_eq!(compute_base_interval(&stats), CEILING_INTERVAL);
+    }
+
+    // -- evaluate_subject tests --
+
+    #[test]
+    fn test_pause_empty_fetches() {
+        let mut stats = make_stats("CS");
+        stats.consecutive_empty_fetches = 3;
+        stats.last_completed = Utc::now() - chrono::Duration::minutes(10);
+        let result = evaluate_subject(&stats, Utc::now(), false);
+        assert_eq!(result, SubjectSchedule::Paused);
+    }
+
+    #[test]
+    fn test_pause_all_failures() {
+        let mut stats = make_stats("CS");
+        stats.recent_success_count = 0;
+        stats.recent_failure_count = 5;
+        stats.last_completed = Utc::now() - chrono::Duration::minutes(10);
+        let result = evaluate_subject(&stats, Utc::now(), false);
+        assert_eq!(result, SubjectSchedule::Paused);
+    }
+
+    #[test]
+    fn test_probe_after_pause() {
+        let mut stats = make_stats("CS");
+        stats.consecutive_empty_fetches = 5;
+        stats.last_completed = Utc::now() - chrono::Duration::hours(7);
+        let result = evaluate_subject(&stats, Utc::now(), false);
+        assert_eq!(result, SubjectSchedule::Eligible(PAUSE_PROBE_INTERVAL));
+    }
+
+    #[test]
+    fn test_read_only_past_term() {
+        let stats = make_stats("CS");
+        let result = evaluate_subject(&stats, Utc::now(), true);
+        assert_eq!(result, SubjectSchedule::ReadOnly);
+    }
+
+    #[test]
+    fn test_cooldown_not_elapsed() {
+        let mut stats = make_stats("CS");
+        stats.avg_change_ratio = 0.15; // floor = 3 min
+        stats.last_completed = Utc::now() - chrono::Duration::seconds(30);
+        // Use a peak-hours timestamp so multiplier = 1
+        let peak = Utc.with_ymd_and_hms(2025, 7, 14, 15, 0, 0).unwrap(); // Mon 10am CT
+        stats.last_completed = peak - chrono::Duration::seconds(30);
+        let result = evaluate_subject(&stats, peak, false);
+        assert!(matches!(result, SubjectSchedule::Cooldown(_)));
+    }
+
+    #[test]
+    fn test_eligible_elapsed() {
+        let mut stats = make_stats("CS");
+        stats.avg_change_ratio = 0.15; // floor = 3 min
+        let peak = Utc.with_ymd_and_hms(2025, 7, 14, 15, 0, 0).unwrap(); // Mon 10am CT
+        stats.last_completed = peak - chrono::Duration::minutes(5);
+        let result = evaluate_subject(&stats, peak, false);
+        assert!(matches!(result, SubjectSchedule::Eligible(_)));
+    }
+
+    // -- time_of_day_multiplier tests --
+
+    #[test]
+    fn test_time_multiplier_peak() {
+        // Monday 10am CT = 15:00 UTC
+        let dt = Utc.with_ymd_and_hms(2025, 7, 14, 15, 0, 0).unwrap();
+        assert_eq!(time_of_day_multiplier(dt), 1);
+    }
+
+    #[test]
+    fn test_time_multiplier_offpeak() {
+        // Monday 8pm CT = 01:00 UTC next day, but let's use Tuesday 01:00 UTC = Mon 8pm CT
+        let dt = Utc.with_ymd_and_hms(2025, 7, 15, 1, 0, 0).unwrap();
+        assert_eq!(time_of_day_multiplier(dt), 2);
+    }
+
+    #[test]
+    fn test_time_multiplier_night() {
+        // 3am CT = 08:00 UTC
+        let dt = Utc.with_ymd_and_hms(2025, 7, 14, 8, 0, 0).unwrap();
+        assert_eq!(time_of_day_multiplier(dt), 4);
+    }
+
+    #[test]
+    fn test_time_multiplier_weekend() {
+        // Saturday noon CT = 17:00 UTC
+        let dt = Utc.with_ymd_and_hms(2025, 7, 12, 17, 0, 0).unwrap();
+        assert_eq!(time_of_day_multiplier(dt), 4);
+    }
+}
@@ -1,7 +1,7 @@
 pub mod subject;

 use crate::banner::BannerApi;
-use crate::data::models::TargetType;
+use crate::data::models::{TargetType, UpsertCounts};
 use crate::error::Result;
 use serde::{Deserialize, Serialize};
 use sqlx::PgPool;
@@ -32,8 +32,9 @@ pub trait Job: Send + Sync {
    #[allow(dead_code)]
    fn target_type(&self) -> TargetType;

-    /// Process the job with the given API client and database pool
-    async fn process(&self, banner_api: &BannerApi, db_pool: &PgPool) -> Result<()>;
+    /// Process the job with the given API client and database pool.
+    /// Returns upsert effectiveness counts on success.
+    async fn process(&self, banner_api: &BannerApi, db_pool: &PgPool) -> Result<UpsertCounts>;

    /// Get a human-readable description of the job
    fn description(&self) -> String;
@@ -1,7 +1,7 @@
 use super::Job;
 use crate::banner::{BannerApi, SearchQuery, Term};
 use crate::data::batch::batch_upsert_courses;
-use crate::data::models::TargetType;
+use crate::data::models::{TargetType, UpsertCounts};
 use crate::error::Result;
 use serde::{Deserialize, Serialize};
 use sqlx::PgPool;
@@ -26,7 +26,7 @@ impl Job for SubjectJob {
    }

    #[tracing::instrument(skip(self, banner_api, db_pool), fields(subject = %self.subject))]
-    async fn process(&self, banner_api: &BannerApi, db_pool: &PgPool) -> Result<()> {
+    async fn process(&self, banner_api: &BannerApi, db_pool: &PgPool) -> Result<UpsertCounts> {
        let subject_code = &self.subject;

        // Get the current term
@@ -37,17 +37,19 @@ impl Job for SubjectJob {
            .search(&term, &query, "subjectDescription", false)
            .await?;

-        if let Some(courses_from_api) = search_result.data {
+        let counts = if let Some(courses_from_api) = search_result.data {
            info!(
                subject = %subject_code,
                count = courses_from_api.len(),
                "Found courses"
            );
-            batch_upsert_courses(&courses_from_api, db_pool).await?;
-        }
+            batch_upsert_courses(&courses_from_api, db_pool).await?
+        } else {
+            UpsertCounts::default()
+        };

        debug!(subject = %subject_code, "Subject job completed");
-        Ok(())
+        Ok(counts)
    }

    fn description(&self) -> String {
@@ -1,13 +1,17 @@
+pub mod adaptive;
 pub mod jobs;
 pub mod scheduler;
 pub mod worker;

 use crate::banner::BannerApi;
+use crate::data::scrape_jobs;
 use crate::services::Service;
+use crate::state::ReferenceCache;
 use crate::status::{ServiceStatus, ServiceStatusRegistry};
+use crate::web::ws::ScrapeJobEvent;
 use sqlx::PgPool;
 use std::sync::Arc;
-use tokio::sync::broadcast;
+use tokio::sync::{RwLock, broadcast};
 use tokio::task::JoinHandle;
 use tracing::{info, warn};

@@ -21,7 +25,9 @@ use self::worker::Worker;
 pub struct ScraperService {
    db_pool: PgPool,
    banner_api: Arc<BannerApi>,
+    reference_cache: Arc<RwLock<ReferenceCache>>,
    service_statuses: ServiceStatusRegistry,
+    job_events_tx: broadcast::Sender<ScrapeJobEvent>,
    scheduler_handle: Option<JoinHandle<()>>,
    worker_handles: Vec<JoinHandle<()>>,
    shutdown_tx: Option<broadcast::Sender<()>>,
@@ -29,11 +35,19 @@ pub struct ScraperService {

 impl ScraperService {
    /// Creates a new `ScraperService`.
-    pub fn new(db_pool: PgPool, banner_api: Arc<BannerApi>, service_statuses: ServiceStatusRegistry) -> Self {
+    pub fn new(
+        db_pool: PgPool,
+        banner_api: Arc<BannerApi>,
+        reference_cache: Arc<RwLock<ReferenceCache>>,
+        service_statuses: ServiceStatusRegistry,
+        job_events_tx: broadcast::Sender<ScrapeJobEvent>,
+    ) -> Self {
        Self {
            db_pool,
            banner_api,
+            reference_cache,
            service_statuses,
+            job_events_tx,
            scheduler_handle: None,
            worker_handles: Vec::new(),
            shutdown_tx: None,
@@ -41,14 +55,29 @@ impl ScraperService {
    }

    /// Starts the scheduler and a pool of workers.
-    pub fn start(&mut self) {
+    ///
+    /// Force-unlocks any jobs left locked by a previous unclean shutdown before
+    /// spawning workers, so those jobs re-enter the queue immediately.
+    pub async fn start(&mut self) {
+        // Recover jobs left locked by a previous crash/unclean shutdown
+        match scrape_jobs::force_unlock_all(&self.db_pool).await {
+            Ok(0) => {}
+            Ok(count) => warn!(count, "Force-unlocked stale jobs from previous run"),
+            Err(e) => warn!(error = ?e, "Failed to force-unlock stale jobs"),
+        }
+
        info!("ScraperService starting");

        // Create shutdown channel
        let (shutdown_tx, _) = broadcast::channel(1);
        self.shutdown_tx = Some(shutdown_tx.clone());

-        let scheduler = Scheduler::new(self.db_pool.clone(), self.banner_api.clone());
+        let scheduler = Scheduler::new(
+            self.db_pool.clone(),
+            self.banner_api.clone(),
+            self.reference_cache.clone(),
+            self.job_events_tx.clone(),
+        );
        let shutdown_rx = shutdown_tx.subscribe();
        let scheduler_handle = tokio::spawn(async move {
            scheduler.run(shutdown_rx).await;
@@ -58,7 +87,12 @@ impl ScraperService {

        let worker_count = 4; // This could be configurable
        for i in 0..worker_count {
-            let worker = Worker::new(i, self.db_pool.clone(), self.banner_api.clone());
+            let worker = Worker::new(
+                i,
+                self.db_pool.clone(),
+                self.banner_api.clone(),
+                self.job_events_tx.clone(),
+            );
            let shutdown_rx = shutdown_tx.subscribe();
            let worker_handle = tokio::spawn(async move {
                worker.run(shutdown_rx).await;
@@ -80,13 +114,14 @@ impl Service for ScraperService {
    }

    async fn run(&mut self) -> Result<(), anyhow::Error> {
-        self.start();
+        self.start().await;
        std::future::pending::<()>().await;
        Ok(())
    }

    async fn shutdown(&mut self) -> Result<(), anyhow::Error> {
-        self.service_statuses.set("scraper", ServiceStatus::Disabled);
+        self.service_statuses
+            .set("scraper", ServiceStatus::Disabled);
        info!("Shutting down scraper service");

        // Send shutdown signal to all tasks
@@ -1,28 +1,49 @@
 use crate::banner::{BannerApi, Term};
-use crate::data::models::{ScrapePriority, TargetType};
+use crate::data::models::{ReferenceData, ScrapePriority, TargetType};
 use crate::data::scrape_jobs;
 use crate::error::Result;
+use crate::rmp::RmpClient;
+use crate::scraper::adaptive::{SubjectSchedule, SubjectStats, evaluate_subject};
 use crate::scraper::jobs::subject::SubjectJob;
+use crate::state::ReferenceCache;
+use crate::web::ws::{ScrapeJobDto, ScrapeJobEvent};
+use chrono::{DateTime, Utc};
 use serde_json::json;
 use sqlx::PgPool;
+use std::collections::HashMap;
 use std::sync::Arc;
-use std::time::Duration;
-use tokio::sync::broadcast;
+use std::time::{Duration, Instant};
+use tokio::sync::{RwLock, broadcast};
 use tokio::time;
 use tokio_util::sync::CancellationToken;
 use tracing::{debug, error, info, warn};

+/// How often reference data is re-scraped (6 hours).
+const REFERENCE_DATA_INTERVAL: Duration = Duration::from_secs(6 * 60 * 60);
+
+/// How often RMP data is synced (24 hours).
+const RMP_SYNC_INTERVAL: Duration = Duration::from_secs(24 * 60 * 60);
+
 /// Periodically analyzes data and enqueues prioritized scrape jobs.
 pub struct Scheduler {
    db_pool: PgPool,
    banner_api: Arc<BannerApi>,
+    reference_cache: Arc<RwLock<ReferenceCache>>,
+    job_events_tx: broadcast::Sender<ScrapeJobEvent>,
 }

 impl Scheduler {
-    pub fn new(db_pool: PgPool, banner_api: Arc<BannerApi>) -> Self {
+    pub fn new(
+        db_pool: PgPool,
+        banner_api: Arc<BannerApi>,
+        reference_cache: Arc<RwLock<ReferenceCache>>,
+        job_events_tx: broadcast::Sender<ScrapeJobEvent>,
+    ) -> Self {
        Self {
            db_pool,
            banner_api,
+            reference_cache,
+            job_events_tx,
        }
    }

@@ -41,33 +62,69 @@ impl Scheduler {
        let work_interval = Duration::from_secs(60);
        let mut next_run = time::Instant::now();
        let mut current_work: Option<(tokio::task::JoinHandle<()>, CancellationToken)> = None;
+        // Scrape reference data immediately on first cycle
+        let mut last_ref_scrape = Instant::now() - REFERENCE_DATA_INTERVAL;
+        // Sync RMP data immediately on first cycle
+        let mut last_rmp_sync = Instant::now() - RMP_SYNC_INTERVAL;

        loop {
            tokio::select! {
                _ = time::sleep_until(next_run) => {
                    let cancel_token = CancellationToken::new();

+                    let should_scrape_ref = last_ref_scrape.elapsed() >= REFERENCE_DATA_INTERVAL;
+                    let should_sync_rmp = last_rmp_sync.elapsed() >= RMP_SYNC_INTERVAL;
+
                    // Spawn work in separate task to allow graceful cancellation during shutdown.
-                    // Without this, shutdown would have to wait for the full scheduling cycle.
                    let work_handle = tokio::spawn({
                        let db_pool = self.db_pool.clone();
                        let banner_api = self.banner_api.clone();
                        let cancel_token = cancel_token.clone();
+                        let reference_cache = self.reference_cache.clone();
+                        let job_events_tx = self.job_events_tx.clone();

-                        async move {
-                            tokio::select! {
-                                result = Self::schedule_jobs_impl(&db_pool, &banner_api) => {
-                                    if let Err(e) = result {
-                                        error!(error = ?e, "Failed to schedule jobs");
+                                async move {
+                                    tokio::select! {
+                                        _ = async {
+                                            // RMP sync is independent of Banner API — run it
+                                            // concurrently with reference data scraping so it
+                                            // doesn't wait behind rate-limited Banner calls.
+                                            let rmp_fut = async {
+                                                if should_sync_rmp
+                                                    && let Err(e) = Self::sync_rmp_data(&db_pool).await
+                                                {
+                                                    error!(error = ?e, "Failed to sync RMP data");
+                                                }
+                                            };
+
+                                            let ref_fut = async {
+                                                if should_scrape_ref
+                                                    && let Err(e) = Self::scrape_reference_data(&db_pool, &banner_api, &reference_cache).await
+                                                {
+                                                    error!(error = ?e, "Failed to scrape reference data");
+                                                }
+                                            };
+
+                                            tokio::join!(rmp_fut, ref_fut);
+
+                                            if let Err(e) = Self::schedule_jobs_impl(&db_pool, &banner_api, Some(&job_events_tx)).await {
+                                                error!(error = ?e, "Failed to schedule jobs");
+                                            }
+                                        } => {}
+                                        _ = cancel_token.cancelled() => {
+                                            debug!("Scheduling work cancelled gracefully");
+                                        }
                                    }
                                }
-                                _ = cancel_token.cancelled() => {
-                                    debug!("Scheduling work cancelled gracefully");
-                                }
-                            }
-                        }
                    });

+                    if should_scrape_ref {
+                        last_ref_scrape = Instant::now();
+                    }
+                    if should_sync_rmp {
+                        last_rmp_sync = Instant::now();
+                    }
+
                    current_work = Some((work_handle, cancel_token));
                    next_run = time::Instant::now() + work_interval;
                }
@@ -94,18 +151,17 @@ impl Scheduler {

    /// Core scheduling logic that analyzes data and creates scrape jobs.
    ///
-    /// Strategy:
-    /// 1. Fetch all subjects for the current term from Banner API
-    /// 2. Query existing jobs in a single batch query
-    /// 3. Create jobs only for subjects that don't have pending jobs
+    /// Uses adaptive scheduling to determine per-subject scrape intervals based
+    /// on recent change rates, failure patterns, and time of day. Only subjects
+    /// that are eligible (i.e. their cooldown has elapsed) are enqueued.
    ///
    /// This is a static method (not &self) to allow it to be called from spawned tasks.
    #[tracing::instrument(skip_all, fields(term))]
-    async fn schedule_jobs_impl(db_pool: &PgPool, banner_api: &BannerApi) -> Result<()> {
-        // For now, we will implement a simple baseline scheduling strategy:
-        // 1. Get a list of all subjects from the Banner API.
-        // 2. Query existing jobs for all subjects in a single query.
-        // 3. Create new jobs only for subjects that don't have existing jobs.
+    async fn schedule_jobs_impl(
+        db_pool: &PgPool,
+        banner_api: &BannerApi,
+        job_events_tx: Option<&broadcast::Sender<ScrapeJobEvent>>,
+    ) -> Result<()> {
        let term = Term::get_current().inner().to_string();

        tracing::Span::current().record("term", term.as_str());
@@ -117,13 +173,70 @@ impl Scheduler {
            "Retrieved subjects from API"
        );

-        // Create payloads for all subjects
-        let subject_payloads: Vec<_> = subjects
-            .iter()
-            .map(|subject| json!({ "subject": subject.code }))
+        // Fetch per-subject stats and build a lookup map
+        let stats_rows = scrape_jobs::fetch_subject_stats(db_pool).await?;
+        let stats_map: HashMap<String, SubjectStats> = stats_rows
+            .into_iter()
+            .map(|row| {
+                let subject = row.subject.clone();
+                (subject, SubjectStats::from(row))
+            })
            .collect();

-        // Query existing jobs for all subjects in a single query
+        // Evaluate each subject using adaptive scheduling
+        let now = Utc::now();
+        let is_past_term = false; // Scheduler currently only fetches current term subjects
+        let mut eligible_subjects: Vec<String> = Vec::new();
+        let mut cooldown_count: usize = 0;
+        let mut paused_count: usize = 0;
+        let mut read_only_count: usize = 0;
+
+        for subject in &subjects {
+            let stats = stats_map.get(&subject.code).cloned().unwrap_or_else(|| {
+                // Cold start: no history for this subject
+                SubjectStats {
+                    subject: subject.code.clone(),
+                    recent_runs: 0,
+                    avg_change_ratio: 0.0,
+                    consecutive_zero_changes: 0,
+                    consecutive_empty_fetches: 0,
+                    recent_failure_count: 0,
+                    recent_success_count: 0,
+                    last_completed: DateTime::<Utc>::MIN_UTC,
+                }
+            });
+
+            match evaluate_subject(&stats, now, is_past_term) {
+                SubjectSchedule::Eligible(_) => {
+                    eligible_subjects.push(subject.code.clone());
+                }
+                SubjectSchedule::Cooldown(_) => cooldown_count += 1,
+                SubjectSchedule::Paused => paused_count += 1,
+                SubjectSchedule::ReadOnly => read_only_count += 1,
+            }
+        }
+
+        info!(
+            total = subjects.len(),
+            eligible = eligible_subjects.len(),
+            cooldown = cooldown_count,
+            paused = paused_count,
+            read_only = read_only_count,
+            "Adaptive scheduling decisions"
+        );
+
+        if eligible_subjects.is_empty() {
+            debug!("No eligible subjects to schedule");
+            return Ok(());
+        }
+
+        // Create payloads only for eligible subjects
+        let subject_payloads: Vec<_> = eligible_subjects
+            .iter()
+            .map(|code| json!({ "subject": code }))
+            .collect();
+
+        // Query existing jobs for eligible subjects only
        let existing_payloads = scrape_jobs::find_existing_job_payloads(
            TargetType::Subject,
            &subject_payloads,
@@ -131,12 +244,12 @@ impl Scheduler {
        )
        .await?;

-        // Filter out subjects that already have jobs and prepare new jobs
+        // Filter out subjects that already have pending jobs
        let mut skipped_count = 0;
-        let new_jobs: Vec<_> = subjects
+        let new_jobs: Vec<_> = eligible_subjects
            .into_iter()
-            .filter_map(|subject| {
-                let job = SubjectJob::new(subject.code.clone());
+            .filter_map(|subject_code| {
+                let job = SubjectJob::new(subject_code.clone());
                let payload = serde_json::to_value(&job).unwrap();
                let payload_str = payload.to_string();

@@ -144,7 +257,7 @@ impl Scheduler {
                    skipped_count += 1;
                    None
                } else {
-                    Some((payload, subject.code))
+                    Some((payload, subject_code))
                }
            })
            .collect();
@@ -164,10 +277,149 @@ impl Scheduler {
                .map(|(payload, _)| (payload, TargetType::Subject, ScrapePriority::Low))
                .collect();

-            scrape_jobs::batch_insert_jobs(&jobs, db_pool).await?;
+            let inserted = scrape_jobs::batch_insert_jobs(&jobs, db_pool).await?;
+
+            if let Some(tx) = job_events_tx {
+                inserted.iter().for_each(|job| {
+                    debug!(job_id = job.id, "Emitting JobCreated event");
+                    let _ = tx.send(ScrapeJobEvent::JobCreated {
+                        job: ScrapeJobDto::from(job),
+                    });
+                });
+            }
        }

        debug!("Job scheduling complete");
        Ok(())
    }
+
+    /// Fetch all RMP professors, upsert to DB, and auto-match against Banner instructors.
+    #[tracing::instrument(skip_all)]
+    async fn sync_rmp_data(db_pool: &PgPool) -> Result<()> {
+        info!("Starting RMP data sync");
+
+        let client = RmpClient::new();
+        let professors = client.fetch_all_professors().await?;
+        let total = professors.len();
+
+        crate::data::rmp::batch_upsert_rmp_professors(&professors, db_pool).await?;
+        info!(total, "RMP professors upserted");
+
+        let stats = crate::data::rmp_matching::generate_candidates(db_pool).await?;
+        info!(
+            total,
+            stats.total_unmatched,
+            stats.candidates_created,
+            stats.auto_matched,
+            stats.skipped_unparseable,
+            stats.skipped_no_candidates,
+            "RMP sync complete"
+        );
+
+        Ok(())
+    }
+
+    /// Scrape all reference data categories from Banner and upsert to DB, then refresh cache.
+    #[tracing::instrument(skip_all)]
+    async fn scrape_reference_data(
+        db_pool: &PgPool,
+        banner_api: &BannerApi,
+        reference_cache: &Arc<RwLock<ReferenceCache>>,
+    ) -> Result<()> {
+        let term = Term::get_current().inner().to_string();
+        info!(term = %term, "Scraping reference data");
+
+        let mut all_entries = Vec::new();
+
+        // Terms (fetched via session pool, no active session needed)
+        match banner_api.sessions.get_terms("", 1, 500).await {
+            Ok(terms) => {
+                debug!(count = terms.len(), "Fetched terms");
+                all_entries.extend(terms.into_iter().map(|t| ReferenceData {
+                    category: "term".to_string(),
+                    code: t.code,
+                    description: t.description,
+                }));
+            }
+            Err(e) => warn!(error = ?e, "Failed to fetch terms"),
+        }
+
+        // Subjects
+        match banner_api.get_subjects("", &term, 1, 500).await {
+            Ok(pairs) => {
+                debug!(count = pairs.len(), "Fetched subjects");
+                all_entries.extend(pairs.into_iter().map(|p| ReferenceData {
+                    category: "subject".to_string(),
+                    code: p.code,
+                    description: p.description,
+                }));
+            }
+            Err(e) => warn!(error = ?e, "Failed to fetch subjects"),
+        }
+
+        // Campuses
+        match banner_api.get_campuses(&term).await {
+            Ok(pairs) => {
+                debug!(count = pairs.len(), "Fetched campuses");
+                all_entries.extend(pairs.into_iter().map(|p| ReferenceData {
+                    category: "campus".to_string(),
+                    code: p.code,
+                    description: p.description,
+                }));
+            }
+            Err(e) => warn!(error = ?e, "Failed to fetch campuses"),
+        }
+
+        // Instructional methods
+        match banner_api.get_instructional_methods(&term).await {
+            Ok(pairs) => {
+                debug!(count = pairs.len(), "Fetched instructional methods");
+                all_entries.extend(pairs.into_iter().map(|p| ReferenceData {
+                    category: "instructional_method".to_string(),
+                    code: p.code,
+                    description: p.description,
+                }));
+            }
+            Err(e) => warn!(error = ?e, "Failed to fetch instructional methods"),
+        }
+
+        // Parts of term
+        match banner_api.get_parts_of_term(&term).await {
+            Ok(pairs) => {
+                debug!(count = pairs.len(), "Fetched parts of term");
+                all_entries.extend(pairs.into_iter().map(|p| ReferenceData {
+                    category: "part_of_term".to_string(),
+                    code: p.code,
+                    description: p.description,
+                }));
+            }
+            Err(e) => warn!(error = ?e, "Failed to fetch parts of term"),
+        }
+
+        // Attributes
+        match banner_api.get_attributes(&term).await {
+            Ok(pairs) => {
+                debug!(count = pairs.len(), "Fetched attributes");
+                all_entries.extend(pairs.into_iter().map(|p| ReferenceData {
+                    category: "attribute".to_string(),
+                    code: p.code,
+                    description: p.description,
+                }));
+            }
+            Err(e) => warn!(error = ?e, "Failed to fetch attributes"),
+        }
+
+        // Batch upsert all entries
+        let total = all_entries.len();
+        crate::data::reference::batch_upsert(&all_entries, db_pool).await?;
+        info!(total_entries = total, "Reference data upserted to DB");
+
+        // Refresh in-memory cache
+        let all = crate::data::reference::get_all(db_pool).await?;
+        let count = all.len();
+        *reference_cache.write().await = ReferenceCache::from_entries(all);
+        info!(entries = count, "Reference cache refreshed");
+
+        Ok(())
+    }
 }
@@ -1,8 +1,10 @@
 use crate::banner::{BannerApi, BannerApiError};
-use crate::data::models::ScrapeJob;
+use crate::data::models::{ScrapeJob, ScrapeJobStatus, UpsertCounts};
 use crate::data::scrape_jobs;
 use crate::error::Result;
 use crate::scraper::jobs::{JobError, JobType};
+use crate::web::ws::ScrapeJobEvent;
+use chrono::{DateTime, Utc};
 use sqlx::PgPool;
 use std::sync::Arc;
 use std::time::Duration;
@@ -10,6 +12,9 @@ use tokio::sync::broadcast;
 use tokio::time;
 use tracing::{Instrument, debug, error, info, trace, warn};

+/// Maximum time a single job is allowed to run before being considered stuck.
+const JOB_TIMEOUT: Duration = Duration::from_secs(5 * 60);
+
 /// A single worker instance.
 ///
 /// Each worker runs in its own asynchronous task and continuously polls the
@@ -18,14 +23,21 @@ pub struct Worker {
    id: usize, // For logging purposes
    db_pool: PgPool,
    banner_api: Arc<BannerApi>,
+    job_events_tx: broadcast::Sender<ScrapeJobEvent>,
 }

 impl Worker {
-    pub fn new(id: usize, db_pool: PgPool, banner_api: Arc<BannerApi>) -> Self {
+    pub fn new(
+        id: usize,
+        db_pool: PgPool,
+        banner_api: Arc<BannerApi>,
+        job_events_tx: broadcast::Sender<ScrapeJobEvent>,
+    ) -> Self {
        Self {
            id,
            db_pool,
            banner_api,
+            job_events_tx,
        }
    }

@@ -60,22 +72,57 @@ impl Worker {
            let job_id = job.id;
            let retry_count = job.retry_count;
            let max_retries = job.max_retries;
+            let target_type = job.target_type;
+            let payload = job.target_payload.clone();
+            let priority = job.priority;
+            let queued_at = job.queued_at;
+            let started_at = Utc::now();
            let start = std::time::Instant::now();

-            // Process the job, racing against shutdown signal
+            // Emit JobLocked event
+            let locked_at = started_at.to_rfc3339();
+            debug!(job_id, "Emitting JobLocked event");
+            let _ = self.job_events_tx.send(ScrapeJobEvent::JobLocked {
+                id: job_id,
+                locked_at,
+                status: ScrapeJobStatus::Processing,
+            });
+
+            // Process the job, racing against shutdown signal and timeout
            let process_result = tokio::select! {
                _ = shutdown_rx.recv() => {
                    self.handle_shutdown_during_processing(job_id).await;
                    break;
                }
-                result = self.process_job(job) => result
+                result = async {
+                    match time::timeout(JOB_TIMEOUT, self.process_job(job)).await {
+                        Ok(result) => result,
+                        Err(_elapsed) => {
+                            Err(JobError::Recoverable(anyhow::anyhow!(
+                                "job timed out after {}s",
+                                JOB_TIMEOUT.as_secs()
+                            )))
+                        }
+                    }
+                } => result
            };

            let duration = start.elapsed();

            // Handle the job processing result
-            self.handle_job_result(job_id, retry_count, max_retries, process_result, duration)
-                .await;
+            self.handle_job_result(
+                job_id,
+                retry_count,
+                max_retries,
+                process_result,
+                duration,
+                target_type,
+                payload,
+                priority,
+                queued_at,
+                started_at,
+            )
+            .await;
        }
    }

@@ -87,7 +134,7 @@ impl Worker {
        scrape_jobs::fetch_and_lock_job(&self.db_pool).await
    }

-    async fn process_job(&self, job: ScrapeJob) -> Result<(), JobError> {
+    async fn process_job(&self, job: ScrapeJob) -> Result<UpsertCounts, JobError> {
        // Convert the database job to our job type
        let job_type = JobType::from_target_type_and_payload(job.target_type, job.target_payload)
            .map_err(|e| JobError::Unrecoverable(anyhow::anyhow!(e)))?; // Parse errors are unrecoverable
@@ -114,9 +161,7 @@ impl Worker {
            job_impl
                .process(&self.banner_api, &self.db_pool)
                .await
-                .map_err(JobError::Recoverable)?;
-
-            Ok(())
+                .map_err(JobError::Recoverable)
        }
        .instrument(span)
        .await
@@ -130,7 +175,11 @@ impl Worker {
        scrape_jobs::unlock_job(job_id, &self.db_pool).await
    }

-    async fn unlock_and_increment_retry(&self, job_id: i32, max_retries: i32) -> Result<bool> {
+    async fn unlock_and_increment_retry(
+        &self,
+        job_id: i32,
+        max_retries: i32,
+    ) -> Result<Option<chrono::DateTime<chrono::Utc>>> {
        scrape_jobs::unlock_and_increment_retry(job_id, max_retries, &self.db_pool).await
    }

@@ -156,31 +205,97 @@ impl Worker {
    }

    /// Handle the result of job processing
+    #[allow(clippy::too_many_arguments)]
    async fn handle_job_result(
        &self,
        job_id: i32,
        retry_count: i32,
        max_retries: i32,
-        result: Result<(), JobError>,
+        result: Result<UpsertCounts, JobError>,
        duration: std::time::Duration,
+        target_type: crate::data::models::TargetType,
+        payload: serde_json::Value,
+        priority: crate::data::models::ScrapePriority,
+        queued_at: DateTime<Utc>,
+        started_at: DateTime<Utc>,
    ) {
+        let duration_ms = duration.as_millis() as i32;
+
        match result {
-            Ok(()) => {
+            Ok(counts) => {
                debug!(
                    worker_id = self.id,
                    job_id,
                    duration_ms = duration.as_millis(),
+                    courses_fetched = counts.courses_fetched,
+                    courses_changed = counts.courses_changed,
+                    courses_unchanged = counts.courses_unchanged,
                    "Job completed successfully"
                );
+
+                // Log the result
+                if let Err(e) = scrape_jobs::insert_job_result(
+                    target_type,
+                    payload,
+                    priority,
+                    queued_at,
+                    started_at,
+                    duration_ms,
+                    true,
+                    None,
+                    retry_count,
+                    Some(&counts),
+                    &self.db_pool,
+                )
+                .await
+                {
+                    error!(worker_id = self.id, job_id, error = ?e, "Failed to insert job result");
+                }
+
                if let Err(e) = self.delete_job(job_id).await {
                    error!(worker_id = self.id, job_id, error = ?e, "Failed to delete completed job");
                }
+                debug!(job_id, "Emitting JobCompleted event");
+                let _ = self
+                    .job_events_tx
+                    .send(ScrapeJobEvent::JobCompleted { id: job_id });
            }
            Err(JobError::Recoverable(e)) => {
-                self.handle_recoverable_error(job_id, retry_count, max_retries, e, duration)
-                    .await;
+                self.handle_recoverable_error(
+                    job_id,
+                    retry_count,
+                    max_retries,
+                    e,
+                    duration,
+                    target_type,
+                    payload,
+                    priority,
+                    queued_at,
+                    started_at,
+                )
+                .await;
            }
            Err(JobError::Unrecoverable(e)) => {
+                // Log the failed result
+                let err_msg = format!("{e:#}");
+                if let Err(log_err) = scrape_jobs::insert_job_result(
+                    target_type,
+                    payload,
+                    priority,
+                    queued_at,
+                    started_at,
+                    duration_ms,
+                    false,
+                    Some(&err_msg),
+                    retry_count,
+                    None,
+                    &self.db_pool,
+                )
+                .await
+                {
+                    error!(worker_id = self.id, job_id, error = ?log_err, "Failed to insert job result");
+                }
+
                error!(
                    worker_id = self.id,
                    job_id,
@@ -191,11 +306,16 @@ impl Worker {
                if let Err(e) = self.delete_job(job_id).await {
                    error!(worker_id = self.id, job_id, error = ?e, "Failed to delete corrupted job");
                }
+                debug!(job_id, "Emitting JobDeleted event");
+                let _ = self
+                    .job_events_tx
+                    .send(ScrapeJobEvent::JobDeleted { id: job_id });
            }
        }
    }

    /// Handle recoverable errors by logging appropriately and unlocking the job
+    #[allow(clippy::too_many_arguments)]
    async fn handle_recoverable_error(
        &self,
        job_id: i32,
@@ -203,6 +323,11 @@ impl Worker {
        max_retries: i32,
        e: anyhow::Error,
        duration: std::time::Duration,
+        target_type: crate::data::models::TargetType,
+        payload: serde_json::Value,
+        priority: crate::data::models::ScrapePriority,
+        queued_at: DateTime<Utc>,
+        started_at: DateTime<Utc>,
    ) {
        let next_attempt = retry_count.saturating_add(1);
        let remaining_retries = max_retries.saturating_sub(next_attempt);
@@ -233,7 +358,7 @@ impl Worker {

        // Atomically unlock and increment retry count, checking if retry is allowed
        match self.unlock_and_increment_retry(job_id, max_retries).await {
-            Ok(can_retry) if can_retry => {
+            Ok(Some(new_queued_at)) => {
                debug!(
                    worker_id = self.id,
                    job_id,
@@ -241,9 +366,37 @@ impl Worker {
                    remaining_retries = remaining_retries,
                    "Job unlocked for retry"
                );
+                debug!(job_id, "Emitting JobRetried event");
+                let _ = self.job_events_tx.send(ScrapeJobEvent::JobRetried {
+                    id: job_id,
+                    retry_count: next_attempt,
+                    queued_at: new_queued_at.to_rfc3339(),
+                    status: ScrapeJobStatus::Pending,
+                });
+                // Don't log a result yet — the job will be retried
            }
-            Ok(_) => {
-                // Max retries exceeded (detected atomically)
+            Ok(None) => {
+                // Max retries exceeded — log final failure result
+                let duration_ms = duration.as_millis() as i32;
+                let err_msg = format!("{e:#}");
+                if let Err(log_err) = scrape_jobs::insert_job_result(
+                    target_type,
+                    payload,
+                    priority,
+                    queued_at,
+                    started_at,
+                    duration_ms,
+                    false,
+                    Some(&err_msg),
+                    next_attempt,
+                    None,
+                    &self.db_pool,
+                )
+                .await
+                {
+                    error!(worker_id = self.id, job_id, error = ?log_err, "Failed to insert job result");
+                }
+
                error!(
                    worker_id = self.id,
                    job_id,
@@ -256,6 +409,13 @@ impl Worker {
                if let Err(e) = self.delete_job(job_id).await {
                    error!(worker_id = self.id, job_id, error = ?e, "Failed to delete failed job");
                }
+                debug!(job_id, "Emitting JobExhausted and JobDeleted events");
+                let _ = self
+                    .job_events_tx
+                    .send(ScrapeJobEvent::JobExhausted { id: job_id });
+                let _ = self
+                    .job_events_tx
+                    .send(ScrapeJobEvent::JobDeleted { id: job_id });
            }
            Err(e) => {
                error!(worker_id = self.id, job_id, error = ?e, "Failed to unlock and increment retry count");
@@ -1,6 +1,7 @@
 use super::Service;
 use crate::state::AppState;
 use crate::status::ServiceStatus;
+use crate::web::auth::AuthConfig;
 use crate::web::create_router;
 use std::net::SocketAddr;
 use tokio::net::TcpListener;
@@ -11,22 +12,21 @@ use tracing::{info, trace, warn};
 pub struct WebService {
    port: u16,
    app_state: AppState,
+    auth_config: AuthConfig,
    shutdown_tx: Option<broadcast::Sender<()>>,
 }

 impl WebService {
-    pub fn new(port: u16, app_state: AppState) -> Self {
+    pub fn new(port: u16, app_state: AppState, auth_config: AuthConfig) -> Self {
        Self {
            port,
            app_state,
+            auth_config,
            shutdown_tx: None,
        }
    }
    /// Periodically pings the database and updates the "database" service status.
-    async fn db_health_check_loop(
-        state: AppState,
-        mut shutdown_rx: broadcast::Receiver<()>,
-    ) {
+    async fn db_health_check_loop(state: AppState, mut shutdown_rx: broadcast::Receiver<()>) {
        use std::time::Duration;
        let mut interval = tokio::time::interval(Duration::from_secs(30));

@@ -61,12 +61,14 @@ impl Service for WebService {

    async fn run(&mut self) -> Result<(), anyhow::Error> {
        // Create the main router with Banner API routes
-        let app = create_router(self.app_state.clone());
+        let app = create_router(self.app_state.clone(), self.auth_config.clone());

        let addr = SocketAddr::from(([0, 0, 0, 0], self.port));

        let listener = TcpListener::bind(addr).await?;
-        self.app_state.service_statuses.set("web", ServiceStatus::Active);
+        self.app_state
+            .service_statuses
+            .set("web", ServiceStatus::Active);
        info!(
            service = "web",
            address = %addr,
@@ -2,27 +2,116 @@

 use crate::banner::BannerApi;
 use crate::banner::Course;
+use crate::data::models::ReferenceData;
 use crate::status::ServiceStatusRegistry;
+use crate::web::schedule_cache::ScheduleCache;
+use crate::web::session_cache::{OAuthStateStore, SessionCache};
+use crate::web::ws::ScrapeJobEvent;
 use anyhow::Result;
 use sqlx::PgPool;
+use std::collections::HashMap;
 use std::sync::Arc;
+use tokio::sync::{RwLock, broadcast};
+
+/// In-memory cache for reference data (code→description lookups).
+///
+/// Loaded from the `reference_data` table on startup and refreshed periodically.
+/// Uses a two-level HashMap so lookups take `&str` without allocating.
+pub struct ReferenceCache {
+    /// category → (code → description)
+    data: HashMap<String, HashMap<String, String>>,
+}
+
+impl Default for ReferenceCache {
+    fn default() -> Self {
+        Self::new()
+    }
+}
+
+impl ReferenceCache {
+    /// Create an empty cache.
+    pub fn new() -> Self {
+        Self {
+            data: HashMap::new(),
+        }
+    }
+
+    /// Build cache from a list of reference data entries.
+    pub fn from_entries(entries: Vec<ReferenceData>) -> Self {
+        let mut data: HashMap<String, HashMap<String, String>> = HashMap::new();
+        for e in entries {
+            data.entry(e.category)
+                .or_default()
+                .insert(e.code, e.description);
+        }
+        Self { data }
+    }
+
+    /// Look up a description by category and code. Zero allocations.
+    pub fn lookup(&self, category: &str, code: &str) -> Option<&str> {
+        self.data
+            .get(category)
+            .and_then(|codes| codes.get(code))
+            .map(|s| s.as_str())
+    }
+
+    /// Get all `(code, description)` pairs for a category, sorted by description.
+    pub fn entries_for_category(&self, category: &str) -> Vec<(&str, &str)> {
+        let Some(codes) = self.data.get(category) else {
+            return Vec::new();
+        };
+        let mut entries: Vec<(&str, &str)> = codes
+            .iter()
+            .map(|(code, desc)| (code.as_str(), desc.as_str()))
+            .collect();
+        entries.sort_by(|a, b| a.1.cmp(b.1));
+        entries
+    }
+}

 #[derive(Clone)]
 pub struct AppState {
    pub banner_api: Arc<BannerApi>,
    pub db_pool: PgPool,
    pub service_statuses: ServiceStatusRegistry,
+    pub reference_cache: Arc<RwLock<ReferenceCache>>,
+    pub session_cache: SessionCache,
+    pub oauth_state_store: OAuthStateStore,
+    pub schedule_cache: ScheduleCache,
+    pub scrape_job_tx: broadcast::Sender<ScrapeJobEvent>,
 }

 impl AppState {
    pub fn new(banner_api: Arc<BannerApi>, db_pool: PgPool) -> Self {
+        let (scrape_job_tx, _) = broadcast::channel(64);
+        let schedule_cache = ScheduleCache::new(db_pool.clone());
        Self {
+            session_cache: SessionCache::new(db_pool.clone()),
+            oauth_state_store: OAuthStateStore::new(),
            banner_api,
            db_pool,
            service_statuses: ServiceStatusRegistry::new(),
+            reference_cache: Arc::new(RwLock::new(ReferenceCache::new())),
+            schedule_cache,
+            scrape_job_tx,
        }
    }

+    /// Subscribe to scrape job lifecycle events.
+    pub fn scrape_job_events(&self) -> broadcast::Receiver<ScrapeJobEvent> {
+        self.scrape_job_tx.subscribe()
+    }
+
+    /// Initialize the reference cache from the database.
+    pub async fn load_reference_cache(&self) -> Result<()> {
+        let entries = crate::data::reference::get_all(&self.db_pool).await?;
+        let count = entries.len();
+        let cache = ReferenceCache::from_entries(entries);
+        *self.reference_cache.write().await = cache;
+        tracing::info!(entries = count, "Reference cache loaded");
+        Ok(())
+    }
+
    /// Get a course by CRN directly from Banner API
    pub async fn get_course_or_fetch(&self, term: &str, crn: &str) -> Result<Course> {
        self.banner_api
@@ -3,11 +3,14 @@ use std::time::Instant;

 use dashmap::DashMap;
 use serde::Serialize;
+use ts_rs::TS;

 /// Health status of a service.
-#[derive(Debug, Clone, Serialize, PartialEq)]
+#[derive(Debug, Clone, Serialize, PartialEq, TS)]
 #[serde(rename_all = "lowercase")]
+#[ts(export)]
 pub enum ServiceStatus {
+    #[allow(dead_code)]
    Starting,
    Active,
    Connected,
@@ -19,6 +22,7 @@ pub enum ServiceStatus {
 #[derive(Debug, Clone)]
 pub struct StatusEntry {
    pub status: ServiceStatus,
+    #[allow(dead_code)]
    pub updated_at: Instant,
 }

@@ -46,6 +50,7 @@ impl ServiceStatusRegistry {
    }

    /// Returns the current status of a named service, if present.
+    #[allow(dead_code)]
    pub fn get(&self, name: &str) -> Option<ServiceStatus> {
        self.inner.get(name).map(|entry| entry.status.clone())
    }
@@ -0,0 +1,269 @@
+//! Admin API handlers.
+//!
+//! All endpoints require the `AdminUser` extractor, returning 401/403 as needed.
+
+use axum::extract::{Path, State};
+use axum::http::{HeaderMap, StatusCode, header};
+use axum::response::{IntoResponse, Json, Response};
+use chrono::{DateTime, Utc};
+use serde::Deserialize;
+use serde_json::{Value, json};
+
+use crate::data::models::User;
+use crate::state::AppState;
+use crate::web::extractors::AdminUser;
+
+/// `GET /api/admin/status` — Enhanced system status for admins.
+pub async fn admin_status(
+    AdminUser(_user): AdminUser,
+    State(state): State<AppState>,
+) -> Result<Json<Value>, (StatusCode, Json<Value>)> {
+    let (user_count,): (i64,) = sqlx::query_as("SELECT COUNT(*) FROM users")
+        .fetch_one(&state.db_pool)
+        .await
+        .map_err(|e| {
+            tracing::error!(error = %e, "failed to count users");
+            (
+                StatusCode::INTERNAL_SERVER_ERROR,
+                Json(json!({"error": "failed to count users"})),
+            )
+        })?;
+
+    let (session_count,): (i64,) =
+        sqlx::query_as("SELECT COUNT(*) FROM user_sessions WHERE expires_at > now()")
+            .fetch_one(&state.db_pool)
+            .await
+            .map_err(|e| {
+                tracing::error!(error = %e, "failed to count sessions");
+                (
+                    StatusCode::INTERNAL_SERVER_ERROR,
+                    Json(json!({"error": "failed to count sessions"})),
+                )
+            })?;
+
+    let course_count = state.get_course_count().await.map_err(|e| {
+        tracing::error!(error = %e, "failed to count courses");
+        (
+            StatusCode::INTERNAL_SERVER_ERROR,
+            Json(json!({"error": "failed to count courses"})),
+        )
+    })?;
+
+    let (scrape_job_count,): (i64,) = sqlx::query_as("SELECT COUNT(*) FROM scrape_jobs")
+        .fetch_one(&state.db_pool)
+        .await
+        .map_err(|e| {
+            tracing::error!(error = %e, "failed to count scrape jobs");
+            (
+                StatusCode::INTERNAL_SERVER_ERROR,
+                Json(json!({"error": "failed to count scrape jobs"})),
+            )
+        })?;
+
+    let services: Vec<Value> = state
+        .service_statuses
+        .all()
+        .into_iter()
+        .map(|(name, status)| {
+            json!({
+                "name": name,
+                "status": status,
+            })
+        })
+        .collect();
+
+    Ok(Json(json!({
+        "userCount": user_count,
+        "sessionCount": session_count,
+        "courseCount": course_count,
+        "scrapeJobCount": scrape_job_count,
+        "services": services,
+    })))
+}
+
+/// `GET /api/admin/users` — List all users.
+pub async fn list_users(
+    AdminUser(_user): AdminUser,
+    State(state): State<AppState>,
+) -> Result<Json<Vec<User>>, (StatusCode, Json<Value>)> {
+    let users = crate::data::users::list_users(&state.db_pool)
+        .await
+        .map_err(|e| {
+            tracing::error!(error = %e, "failed to list users");
+            (
+                StatusCode::INTERNAL_SERVER_ERROR,
+                Json(json!({"error": "failed to list users"})),
+            )
+        })?;
+
+    Ok(Json(users))
+}
+
+#[derive(Deserialize)]
+pub struct SetAdminBody {
+    is_admin: bool,
+}
+
+/// `PUT /api/admin/users/{discord_id}/admin` — Set admin status for a user.
+pub async fn set_user_admin(
+    AdminUser(_user): AdminUser,
+    State(state): State<AppState>,
+    Path(discord_id): Path<i64>,
+    Json(body): Json<SetAdminBody>,
+) -> Result<Json<User>, (StatusCode, Json<Value>)> {
+    let user = crate::data::users::set_admin(&state.db_pool, discord_id, body.is_admin)
+        .await
+        .map_err(|e| {
+            tracing::error!(error = %e, "failed to set admin status");
+            (
+                StatusCode::INTERNAL_SERVER_ERROR,
+                Json(json!({"error": "failed to set admin status"})),
+            )
+        })?
+        .ok_or_else(|| {
+            (
+                StatusCode::NOT_FOUND,
+                Json(json!({"error": "user not found"})),
+            )
+        })?;
+
+    state.session_cache.evict_user(discord_id);
+
+    Ok(Json(user))
+}
+
+/// `GET /api/admin/scrape-jobs` — List scrape jobs.
+pub async fn list_scrape_jobs(
+    AdminUser(_user): AdminUser,
+    State(state): State<AppState>,
+) -> Result<Json<Value>, (StatusCode, Json<Value>)> {
+    let rows = sqlx::query_as::<_, crate::data::models::ScrapeJob>(
+        "SELECT * FROM scrape_jobs ORDER BY priority DESC, execute_at ASC LIMIT 100",
+    )
+    .fetch_all(&state.db_pool)
+    .await
+    .map_err(|e| {
+        tracing::error!(error = %e, "failed to list scrape jobs");
+        (
+            StatusCode::INTERNAL_SERVER_ERROR,
+            Json(json!({"error": "failed to list scrape jobs"})),
+        )
+    })?;
+
+    let jobs: Vec<Value> = rows
+        .iter()
+        .map(|j| {
+            json!({
+                "id": j.id,
+                "targetType": format!("{:?}", j.target_type),
+                "targetPayload": j.target_payload,
+                "priority": format!("{:?}", j.priority),
+                "executeAt": j.execute_at.to_rfc3339(),
+                "createdAt": j.created_at.to_rfc3339(),
+                "lockedAt": j.locked_at.map(|t| t.to_rfc3339()),
+                "retryCount": j.retry_count,
+                "maxRetries": j.max_retries,
+                "queuedAt": j.queued_at.to_rfc3339(),
+                "status": j.status(),
+            })
+        })
+        .collect();
+
+    Ok(Json(json!({ "jobs": jobs })))
+}
+
+/// Row returned by the audit-log query (audit + joined course fields).
+#[derive(sqlx::FromRow, Debug)]
+struct AuditRow {
+    id: i32,
+    course_id: i32,
+    timestamp: chrono::DateTime<chrono::Utc>,
+    field_changed: String,
+    old_value: String,
+    new_value: String,
+    // Joined from courses table (nullable in case the course was deleted)
+    subject: Option<String>,
+    course_number: Option<String>,
+    crn: Option<String>,
+    title: Option<String>,
+}
+
+/// Format a `DateTime<Utc>` as an HTTP-date (RFC 2822) for Last-Modified headers.
+fn to_http_date(dt: &DateTime<Utc>) -> String {
+    dt.format("%a, %d %b %Y %H:%M:%S GMT").to_string()
+}
+
+/// Parse an `If-Modified-Since` header value into a `DateTime<Utc>`.
+fn parse_if_modified_since(headers: &HeaderMap) -> Option<DateTime<Utc>> {
+    let val = headers.get(header::IF_MODIFIED_SINCE)?.to_str().ok()?;
+    DateTime::parse_from_rfc2822(val)
+        .ok()
+        .map(|dt| dt.with_timezone(&Utc))
+}
+
+/// `GET /api/admin/audit-log` — List recent audit entries.
+///
+/// Supports `If-Modified-Since`: returns 304 when the newest entry hasn't changed.
+pub async fn list_audit_log(
+    AdminUser(_user): AdminUser,
+    headers: HeaderMap,
+    State(state): State<AppState>,
+) -> Result<Response, (StatusCode, Json<Value>)> {
+    let rows = sqlx::query_as::<_, AuditRow>(
+        "SELECT a.id, a.course_id, a.timestamp, a.field_changed, a.old_value, a.new_value, \
+                c.subject, c.course_number, c.crn, c.title \
+         FROM course_audits a \
+         LEFT JOIN courses c ON c.id = a.course_id \
+         ORDER BY a.timestamp DESC LIMIT 200",
+    )
+    .fetch_all(&state.db_pool)
+    .await
+    .map_err(|e| {
+        tracing::error!(error = %e, "failed to list audit log");
+        (
+            StatusCode::INTERNAL_SERVER_ERROR,
+            Json(json!({"error": "failed to list audit log"})),
+        )
+    })?;
+
+    // Determine the latest timestamp across all rows (query is DESC so first row is newest)
+    let latest = rows.first().map(|r| r.timestamp);
+
+    // If the client sent If-Modified-Since and our data hasn't changed, return 304
+    if let (Some(since), Some(latest_ts)) = (parse_if_modified_since(&headers), latest) {
+        // Truncate to seconds for comparison (HTTP dates have second precision)
+        if latest_ts.timestamp() <= since.timestamp() {
+            let mut resp = StatusCode::NOT_MODIFIED.into_response();
+            if let Ok(val) = to_http_date(&latest_ts).parse() {
+                resp.headers_mut().insert(header::LAST_MODIFIED, val);
+            }
+            return Ok(resp);
+        }
+    }
+
+    let entries: Vec<Value> = rows
+        .iter()
+        .map(|a| {
+            json!({
+                "id": a.id,
+                "courseId": a.course_id,
+                "timestamp": a.timestamp.to_rfc3339(),
+                "fieldChanged": a.field_changed,
+                "oldValue": a.old_value,
+                "newValue": a.new_value,
+                "subject": a.subject,
+                "courseNumber": a.course_number,
+                "crn": a.crn,
+                "courseTitle": a.title,
+            })
+        })
+        .collect();
+
+    let mut resp = Json(json!({ "entries": entries })).into_response();
+    if let Some(latest_ts) = latest
+        && let Ok(val) = to_http_date(&latest_ts).parse()
+    {
+        resp.headers_mut().insert(header::LAST_MODIFIED, val);
+    }
+    Ok(resp)
+}
@@ -0,0 +1,865 @@
+//! Admin API handlers for RMP instructor matching management.
+
+use axum::extract::{Path, Query, State};
+use axum::http::StatusCode;
+use axum::response::Json;
+use serde::{Deserialize, Serialize};
+use serde_json::{Value, json};
+use ts_rs::TS;
+
+use crate::state::AppState;
+use crate::web::extractors::AdminUser;
+
+// ---------------------------------------------------------------------------
+// Query / body types
+// ---------------------------------------------------------------------------
+
+#[derive(Deserialize)]
+pub struct ListInstructorsParams {
+    status: Option<String>,
+    search: Option<String>,
+    page: Option<i32>,
+    per_page: Option<i32>,
+    sort: Option<String>,
+}
+
+#[derive(Deserialize)]
+#[serde(rename_all = "camelCase")]
+pub struct MatchBody {
+    rmp_legacy_id: i32,
+}
+
+#[derive(Deserialize)]
+#[serde(rename_all = "camelCase")]
+pub struct RejectCandidateBody {
+    rmp_legacy_id: i32,
+}
+
+// ---------------------------------------------------------------------------
+// Response types
+// ---------------------------------------------------------------------------
+
+/// Simple acknowledgement response for mutating operations.
+#[derive(Debug, Clone, Serialize, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export)]
+pub struct OkResponse {
+    pub ok: bool,
+}
+
+/// A top-candidate summary shown in the instructor list view.
+#[derive(Debug, Clone, Serialize, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export)]
+pub struct TopCandidateResponse {
+    pub rmp_legacy_id: i32,
+    pub score: Option<f32>,
+    #[ts(as = "Option<std::collections::HashMap<String, f32>>")]
+    pub score_breakdown: Option<serde_json::Value>,
+    pub first_name: Option<String>,
+    pub last_name: Option<String>,
+    pub department: Option<String>,
+    pub avg_rating: Option<f32>,
+    pub num_ratings: Option<i32>,
+}
+
+/// An instructor row in the paginated list.
+#[derive(Debug, Clone, Serialize, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export)]
+pub struct InstructorListItem {
+    pub id: i32,
+    pub display_name: String,
+    pub email: String,
+    pub rmp_match_status: String,
+    #[ts(as = "i32")]
+    pub rmp_link_count: i64,
+    #[ts(as = "i32")]
+    pub candidate_count: i64,
+    #[ts(as = "i32")]
+    pub course_subject_count: i64,
+    pub top_candidate: Option<TopCandidateResponse>,
+}
+
+/// Aggregate status counts for the instructor list.
+#[derive(Debug, Clone, Serialize, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export)]
+pub struct InstructorStats {
+    #[ts(as = "i32")]
+    pub total: i64,
+    #[ts(as = "i32")]
+    pub unmatched: i64,
+    #[ts(as = "i32")]
+    pub auto: i64,
+    #[ts(as = "i32")]
+    pub confirmed: i64,
+    #[ts(as = "i32")]
+    pub rejected: i64,
+    #[ts(as = "i32")]
+    pub with_candidates: i64,
+}
+
+/// Response for `GET /api/admin/instructors`.
+#[derive(Debug, Clone, Serialize, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export)]
+pub struct ListInstructorsResponse {
+    pub instructors: Vec<InstructorListItem>,
+    #[ts(as = "i32")]
+    pub total: i64,
+    pub page: i32,
+    pub per_page: i32,
+    pub stats: InstructorStats,
+}
+
+/// Instructor summary in the detail view.
+#[derive(Debug, Clone, Serialize, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export)]
+pub struct InstructorDetail {
+    pub id: i32,
+    pub display_name: String,
+    pub email: String,
+    pub rmp_match_status: String,
+    pub subjects_taught: Vec<String>,
+    #[ts(as = "i32")]
+    pub course_count: i64,
+}
+
+/// A linked RMP profile in the detail view.
+#[derive(Debug, Clone, Serialize, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export)]
+pub struct LinkedRmpProfile {
+    pub link_id: i32,
+    pub legacy_id: i32,
+    pub first_name: Option<String>,
+    pub last_name: Option<String>,
+    pub department: Option<String>,
+    pub avg_rating: Option<f32>,
+    pub avg_difficulty: Option<f32>,
+    pub num_ratings: Option<i32>,
+    pub would_take_again_pct: Option<f32>,
+}
+
+/// A match candidate in the detail view.
+#[derive(Debug, Clone, Serialize, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export)]
+pub struct CandidateResponse {
+    pub id: i32,
+    pub rmp_legacy_id: i32,
+    pub first_name: Option<String>,
+    pub last_name: Option<String>,
+    pub department: Option<String>,
+    pub avg_rating: Option<f32>,
+    pub avg_difficulty: Option<f32>,
+    pub num_ratings: Option<i32>,
+    pub would_take_again_pct: Option<f32>,
+    pub score: Option<f32>,
+    #[ts(as = "Option<std::collections::HashMap<String, f32>>")]
+    pub score_breakdown: Option<serde_json::Value>,
+    pub status: String,
+}
+
+/// Response for `GET /api/admin/instructors/{id}` and `POST .../match`.
+#[derive(Debug, Clone, Serialize, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export)]
+pub struct InstructorDetailResponse {
+    pub instructor: InstructorDetail,
+    pub current_matches: Vec<LinkedRmpProfile>,
+    pub candidates: Vec<CandidateResponse>,
+}
+
+/// Response for `POST /api/admin/rmp/rescore`.
+#[derive(Debug, Clone, Serialize, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export)]
+pub struct RescoreResponse {
+    pub total_unmatched: usize,
+    pub candidates_created: usize,
+    pub auto_matched: usize,
+    pub skipped_unparseable: usize,
+    pub skipped_no_candidates: usize,
+}
+
+// ---------------------------------------------------------------------------
+// Helper: map sqlx errors to the standard admin error tuple
+// ---------------------------------------------------------------------------
+
+fn db_error(context: &str, e: sqlx::Error) -> (StatusCode, Json<Value>) {
+    tracing::error!(error = %e, "{context}");
+    (
+        StatusCode::INTERNAL_SERVER_ERROR,
+        Json(json!({"error": context})),
+    )
+}
+
+// ---------------------------------------------------------------------------
+// Row types for SQL queries
+// ---------------------------------------------------------------------------
+
+#[derive(sqlx::FromRow)]
+struct InstructorRow {
+    id: i32,
+    display_name: String,
+    email: String,
+    rmp_match_status: String,
+    rmp_link_count: Option<i64>,
+    top_candidate_rmp_id: Option<i32>,
+    top_candidate_score: Option<f32>,
+    top_candidate_breakdown: Option<serde_json::Value>,
+    tc_first_name: Option<String>,
+    tc_last_name: Option<String>,
+    tc_department: Option<String>,
+    tc_avg_rating: Option<f32>,
+    tc_num_ratings: Option<i32>,
+    candidate_count: Option<i64>,
+    course_subject_count: Option<i64>,
+}
+
+#[derive(sqlx::FromRow)]
+struct StatusCount {
+    rmp_match_status: String,
+    count: i64,
+}
+
+#[derive(sqlx::FromRow)]
+struct CandidateRow {
+    id: i32,
+    rmp_legacy_id: i32,
+    score: Option<f32>,
+    score_breakdown: Option<serde_json::Value>,
+    status: String,
+    first_name: Option<String>,
+    last_name: Option<String>,
+    department: Option<String>,
+    avg_rating: Option<f32>,
+    avg_difficulty: Option<f32>,
+    num_ratings: Option<i32>,
+    would_take_again_pct: Option<f32>,
+}
+
+#[derive(sqlx::FromRow)]
+struct LinkedRmpProfileRow {
+    link_id: i32,
+    legacy_id: i32,
+    first_name: Option<String>,
+    last_name: Option<String>,
+    department: Option<String>,
+    avg_rating: Option<f32>,
+    avg_difficulty: Option<f32>,
+    num_ratings: Option<i32>,
+    would_take_again_pct: Option<f32>,
+}
+
+// ---------------------------------------------------------------------------
+// 1. GET /api/admin/instructors — paginated list with filtering
+// ---------------------------------------------------------------------------
+
+/// `GET /api/admin/instructors` — List instructors with filtering and pagination.
+pub async fn list_instructors(
+    AdminUser(_user): AdminUser,
+    State(state): State<AppState>,
+    Query(params): Query<ListInstructorsParams>,
+) -> Result<Json<ListInstructorsResponse>, (StatusCode, Json<Value>)> {
+    let page = params.page.unwrap_or(1).max(1);
+    let per_page = params.per_page.unwrap_or(50).clamp(1, 100);
+    let offset = (page - 1) * per_page;
+
+    let sort_clause = match params.sort.as_deref() {
+        Some("name_asc") => "i.display_name ASC",
+        Some("name_desc") => "i.display_name DESC",
+        Some("status") => "i.rmp_match_status ASC, i.display_name ASC",
+        _ => "tc.score DESC NULLS LAST, i.display_name ASC",
+    };
+
+    // Build WHERE clause
+    let mut conditions = Vec::new();
+    let mut bind_idx = 0u32;
+
+    if params.status.is_some() {
+        bind_idx += 1;
+        conditions.push(format!("i.rmp_match_status = ${bind_idx}"));
+    }
+    if params.search.is_some() {
+        bind_idx += 1;
+        conditions.push(format!(
+            "(i.display_name ILIKE ${bind_idx} OR i.email ILIKE ${bind_idx})"
+        ));
+    }
+
+    let where_clause = if conditions.is_empty() {
+        String::new()
+    } else {
+        format!("WHERE {}", conditions.join(" AND "))
+    };
+
+    let query_str = format!(
+        r#"
+        SELECT
+            i.id, i.display_name, i.email, i.rmp_match_status,
+            (SELECT COUNT(*) FROM instructor_rmp_links irl WHERE irl.instructor_id = i.id) as rmp_link_count,
+            tc.rmp_legacy_id as top_candidate_rmp_id,
+            tc.score as top_candidate_score,
+            tc.score_breakdown as top_candidate_breakdown,
+            rp.first_name as tc_first_name,
+            rp.last_name as tc_last_name,
+            rp.department as tc_department,
+            rp.avg_rating as tc_avg_rating,
+            rp.num_ratings as tc_num_ratings,
+            (SELECT COUNT(*) FROM rmp_match_candidates mc WHERE mc.instructor_id = i.id AND mc.status = 'pending') as candidate_count,
+            (SELECT COUNT(DISTINCT c.subject) FROM course_instructors ci JOIN courses c ON c.id = ci.course_id WHERE ci.instructor_id = i.id) as course_subject_count
+        FROM instructors i
+        LEFT JOIN LATERAL (
+            SELECT mc.rmp_legacy_id, mc.score, mc.score_breakdown
+            FROM rmp_match_candidates mc
+            WHERE mc.instructor_id = i.id AND mc.status = 'pending'
+            ORDER BY mc.score DESC
+            LIMIT 1
+        ) tc ON true
+        LEFT JOIN rmp_professors rp ON rp.legacy_id = tc.rmp_legacy_id
+        {where_clause}
+        ORDER BY {sort_clause}
+        LIMIT {per_page} OFFSET {offset}
+        "#
+    );
+
+    // Build the query with dynamic binds
+    let mut query = sqlx::query_as::<_, InstructorRow>(&query_str);
+    if let Some(ref status) = params.status {
+        query = query.bind(status);
+    }
+    if let Some(ref search) = params.search {
+        query = query.bind(format!("%{search}%"));
+    }
+
+    let rows = query
+        .fetch_all(&state.db_pool)
+        .await
+        .map_err(|e| db_error("failed to list instructors", e))?;
+
+    // Count total with filters
+    let count_query_str = format!("SELECT COUNT(*) FROM instructors i {where_clause}");
+    let mut count_query = sqlx::query_as::<_, (i64,)>(&count_query_str);
+    if let Some(ref status) = params.status {
+        count_query = count_query.bind(status);
+    }
+    if let Some(ref search) = params.search {
+        count_query = count_query.bind(format!("%{search}%"));
+    }
+
+    let (total,) = count_query
+        .fetch_one(&state.db_pool)
+        .await
+        .map_err(|e| db_error("failed to count instructors", e))?;
+
+    // Aggregate stats (unfiltered)
+    let stats_rows = sqlx::query_as::<_, StatusCount>(
+        "SELECT rmp_match_status, COUNT(*) as count FROM instructors GROUP BY rmp_match_status",
+    )
+    .fetch_all(&state.db_pool)
+    .await
+    .map_err(|e| db_error("failed to get instructor stats", e))?;
+
+    // Count instructors with at least one candidate (for progress bar denominator)
+    let (with_candidates,): (i64,) =
+        sqlx::query_as("SELECT COUNT(DISTINCT instructor_id) FROM rmp_match_candidates")
+            .fetch_one(&state.db_pool)
+            .await
+            .map_err(|e| db_error("failed to count instructors with candidates", e))?;
+
+    let mut stats = InstructorStats {
+        total: 0,
+        unmatched: 0,
+        auto: 0,
+        confirmed: 0,
+        rejected: 0,
+        with_candidates,
+    };
+    for row in &stats_rows {
+        stats.total += row.count;
+        match row.rmp_match_status.as_str() {
+            "unmatched" => stats.unmatched = row.count,
+            "auto" => stats.auto = row.count,
+            "confirmed" => stats.confirmed = row.count,
+            "rejected" => stats.rejected = row.count,
+            _ => {}
+        }
+    }
+
+    let instructors: Vec<InstructorListItem> = rows
+        .iter()
+        .map(|r| {
+            let top_candidate = r.top_candidate_rmp_id.map(|rmp_id| TopCandidateResponse {
+                rmp_legacy_id: rmp_id,
+                score: r.top_candidate_score,
+                score_breakdown: r.top_candidate_breakdown.clone(),
+                first_name: r.tc_first_name.clone(),
+                last_name: r.tc_last_name.clone(),
+                department: r.tc_department.clone(),
+                avg_rating: r.tc_avg_rating,
+                num_ratings: r.tc_num_ratings,
+            });
+
+            InstructorListItem {
+                id: r.id,
+                display_name: r.display_name.clone(),
+                email: r.email.clone(),
+                rmp_match_status: r.rmp_match_status.clone(),
+                rmp_link_count: r.rmp_link_count.unwrap_or(0),
+                candidate_count: r.candidate_count.unwrap_or(0),
+                course_subject_count: r.course_subject_count.unwrap_or(0),
+                top_candidate,
+            }
+        })
+        .collect();
+
+    Ok(Json(ListInstructorsResponse {
+        instructors,
+        total,
+        page,
+        per_page,
+        stats,
+    }))
+}
+
+// ---------------------------------------------------------------------------
+// 2. GET /api/admin/instructors/{id} — full detail
+// ---------------------------------------------------------------------------
+
+/// `GET /api/admin/instructors/{id}` — Full instructor detail with candidates.
+pub async fn get_instructor(
+    AdminUser(_user): AdminUser,
+    State(state): State<AppState>,
+    Path(id): Path<i32>,
+) -> Result<Json<InstructorDetailResponse>, (StatusCode, Json<Value>)> {
+    build_instructor_detail(&state, id).await
+}
+
+/// Shared helper that builds the full instructor detail response.
+async fn build_instructor_detail(
+    state: &AppState,
+    id: i32,
+) -> Result<Json<InstructorDetailResponse>, (StatusCode, Json<Value>)> {
+    // Fetch instructor
+    let instructor: Option<(i32, String, String, String)> = sqlx::query_as(
+        "SELECT id, display_name, email, rmp_match_status FROM instructors WHERE id = $1",
+    )
+    .bind(id)
+    .fetch_optional(&state.db_pool)
+    .await
+    .map_err(|e| db_error("failed to fetch instructor", e))?;
+
+    let (inst_id, display_name, email, rmp_match_status) = instructor.ok_or_else(|| {
+        (
+            StatusCode::NOT_FOUND,
+            Json(json!({"error": "instructor not found"})),
+        )
+    })?;
+
+    // Subjects taught
+    let subjects: Vec<(String,)> = sqlx::query_as(
+        "SELECT DISTINCT c.subject FROM course_instructors ci JOIN courses c ON c.id = ci.course_id WHERE ci.instructor_id = $1 ORDER BY c.subject",
+    )
+    .bind(inst_id)
+    .fetch_all(&state.db_pool)
+    .await
+    .map_err(|e| db_error("failed to fetch subjects", e))?;
+
+    // Course count
+    let (course_count,): (i64,) = sqlx::query_as(
+        "SELECT COUNT(DISTINCT ci.course_id) FROM course_instructors ci WHERE ci.instructor_id = $1",
+    )
+    .bind(inst_id)
+    .fetch_one(&state.db_pool)
+    .await
+    .map_err(|e| db_error("failed to count courses", e))?;
+
+    // Candidates with RMP professor info
+    let candidates = sqlx::query_as::<_, CandidateRow>(
+        r#"
+        SELECT mc.id, mc.rmp_legacy_id, mc.score, mc.score_breakdown, mc.status,
+               rp.first_name, rp.last_name, rp.department,
+               rp.avg_rating, rp.avg_difficulty, rp.num_ratings, rp.would_take_again_pct
+        FROM rmp_match_candidates mc
+        JOIN rmp_professors rp ON rp.legacy_id = mc.rmp_legacy_id
+        WHERE mc.instructor_id = $1
+        ORDER BY mc.score DESC
+        "#,
+    )
+    .bind(inst_id)
+    .fetch_all(&state.db_pool)
+    .await
+    .map_err(|e| db_error("failed to fetch candidates", e))?;
+
+    // Current matches (all linked RMP profiles)
+    let current_matches = sqlx::query_as::<_, LinkedRmpProfileRow>(
+        r#"
+        SELECT irl.id as link_id,
+               rp.legacy_id, rp.first_name, rp.last_name, rp.department,
+               rp.avg_rating, rp.avg_difficulty, rp.num_ratings, rp.would_take_again_pct
+        FROM instructor_rmp_links irl
+        JOIN rmp_professors rp ON rp.legacy_id = irl.rmp_legacy_id
+        WHERE irl.instructor_id = $1
+        ORDER BY rp.num_ratings DESC NULLS LAST
+        "#,
+    )
+    .bind(inst_id)
+    .fetch_all(&state.db_pool)
+    .await
+    .map_err(|e| db_error("failed to fetch linked rmp profiles", e))?;
+
+    let current_matches_resp: Vec<LinkedRmpProfile> = current_matches
+        .into_iter()
+        .map(|p| LinkedRmpProfile {
+            link_id: p.link_id,
+            legacy_id: p.legacy_id,
+            first_name: p.first_name,
+            last_name: p.last_name,
+            department: p.department,
+            avg_rating: p.avg_rating,
+            avg_difficulty: p.avg_difficulty,
+            num_ratings: p.num_ratings,
+            would_take_again_pct: p.would_take_again_pct,
+        })
+        .collect();
+
+    let candidates_resp: Vec<CandidateResponse> = candidates
+        .into_iter()
+        .map(|c| CandidateResponse {
+            id: c.id,
+            rmp_legacy_id: c.rmp_legacy_id,
+            first_name: c.first_name,
+            last_name: c.last_name,
+            department: c.department,
+            avg_rating: c.avg_rating,
+            avg_difficulty: c.avg_difficulty,
+            num_ratings: c.num_ratings,
+            would_take_again_pct: c.would_take_again_pct,
+            score: c.score,
+            score_breakdown: c.score_breakdown,
+            status: c.status,
+        })
+        .collect();
+
+    Ok(Json(InstructorDetailResponse {
+        instructor: InstructorDetail {
+            id: inst_id,
+            display_name,
+            email,
+            rmp_match_status,
+            subjects_taught: subjects.into_iter().map(|(s,)| s).collect(),
+            course_count,
+        },
+        current_matches: current_matches_resp,
+        candidates: candidates_resp,
+    }))
+}
+
+// ---------------------------------------------------------------------------
+// 3. POST /api/admin/instructors/{id}/match — accept a candidate
+// ---------------------------------------------------------------------------
+
+/// `POST /api/admin/instructors/{id}/match` — Accept a candidate match.
+pub async fn match_instructor(
+    AdminUser(user): AdminUser,
+    State(state): State<AppState>,
+    Path(id): Path<i32>,
+    Json(body): Json<MatchBody>,
+) -> Result<Json<InstructorDetailResponse>, (StatusCode, Json<Value>)> {
+    // Verify the candidate exists and is pending
+    let candidate: Option<(i32,)> = sqlx::query_as(
+        "SELECT id FROM rmp_match_candidates WHERE instructor_id = $1 AND rmp_legacy_id = $2 AND status = 'pending'",
+    )
+    .bind(id)
+    .bind(body.rmp_legacy_id)
+    .fetch_optional(&state.db_pool)
+    .await
+    .map_err(|e| db_error("failed to check candidate", e))?;
+
+    if candidate.is_none() {
+        return Err((
+            StatusCode::NOT_FOUND,
+            Json(json!({"error": "pending candidate not found for this instructor"})),
+        ));
+    }
+
+    // Check if this RMP profile is already linked to a different instructor
+    let conflict: Option<(i32,)> = sqlx::query_as(
+        "SELECT instructor_id FROM instructor_rmp_links WHERE rmp_legacy_id = $1 AND instructor_id != $2",
+    )
+    .bind(body.rmp_legacy_id)
+    .bind(id)
+    .fetch_optional(&state.db_pool)
+    .await
+    .map_err(|e| db_error("failed to check rmp uniqueness", e))?;
+
+    if let Some((other_id,)) = conflict {
+        return Err((
+            StatusCode::CONFLICT,
+            Json(json!({
+                "error": "RMP profile already linked to another instructor",
+                "conflictingInstructorId": other_id,
+            })),
+        ));
+    }
+
+    let mut tx = state
+        .db_pool
+        .begin()
+        .await
+        .map_err(|e| db_error("failed to begin transaction", e))?;
+
+    // Insert link into instructor_rmp_links
+    sqlx::query(
+        "INSERT INTO instructor_rmp_links (instructor_id, rmp_legacy_id, created_by, source) VALUES ($1, $2, $3, 'manual') ON CONFLICT (rmp_legacy_id) DO NOTHING",
+    )
+    .bind(id)
+    .bind(body.rmp_legacy_id)
+    .bind(user.discord_id)
+    .execute(&mut *tx)
+    .await
+    .map_err(|e| db_error("failed to insert rmp link", e))?;
+
+    // Update instructor match status
+    sqlx::query("UPDATE instructors SET rmp_match_status = 'confirmed' WHERE id = $1")
+        .bind(id)
+        .execute(&mut *tx)
+        .await
+        .map_err(|e| db_error("failed to update instructor match status", e))?;
+
+    // Accept the candidate
+    sqlx::query(
+        "UPDATE rmp_match_candidates SET status = 'accepted', resolved_at = NOW(), resolved_by = $1 WHERE instructor_id = $2 AND rmp_legacy_id = $3",
+    )
+    .bind(user.discord_id)
+    .bind(id)
+    .bind(body.rmp_legacy_id)
+    .execute(&mut *tx)
+    .await
+    .map_err(|e| db_error("failed to accept candidate", e))?;
+
+    tx.commit()
+        .await
+        .map_err(|e| db_error("failed to commit transaction", e))?;
+
+    build_instructor_detail(&state, id).await
+}
+
+// ---------------------------------------------------------------------------
+// 4. POST /api/admin/instructors/{id}/reject-candidate — reject one candidate
+// ---------------------------------------------------------------------------
+
+/// `POST /api/admin/instructors/{id}/reject-candidate` — Reject a single candidate.
+pub async fn reject_candidate(
+    AdminUser(user): AdminUser,
+    State(state): State<AppState>,
+    Path(id): Path<i32>,
+    Json(body): Json<RejectCandidateBody>,
+) -> Result<Json<OkResponse>, (StatusCode, Json<Value>)> {
+    let result = sqlx::query(
+        "UPDATE rmp_match_candidates SET status = 'rejected', resolved_at = NOW(), resolved_by = $1 WHERE instructor_id = $2 AND rmp_legacy_id = $3 AND status = 'pending'",
+    )
+    .bind(user.discord_id)
+    .bind(id)
+    .bind(body.rmp_legacy_id)
+    .execute(&state.db_pool)
+    .await
+    .map_err(|e| db_error("failed to reject candidate", e))?;
+
+    if result.rows_affected() == 0 {
+        return Err((
+            StatusCode::NOT_FOUND,
+            Json(json!({"error": "pending candidate not found"})),
+        ));
+    }
+
+    Ok(Json(OkResponse { ok: true }))
+}
+
+// ---------------------------------------------------------------------------
+// 5. POST /api/admin/instructors/{id}/reject-all — no valid match
+// ---------------------------------------------------------------------------
+
+/// `POST /api/admin/instructors/{id}/reject-all` — Mark instructor as having no valid RMP match.
+pub async fn reject_all(
+    AdminUser(user): AdminUser,
+    State(state): State<AppState>,
+    Path(id): Path<i32>,
+) -> Result<Json<OkResponse>, (StatusCode, Json<Value>)> {
+    let mut tx = state
+        .db_pool
+        .begin()
+        .await
+        .map_err(|e| db_error("failed to begin transaction", e))?;
+
+    // Check current status — cannot reject an instructor with confirmed matches
+    let current_status: Option<(String,)> =
+        sqlx::query_as("SELECT rmp_match_status FROM instructors WHERE id = $1")
+            .bind(id)
+            .fetch_optional(&mut *tx)
+            .await
+            .map_err(|e| db_error("failed to fetch instructor status", e))?;
+
+    let (status,) = current_status.ok_or_else(|| {
+        (
+            StatusCode::NOT_FOUND,
+            Json(json!({"error": "instructor not found"})),
+        )
+    })?;
+
+    if status == "confirmed" {
+        return Err((
+            StatusCode::CONFLICT,
+            Json(
+                json!({"error": "cannot reject instructor with confirmed matches — unmatch first"}),
+            ),
+        ));
+    }
+
+    // Update instructor status
+    sqlx::query("UPDATE instructors SET rmp_match_status = 'rejected' WHERE id = $1")
+        .bind(id)
+        .execute(&mut *tx)
+        .await
+        .map_err(|e| db_error("failed to update instructor status", e))?;
+
+    // Reject all pending candidates
+    sqlx::query(
+        "UPDATE rmp_match_candidates SET status = 'rejected', resolved_at = NOW(), resolved_by = $1 WHERE instructor_id = $2 AND status = 'pending'",
+    )
+    .bind(user.discord_id)
+    .bind(id)
+    .execute(&mut *tx)
+    .await
+    .map_err(|e| db_error("failed to reject candidates", e))?;
+
+    tx.commit()
+        .await
+        .map_err(|e| db_error("failed to commit transaction", e))?;
+
+    Ok(Json(OkResponse { ok: true }))
+}
+
+// ---------------------------------------------------------------------------
+// 6. POST /api/admin/instructors/{id}/unmatch — remove current match
+// ---------------------------------------------------------------------------
+
+/// Body for unmatch — optional `rmpLegacyId` to remove a specific link.
+/// If omitted (or null), all links are removed.
+#[derive(Deserialize, Default)]
+#[serde(rename_all = "camelCase")]
+pub struct UnmatchBody {
+    rmp_legacy_id: Option<i32>,
+}
+
+/// `POST /api/admin/instructors/{id}/unmatch` — Remove RMP link(s).
+///
+/// Send `{ "rmpLegacyId": N }` to remove a specific link, or an empty body / `{}`
+/// to remove all links for the instructor.
+pub async fn unmatch_instructor(
+    AdminUser(_user): AdminUser,
+    State(state): State<AppState>,
+    Path(id): Path<i32>,
+    body: Option<Json<UnmatchBody>>,
+) -> Result<Json<OkResponse>, (StatusCode, Json<Value>)> {
+    let rmp_legacy_id = body.and_then(|b| b.rmp_legacy_id);
+
+    let mut tx = state
+        .db_pool
+        .begin()
+        .await
+        .map_err(|e| db_error("failed to begin transaction", e))?;
+
+    // Verify instructor exists
+    let exists: Option<(i32,)> = sqlx::query_as("SELECT id FROM instructors WHERE id = $1")
+        .bind(id)
+        .fetch_optional(&mut *tx)
+        .await
+        .map_err(|e| db_error("failed to check instructor", e))?;
+
+    if exists.is_none() {
+        return Err((
+            StatusCode::NOT_FOUND,
+            Json(json!({"error": "instructor not found"})),
+        ));
+    }
+
+    // Delete specific link or all links
+    if let Some(legacy_id) = rmp_legacy_id {
+        let result = sqlx::query(
+            "DELETE FROM instructor_rmp_links WHERE instructor_id = $1 AND rmp_legacy_id = $2",
+        )
+        .bind(id)
+        .bind(legacy_id)
+        .execute(&mut *tx)
+        .await
+        .map_err(|e| db_error("failed to remove rmp link", e))?;
+
+        if result.rows_affected() == 0 {
+            return Err((
+                StatusCode::NOT_FOUND,
+                Json(json!({"error": "link not found for this instructor"})),
+            ));
+        }
+    } else {
+        sqlx::query("DELETE FROM instructor_rmp_links WHERE instructor_id = $1")
+            .bind(id)
+            .execute(&mut *tx)
+            .await
+            .map_err(|e| db_error("failed to remove rmp links", e))?;
+    }
+
+    // Check if any links remain; update status accordingly
+    let (remaining,): (i64,) =
+        sqlx::query_as("SELECT COUNT(*) FROM instructor_rmp_links WHERE instructor_id = $1")
+            .bind(id)
+            .fetch_one(&mut *tx)
+            .await
+            .map_err(|e| db_error("failed to count remaining links", e))?;
+
+    if remaining == 0 {
+        sqlx::query("UPDATE instructors SET rmp_match_status = 'unmatched' WHERE id = $1")
+            .bind(id)
+            .execute(&mut *tx)
+            .await
+            .map_err(|e| db_error("failed to update instructor status", e))?;
+    }
+
+    tx.commit()
+        .await
+        .map_err(|e| db_error("failed to commit transaction", e))?;
+
+    Ok(Json(OkResponse { ok: true }))
+}
+
+// ---------------------------------------------------------------------------
+// 7. POST /api/admin/rmp/rescore — re-run candidate generation
+// ---------------------------------------------------------------------------
+
+/// `POST /api/admin/rmp/rescore` — Re-run RMP candidate generation.
+pub async fn rescore(
+    AdminUser(_user): AdminUser,
+    State(state): State<AppState>,
+) -> Result<Json<RescoreResponse>, (StatusCode, Json<Value>)> {
+    let stats = crate::data::rmp_matching::generate_candidates(&state.db_pool)
+        .await
+        .map_err(|e| {
+            tracing::error!(error = %e, "failed to run candidate generation");
+            (
+                StatusCode::INTERNAL_SERVER_ERROR,
+                Json(json!({"error": "candidate generation failed"})),
+            )
+        })?;
+
+    Ok(Json(RescoreResponse {
+        total_unmatched: stats.total_unmatched,
+        candidates_created: stats.candidates_created,
+        auto_matched: stats.auto_matched,
+        skipped_unparseable: stats.skipped_unparseable,
+        skipped_no_candidates: stats.skipped_no_candidates,
+    }))
+}
@@ -0,0 +1,523 @@
+//! Admin API handlers for scraper observability.
+//!
+//! All endpoints require the `AdminUser` extractor, returning 401/403 as needed.
+
+use axum::extract::{Path, Query, State};
+use axum::http::StatusCode;
+use axum::response::Json;
+use chrono::{DateTime, Utc};
+use serde::{Deserialize, Serialize};
+use serde_json::json;
+use sqlx::Row;
+use ts_rs::TS;
+
+use crate::banner::models::terms::Term;
+use crate::data::scrape_jobs;
+use crate::scraper::adaptive::{self, SubjectSchedule, SubjectStats};
+use crate::state::AppState;
+use crate::web::extractors::AdminUser;
+
+type ApiError = (StatusCode, Json<serde_json::Value>);
+
+fn parse_period(period: &str) -> Result<chrono::Duration, ApiError> {
+    match period {
+        "1h" => Ok(chrono::Duration::hours(1)),
+        "6h" => Ok(chrono::Duration::hours(6)),
+        "24h" => Ok(chrono::Duration::hours(24)),
+        "7d" => Ok(chrono::Duration::days(7)),
+        "30d" => Ok(chrono::Duration::days(30)),
+        _ => Err((
+            StatusCode::BAD_REQUEST,
+            Json(
+                json!({"error": format!("Invalid period '{period}'. Valid: 1h, 6h, 24h, 7d, 30d")}),
+            ),
+        )),
+    }
+}
+
+fn period_to_interval_str(period: &str) -> &'static str {
+    match period {
+        "1h" => "1 hour",
+        "6h" => "6 hours",
+        "24h" => "24 hours",
+        "7d" => "7 days",
+        "30d" => "30 days",
+        _ => "24 hours",
+    }
+}
+
+fn parse_bucket(bucket: &str) -> Result<&'static str, ApiError> {
+    match bucket {
+        "1m" => Ok("1 minute"),
+        "5m" => Ok("5 minutes"),
+        "15m" => Ok("15 minutes"),
+        "1h" => Ok("1 hour"),
+        "6h" => Ok("6 hours"),
+        _ => Err((
+            StatusCode::BAD_REQUEST,
+            Json(
+                json!({"error": format!("Invalid bucket '{bucket}'. Valid: 1m, 5m, 15m, 1h, 6h")}),
+            ),
+        )),
+    }
+}
+
+fn default_bucket_for_period(period: &str) -> &'static str {
+    match period {
+        "1h" => "1m",
+        "6h" => "5m",
+        "24h" => "15m",
+        "7d" => "1h",
+        "30d" => "6h",
+        _ => "15m",
+    }
+}
+
+// ---------------------------------------------------------------------------
+// Endpoint 1: GET /api/admin/scraper/stats
+// ---------------------------------------------------------------------------
+
+#[derive(Deserialize)]
+pub struct StatsParams {
+    #[serde(default = "default_period")]
+    period: String,
+}
+
+fn default_period() -> String {
+    "24h".to_string()
+}
+
+#[derive(Serialize, TS)]
+#[ts(export)]
+#[serde(rename_all = "camelCase")]
+pub struct ScraperStatsResponse {
+    period: String,
+    #[ts(type = "number")]
+    total_scrapes: i64,
+    #[ts(type = "number")]
+    successful_scrapes: i64,
+    #[ts(type = "number")]
+    failed_scrapes: i64,
+    success_rate: Option<f64>,
+    avg_duration_ms: Option<f64>,
+    #[ts(type = "number")]
+    total_courses_changed: i64,
+    #[ts(type = "number")]
+    total_courses_fetched: i64,
+    #[ts(type = "number")]
+    total_audits_generated: i64,
+    #[ts(type = "number")]
+    pending_jobs: i64,
+    #[ts(type = "number")]
+    locked_jobs: i64,
+}
+
+pub async fn scraper_stats(
+    _admin: AdminUser,
+    State(state): State<AppState>,
+    Query(params): Query<StatsParams>,
+) -> Result<Json<ScraperStatsResponse>, ApiError> {
+    let _duration = parse_period(&params.period)?;
+    let interval_str = period_to_interval_str(&params.period);
+
+    let row = sqlx::query(
+        "SELECT \
+            COUNT(*) AS total_scrapes, \
+            COUNT(*) FILTER (WHERE success) AS successful_scrapes, \
+            COUNT(*) FILTER (WHERE NOT success) AS failed_scrapes, \
+            (AVG(duration_ms) FILTER (WHERE success))::FLOAT8 AS avg_duration_ms, \
+            COALESCE(SUM(courses_changed) FILTER (WHERE success), 0) AS total_courses_changed, \
+            COALESCE(SUM(courses_fetched) FILTER (WHERE success), 0) AS total_courses_fetched, \
+            COALESCE(SUM(audits_generated) FILTER (WHERE success), 0) AS total_audits_generated \
+         FROM scrape_job_results \
+         WHERE completed_at > NOW() - $1::interval",
+    )
+    .bind(interval_str)
+    .fetch_one(&state.db_pool)
+    .await
+    .map_err(|e| {
+        tracing::error!(error = %e, "Failed to fetch scraper stats");
+        (
+            StatusCode::INTERNAL_SERVER_ERROR,
+            Json(json!({"error": "Failed to fetch scraper stats"})),
+        )
+    })?;
+
+    let total_scrapes: i64 = row.get("total_scrapes");
+    let successful_scrapes: i64 = row.get("successful_scrapes");
+    let failed_scrapes: i64 = row.get("failed_scrapes");
+    let avg_duration_ms: Option<f64> = row.get("avg_duration_ms");
+    let total_courses_changed: i64 = row.get("total_courses_changed");
+    let total_courses_fetched: i64 = row.get("total_courses_fetched");
+    let total_audits_generated: i64 = row.get("total_audits_generated");
+
+    let queue_row = sqlx::query(
+        "SELECT \
+            COUNT(*) FILTER (WHERE locked_at IS NULL) AS pending_jobs, \
+            COUNT(*) FILTER (WHERE locked_at IS NOT NULL) AS locked_jobs \
+         FROM scrape_jobs",
+    )
+    .fetch_one(&state.db_pool)
+    .await
+    .map_err(|e| {
+        tracing::error!(error = %e, "Failed to fetch queue stats");
+        (
+            StatusCode::INTERNAL_SERVER_ERROR,
+            Json(json!({"error": "Failed to fetch queue stats"})),
+        )
+    })?;
+
+    let pending_jobs: i64 = queue_row.get("pending_jobs");
+    let locked_jobs: i64 = queue_row.get("locked_jobs");
+
+    let success_rate = if total_scrapes > 0 {
+        Some(successful_scrapes as f64 / total_scrapes as f64)
+    } else {
+        None
+    };
+
+    Ok(Json(ScraperStatsResponse {
+        period: params.period,
+        total_scrapes,
+        successful_scrapes,
+        failed_scrapes,
+        success_rate,
+        avg_duration_ms,
+        total_courses_changed,
+        total_courses_fetched,
+        total_audits_generated,
+        pending_jobs,
+        locked_jobs,
+    }))
+}
+
+// ---------------------------------------------------------------------------
+// Endpoint 2: GET /api/admin/scraper/timeseries
+// ---------------------------------------------------------------------------
+
+#[derive(Deserialize)]
+pub struct TimeseriesParams {
+    #[serde(default = "default_period")]
+    period: String,
+    bucket: Option<String>,
+}
+
+#[derive(Serialize, TS)]
+#[ts(export)]
+#[serde(rename_all = "camelCase")]
+pub struct TimeseriesResponse {
+    period: String,
+    bucket: String,
+    points: Vec<TimeseriesPoint>,
+}
+
+#[derive(Serialize, TS)]
+#[ts(export)]
+#[serde(rename_all = "camelCase")]
+pub struct TimeseriesPoint {
+    timestamp: DateTime<Utc>,
+    #[ts(type = "number")]
+    scrape_count: i64,
+    #[ts(type = "number")]
+    success_count: i64,
+    #[ts(type = "number")]
+    error_count: i64,
+    #[ts(type = "number")]
+    courses_changed: i64,
+    avg_duration_ms: f64,
+}
+
+pub async fn scraper_timeseries(
+    _admin: AdminUser,
+    State(state): State<AppState>,
+    Query(params): Query<TimeseriesParams>,
+) -> Result<Json<TimeseriesResponse>, ApiError> {
+    let _duration = parse_period(&params.period)?;
+    let period_interval = period_to_interval_str(&params.period);
+
+    let bucket_code = match &params.bucket {
+        Some(b) => {
+            // Validate the bucket
+            parse_bucket(b)?;
+            b.as_str()
+        }
+        None => default_bucket_for_period(&params.period),
+    };
+    let bucket_interval = parse_bucket(bucket_code)?;
+
+    let rows = sqlx::query(
+        "WITH buckets AS ( \
+            SELECT generate_series( \
+                date_bin($1::interval, NOW() - $2::interval, '2020-01-01'::timestamptz), \
+                date_bin($1::interval, NOW(), '2020-01-01'::timestamptz), \
+                $1::interval \
+            ) AS bucket_start \
+         ), \
+         raw AS ( \
+            SELECT date_bin($1::interval, completed_at, '2020-01-01'::timestamptz) AS bucket_start, \
+                   COUNT(*)::BIGINT AS scrape_count, \
+                   COUNT(*) FILTER (WHERE success)::BIGINT AS success_count, \
+                   COUNT(*) FILTER (WHERE NOT success)::BIGINT AS error_count, \
+                   COALESCE(SUM(courses_changed) FILTER (WHERE success), 0)::BIGINT AS courses_changed, \
+                   COALESCE(AVG(duration_ms) FILTER (WHERE success), 0)::FLOAT8 AS avg_duration_ms \
+            FROM scrape_job_results \
+            WHERE completed_at > NOW() - $2::interval \
+            GROUP BY 1 \
+         ) \
+         SELECT b.bucket_start, \
+                COALESCE(r.scrape_count, 0) AS scrape_count, \
+                COALESCE(r.success_count, 0) AS success_count, \
+                COALESCE(r.error_count, 0) AS error_count, \
+                COALESCE(r.courses_changed, 0) AS courses_changed, \
+                COALESCE(r.avg_duration_ms, 0) AS avg_duration_ms \
+         FROM buckets b \
+         LEFT JOIN raw r ON b.bucket_start = r.bucket_start \
+         ORDER BY b.bucket_start",
+    )
+    .bind(bucket_interval)
+    .bind(period_interval)
+    .fetch_all(&state.db_pool)
+    .await
+    .map_err(|e| {
+        tracing::error!(error = %e, "Failed to fetch scraper timeseries");
+        (
+            StatusCode::INTERNAL_SERVER_ERROR,
+            Json(json!({"error": "Failed to fetch scraper timeseries"})),
+        )
+    })?;
+
+    let points = rows
+        .iter()
+        .map(|row| TimeseriesPoint {
+            timestamp: row.get("bucket_start"),
+            scrape_count: row.get("scrape_count"),
+            success_count: row.get("success_count"),
+            error_count: row.get("error_count"),
+            courses_changed: row.get("courses_changed"),
+            avg_duration_ms: row.get("avg_duration_ms"),
+        })
+        .collect();
+
+    Ok(Json(TimeseriesResponse {
+        period: params.period,
+        bucket: bucket_code.to_string(),
+        points,
+    }))
+}
+
+// ---------------------------------------------------------------------------
+// Endpoint 3: GET /api/admin/scraper/subjects
+// ---------------------------------------------------------------------------
+
+#[derive(Serialize, TS)]
+#[ts(export)]
+#[serde(rename_all = "camelCase")]
+pub struct SubjectsResponse {
+    subjects: Vec<SubjectSummary>,
+}
+
+#[derive(Serialize, TS)]
+#[ts(export)]
+#[serde(rename_all = "camelCase")]
+pub struct SubjectSummary {
+    subject: String,
+    subject_description: Option<String>,
+    #[ts(type = "number")]
+    tracked_course_count: i64,
+    schedule_state: String,
+    #[ts(type = "number")]
+    current_interval_secs: u64,
+    time_multiplier: u32,
+    last_scraped: DateTime<Utc>,
+    next_eligible_at: Option<DateTime<Utc>>,
+    #[ts(type = "number | null")]
+    cooldown_remaining_secs: Option<u64>,
+    avg_change_ratio: f64,
+    #[ts(type = "number")]
+    consecutive_zero_changes: i64,
+    #[ts(type = "number")]
+    recent_runs: i64,
+    #[ts(type = "number")]
+    recent_failures: i64,
+}
+
+pub async fn scraper_subjects(
+    _admin: AdminUser,
+    State(state): State<AppState>,
+) -> Result<Json<SubjectsResponse>, ApiError> {
+    let raw_stats = scrape_jobs::fetch_subject_stats(&state.db_pool)
+        .await
+        .map_err(|e| {
+            tracing::error!(error = %e, "Failed to fetch subject stats");
+            (
+                StatusCode::INTERNAL_SERVER_ERROR,
+                Json(json!({"error": "Failed to fetch subject stats"})),
+            )
+        })?;
+
+    let now = Utc::now();
+    let multiplier = adaptive::time_of_day_multiplier(now);
+
+    // Look up subject descriptions from the reference cache
+    let ref_cache = state.reference_cache.read().await;
+
+    // Count tracked courses per subject for the current term
+    let term = Term::get_current().inner().to_string();
+    let course_counts: std::collections::HashMap<String, i64> = sqlx::query_as(
+        "SELECT subject, COUNT(*)::BIGINT AS cnt FROM courses WHERE term_code = $1 GROUP BY subject",
+    )
+    .bind(&term)
+    .fetch_all(&state.db_pool)
+    .await
+    .map_err(|e| {
+        tracing::error!(error = %e, "Failed to fetch course counts");
+        (
+            StatusCode::INTERNAL_SERVER_ERROR,
+            Json(json!({"error": "Failed to fetch course counts"})),
+        )
+    })?
+    .into_iter()
+    .map(|(subject, cnt): (String, i64)| (subject, cnt))
+    .collect();
+
+    let subjects = raw_stats
+        .into_iter()
+        .map(|row| {
+            let stats: SubjectStats = row.into();
+            let schedule = adaptive::evaluate_subject(&stats, now, false);
+            let base_interval = adaptive::compute_base_interval(&stats);
+
+            let schedule_state = match &schedule {
+                SubjectSchedule::Eligible(_) => "eligible",
+                SubjectSchedule::Cooldown(_) => "cooldown",
+                SubjectSchedule::Paused => "paused",
+                SubjectSchedule::ReadOnly => "read_only",
+            };
+
+            let current_interval_secs = base_interval.as_secs() * multiplier as u64;
+
+            let (next_eligible_at, cooldown_remaining_secs) = match &schedule {
+                SubjectSchedule::Eligible(_) => (Some(now), Some(0)),
+                SubjectSchedule::Cooldown(remaining) => {
+                    let remaining_secs = remaining.as_secs();
+                    (
+                        Some(now + chrono::Duration::seconds(remaining_secs as i64)),
+                        Some(remaining_secs),
+                    )
+                }
+                SubjectSchedule::Paused | SubjectSchedule::ReadOnly => (None, None),
+            };
+
+            let subject_description = ref_cache
+                .lookup("subject", &stats.subject)
+                .map(|s| s.to_string());
+
+            let tracked_course_count = course_counts.get(&stats.subject).copied().unwrap_or(0);
+
+            SubjectSummary {
+                subject: stats.subject,
+                subject_description,
+                tracked_course_count,
+                schedule_state: schedule_state.to_string(),
+                current_interval_secs,
+                time_multiplier: multiplier,
+                last_scraped: stats.last_completed,
+                next_eligible_at,
+                cooldown_remaining_secs,
+                avg_change_ratio: stats.avg_change_ratio,
+                consecutive_zero_changes: stats.consecutive_zero_changes,
+                recent_runs: stats.recent_runs,
+                recent_failures: stats.recent_failure_count,
+            }
+        })
+        .collect();
+
+    Ok(Json(SubjectsResponse { subjects }))
+}
+
+// ---------------------------------------------------------------------------
+// Endpoint 4: GET /api/admin/scraper/subjects/{subject}
+// ---------------------------------------------------------------------------
+
+#[derive(Deserialize)]
+pub struct SubjectDetailParams {
+    #[serde(default = "default_detail_limit")]
+    limit: i32,
+}
+
+fn default_detail_limit() -> i32 {
+    50
+}
+
+#[derive(Serialize, TS)]
+#[ts(export)]
+#[serde(rename_all = "camelCase")]
+pub struct SubjectDetailResponse {
+    subject: String,
+    results: Vec<SubjectResultEntry>,
+}
+
+#[derive(Serialize, TS)]
+#[ts(export)]
+#[serde(rename_all = "camelCase")]
+pub struct SubjectResultEntry {
+    #[ts(type = "number")]
+    id: i64,
+    completed_at: DateTime<Utc>,
+    duration_ms: i32,
+    success: bool,
+    error_message: Option<String>,
+    courses_fetched: Option<i32>,
+    courses_changed: Option<i32>,
+    courses_unchanged: Option<i32>,
+    audits_generated: Option<i32>,
+    metrics_generated: Option<i32>,
+}
+
+pub async fn scraper_subject_detail(
+    _admin: AdminUser,
+    State(state): State<AppState>,
+    Path(subject): Path<String>,
+    Query(params): Query<SubjectDetailParams>,
+) -> Result<Json<SubjectDetailResponse>, ApiError> {
+    let limit = params.limit.clamp(1, 200);
+
+    let rows = sqlx::query(
+        "SELECT id, completed_at, duration_ms, success, error_message, \
+                courses_fetched, courses_changed, courses_unchanged, \
+                audits_generated, metrics_generated \
+         FROM scrape_job_results \
+         WHERE target_type = 'Subject' AND payload->>'subject' = $1 \
+         ORDER BY completed_at DESC \
+         LIMIT $2",
+    )
+    .bind(&subject)
+    .bind(limit)
+    .fetch_all(&state.db_pool)
+    .await
+    .map_err(|e| {
+        tracing::error!(error = %e, subject = %subject, "Failed to fetch subject detail");
+        (
+            StatusCode::INTERNAL_SERVER_ERROR,
+            Json(json!({"error": "Failed to fetch subject detail"})),
+        )
+    })?;
+
+    let results = rows
+        .iter()
+        .map(|row| SubjectResultEntry {
+            id: row.get("id"),
+            completed_at: row.get("completed_at"),
+            duration_ms: row.get("duration_ms"),
+            success: row.get("success"),
+            error_message: row.get("error_message"),
+            courses_fetched: row.get("courses_fetched"),
+            courses_changed: row.get("courses_changed"),
+            courses_unchanged: row.get("courses_unchanged"),
+            audits_generated: row.get("audits_generated"),
+            metrics_generated: row.get("metrics_generated"),
+        })
+        .collect();
+
+    Ok(Json(SubjectDetailResponse { subject, results }))
+}
@@ -1,14 +1,18 @@
-//! Embedded assets for the web frontend
+//! Embedded assets for the web frontend.
 //!
-//! This module handles serving static assets that are embedded into the binary
-//! at compile time using rust-embed.
+//! Serves static assets embedded into the binary at compile time using rust-embed.
+//! Supports content negotiation for pre-compressed variants (.br, .gz, .zst)
+//! generated at build time by `web/scripts/compress-assets.ts`.

+use axum::http::{HeaderMap, HeaderValue, header};
 use dashmap::DashMap;
 use rapidhash::v3::rapidhash_v3;
 use rust_embed::RustEmbed;
 use std::fmt;
 use std::sync::LazyLock;

+use super::encoding::{COMPRESSION_MIN_SIZE, ContentEncoding, parse_accepted_encodings};
+
 /// Embedded web assets from the dist directory
 #[derive(RustEmbed)]
 #[folder = "web/dist/"]
@@ -21,17 +25,15 @@ pub struct WebAssets;
 pub struct AssetHash(u64);

 impl AssetHash {
-    /// Create a new AssetHash from u64 value
    pub fn new(hash: u64) -> Self {
        Self(hash)
    }

-    /// Get the hash as a hex string
    pub fn to_hex(&self) -> String {
        format!("{:016x}", self.0)
    }

-    /// Get the hash as a quoted hex string
+    /// Get the hash as a quoted hex string (for ETag headers)
    pub fn quoted(&self) -> String {
        format!("\"{}\"", self.to_hex())
    }
@@ -51,12 +53,8 @@ pub struct AssetMetadata {
 }

 impl AssetMetadata {
-    /// Check if the etag matches the asset hash
    pub fn etag_matches(&self, etag: &str) -> bool {
-        // Remove quotes if present (ETags are typically quoted)
        let etag = etag.trim_matches('"');
-
-        // ETags generated from u64 hex should be 16 characters
        etag.len() == 16
            && u64::from_str_radix(etag, 16)
                .map(|parsed| parsed == self.hash.0)
@@ -68,28 +66,125 @@ impl AssetMetadata {
 static ASSET_CACHE: LazyLock<DashMap<String, AssetMetadata>> = LazyLock::new(DashMap::new);

 /// Get cached asset metadata for a file path, caching on-demand
-/// Returns AssetMetadata containing MIME type and RapidHash hash
 pub fn get_asset_metadata_cached(path: &str, content: &[u8]) -> AssetMetadata {
-    // Check cache first
    if let Some(cached) = ASSET_CACHE.get(path) {
        return cached.value().clone();
    }

-    // Calculate MIME type
    let mime_type = mime_guess::from_path(path)
        .first()
        .map(|mime| mime.to_string());
-
-    // Calculate RapidHash hash (using u64 native output size)
-    let hash_value = rapidhash_v3(content);
-    let hash = AssetHash::new(hash_value);
-
+    let hash = AssetHash::new(rapidhash_v3(content));
    let metadata = AssetMetadata { mime_type, hash };

-    // Only cache if we haven't exceeded the limit
    if ASSET_CACHE.len() < 1000 {
        ASSET_CACHE.insert(path.to_string(), metadata.clone());
    }

    metadata
 }
+
+/// Set appropriate `Cache-Control` header based on the asset path.
+///
+/// SvelteKit outputs fingerprinted assets under `_app/immutable/` which are
+/// safe to cache indefinitely. Other assets get shorter cache durations.
+fn set_cache_control(headers: &mut HeaderMap, path: &str) {
+    let cache_control = if path.contains("immutable/") {
+        // SvelteKit fingerprinted assets — cache forever
+        "public, max-age=31536000, immutable"
+    } else if path == "index.html" || path.ends_with(".html") {
+        "public, max-age=300"
+    } else {
+        match path.rsplit_once('.').map(|(_, ext)| ext) {
+            Some("css" | "js") => "public, max-age=86400",
+            Some("png" | "jpg" | "jpeg" | "gif" | "svg" | "ico") => "public, max-age=2592000",
+            _ => "public, max-age=3600",
+        }
+    };
+
+    if let Ok(value) = HeaderValue::from_str(cache_control) {
+        headers.insert(header::CACHE_CONTROL, value);
+    }
+}
+
+/// Serve an embedded asset with content encoding negotiation.
+///
+/// Tries pre-compressed variants (.br, .gz, .zst) in the order preferred by
+/// the client's `Accept-Encoding` header, falling back to the uncompressed
+/// original. Returns `None` if the asset doesn't exist at all.
+pub fn try_serve_asset_with_encoding(
+    path: &str,
+    request_headers: &HeaderMap,
+) -> Option<axum::response::Response> {
+    use axum::response::IntoResponse;
+
+    let asset_path = path.strip_prefix('/').unwrap_or(path);
+
+    // Get the uncompressed original first (for metadata: MIME type, ETag)
+    let original = WebAssets::get(asset_path)?;
+    let metadata = get_asset_metadata_cached(asset_path, &original.data);
+
+    // Check ETag for conditional requests (304 Not Modified)
+    if let Some(etag) = request_headers.get(header::IF_NONE_MATCH)
+        && etag.to_str().is_ok_and(|s| metadata.etag_matches(s))
+    {
+        return Some(axum::http::StatusCode::NOT_MODIFIED.into_response());
+    }
+
+    let mime_type = metadata
+        .mime_type
+        .unwrap_or_else(|| "application/octet-stream".to_string());
+
+    // Only attempt pre-compressed variants for files above the compression
+    // threshold — the build script skips smaller files too.
+    let accepted_encodings = if original.data.len() >= COMPRESSION_MIN_SIZE {
+        parse_accepted_encodings(request_headers)
+    } else {
+        vec![ContentEncoding::Identity]
+    };
+
+    for encoding in &accepted_encodings {
+        if *encoding == ContentEncoding::Identity {
+            continue;
+        }
+
+        let compressed_path = format!("{}{}", asset_path, encoding.extension());
+        if let Some(compressed) = WebAssets::get(&compressed_path) {
+            let mut response_headers = HeaderMap::new();
+
+            if let Ok(ct) = HeaderValue::from_str(&mime_type) {
+                response_headers.insert(header::CONTENT_TYPE, ct);
+            }
+            if let Some(ce) = encoding.header_value() {
+                response_headers.insert(header::CONTENT_ENCODING, ce);
+            }
+            if let Ok(etag_val) = HeaderValue::from_str(&metadata.hash.quoted()) {
+                response_headers.insert(header::ETAG, etag_val);
+            }
+            // Vary so caches distinguish by encoding
+            response_headers.insert(header::VARY, HeaderValue::from_static("Accept-Encoding"));
+            set_cache_control(&mut response_headers, asset_path);
+
+            return Some(
+                (
+                    axum::http::StatusCode::OK,
+                    response_headers,
+                    compressed.data,
+                )
+                    .into_response(),
+            );
+        }
+    }
+
+    // No compressed variant found — serve uncompressed original
+    let mut response_headers = HeaderMap::new();
+    if let Ok(ct) = HeaderValue::from_str(&mime_type) {
+        response_headers.insert(header::CONTENT_TYPE, ct);
+    }
+    if let Ok(etag_val) = HeaderValue::from_str(&metadata.hash.quoted()) {
+        response_headers.insert(header::ETAG, etag_val);
+    }
+    set_cache_control(&mut response_headers, asset_path);
+
+    Some((axum::http::StatusCode::OK, response_headers, original.data).into_response())
+}
@@ -0,0 +1,300 @@
+//! Discord OAuth2 authentication handlers.
+//!
+//! Provides login, callback, logout, and session introspection endpoints
+//! for Discord OAuth2 authentication flow.
+
+use axum::extract::{Extension, Query, State};
+use axum::http::{HeaderMap, StatusCode, header};
+use axum::response::{IntoResponse, Json, Redirect, Response};
+use serde::Deserialize;
+use serde_json::{Value, json};
+use std::time::Duration;
+use tracing::{error, info, warn};
+
+use crate::state::AppState;
+
+/// OAuth configuration passed as an Axum Extension.
+#[derive(Clone)]
+pub struct AuthConfig {
+    pub client_id: String,
+    pub client_secret: String,
+    /// Optional base URL override (e.g. "https://banner.xevion.dev").
+    /// When `None`, the redirect URI is derived from the request's Origin/Host header.
+    pub redirect_base: Option<String>,
+}
+
+const CALLBACK_PATH: &str = "/api/auth/callback";
+
+/// Derive the origin (scheme + host + port) the user's browser is actually on.
+///
+/// Priority:
+/// 1. Configured `redirect_base` (production override)
+/// 2. `Referer` header — preserves the real browser origin even through
+///    reverse proxies that rewrite `Host` (e.g. Vite dev proxy with
+///    `changeOrigin: true`)
+/// 3. `Origin` header (present on POST / CORS requests)
+/// 4. `Host` header (last resort, may be rewritten by proxies)
+fn resolve_origin(auth_config: &AuthConfig, headers: &HeaderMap) -> String {
+    if let Some(base) = &auth_config.redirect_base {
+        return base.trim_end_matches('/').to_owned();
+    }
+
+    // Referer carries the full browser URL; extract just the origin.
+    if let Some(referer) = headers.get(header::REFERER).and_then(|v| v.to_str().ok())
+        && let Ok(parsed) = url::Url::parse(referer)
+    {
+        let origin = parsed.origin().unicode_serialization();
+        if origin != "null" {
+            return origin;
+        }
+    }
+
+    if let Some(origin) = headers.get("origin").and_then(|v| v.to_str().ok()) {
+        return origin.trim_end_matches('/').to_owned();
+    }
+
+    if let Some(host) = headers.get(header::HOST).and_then(|v| v.to_str().ok()) {
+        return format!("http://{host}");
+    }
+
+    "http://localhost:8080".to_owned()
+}
+
+#[derive(Deserialize)]
+pub struct CallbackParams {
+    code: String,
+    state: String,
+}
+
+#[derive(Deserialize)]
+struct TokenResponse {
+    access_token: String,
+}
+
+#[derive(Deserialize)]
+struct DiscordUser {
+    id: String,
+    username: String,
+    avatar: Option<String>,
+}
+
+/// Extract the `session` cookie value from request headers.
+fn extract_session_token(headers: &HeaderMap) -> Option<String> {
+    headers
+        .get(header::COOKIE)?
+        .to_str()
+        .ok()?
+        .split(';')
+        .find_map(|cookie| {
+            let cookie = cookie.trim();
+            cookie.strip_prefix("session=").map(|v| v.to_owned())
+        })
+}
+
+/// Build a `Set-Cookie` header value for the session cookie.
+fn session_cookie(token: &str, max_age: i64, secure: bool) -> String {
+    let mut cookie = format!("session={token}; HttpOnly; SameSite=Lax; Path=/; Max-Age={max_age}");
+    if secure {
+        cookie.push_str("; Secure");
+    }
+    cookie
+}
+
+/// `GET /api/auth/login` — Redirect to Discord OAuth2 authorization page.
+pub async fn auth_login(
+    State(state): State<AppState>,
+    Extension(auth_config): Extension<AuthConfig>,
+    headers: HeaderMap,
+) -> Redirect {
+    let origin = resolve_origin(&auth_config, &headers);
+    let redirect_uri = format!("{origin}{CALLBACK_PATH}");
+    let csrf_state = state.oauth_state_store.generate(origin);
+    let redirect_uri_encoded = urlencoding::encode(&redirect_uri);
+
+    let url = format!(
+        "https://discord.com/oauth2/authorize\
+         ?client_id={}\
+         &redirect_uri={redirect_uri_encoded}\
+         &response_type=code\
+         &scope=identify\
+         &state={csrf_state}",
+        auth_config.client_id,
+    );
+
+    Redirect::temporary(&url)
+}
+
+/// `GET /api/auth/callback` — Handle Discord OAuth2 callback.
+pub async fn auth_callback(
+    State(state): State<AppState>,
+    Extension(auth_config): Extension<AuthConfig>,
+    Query(params): Query<CallbackParams>,
+) -> Result<Response, (StatusCode, Json<Value>)> {
+    // 1. Validate CSRF state and recover the origin used during login
+    let origin = state
+        .oauth_state_store
+        .validate(&params.state)
+        .ok_or_else(|| {
+            warn!("OAuth callback with invalid CSRF state");
+            (
+                StatusCode::BAD_REQUEST,
+                Json(json!({ "error": "Invalid OAuth state" })),
+            )
+        })?;
+
+    // 2. Exchange authorization code for access token
+    let redirect_uri = format!("{origin}{CALLBACK_PATH}");
+    let client = reqwest::Client::new();
+    let token_response = client
+        .post("https://discord.com/api/oauth2/token")
+        .form(&[
+            ("client_id", auth_config.client_id.as_str()),
+            ("client_secret", auth_config.client_secret.as_str()),
+            ("grant_type", "authorization_code"),
+            ("code", params.code.as_str()),
+            ("redirect_uri", redirect_uri.as_str()),
+        ])
+        .send()
+        .await
+        .map_err(|e| {
+            error!(error = %e, "failed to exchange OAuth code for token");
+            (
+                StatusCode::BAD_GATEWAY,
+                Json(json!({ "error": "Failed to exchange code with Discord" })),
+            )
+        })?;
+
+    if !token_response.status().is_success() {
+        let status = token_response.status();
+        let body = token_response.text().await.unwrap_or_default();
+        error!(%status, %body, "Discord token exchange returned error");
+        return Err((
+            StatusCode::BAD_GATEWAY,
+            Json(json!({ "error": "Discord token exchange failed" })),
+        ));
+    }
+
+    let token_data: TokenResponse = token_response.json().await.map_err(|e| {
+        error!(error = %e, "failed to parse Discord token response");
+        (
+            StatusCode::BAD_GATEWAY,
+            Json(json!({ "error": "Invalid token response from Discord" })),
+        )
+    })?;
+
+    // 3. Fetch Discord user profile
+    let discord_user: DiscordUser = client
+        .get("https://discord.com/api/users/@me")
+        .bearer_auth(&token_data.access_token)
+        .send()
+        .await
+        .map_err(|e| {
+            error!(error = %e, "failed to fetch Discord user profile");
+            (
+                StatusCode::BAD_GATEWAY,
+                Json(json!({ "error": "Failed to fetch Discord profile" })),
+            )
+        })?
+        .json()
+        .await
+        .map_err(|e| {
+            error!(error = %e, "failed to parse Discord user profile");
+            (
+                StatusCode::BAD_GATEWAY,
+                Json(json!({ "error": "Invalid user profile from Discord" })),
+            )
+        })?;
+
+    let discord_id: i64 = discord_user.id.parse().map_err(|_| {
+        error!(id = %discord_user.id, "Discord user ID is not a valid i64");
+        (
+            StatusCode::BAD_GATEWAY,
+            Json(json!({ "error": "Invalid Discord user ID" })),
+        )
+    })?;
+
+    // 4. Upsert user
+    let user = crate::data::users::upsert_user(
+        &state.db_pool,
+        discord_id,
+        &discord_user.username,
+        discord_user.avatar.as_deref(),
+    )
+    .await
+    .map_err(|e| {
+        error!(error = %e, "failed to upsert user");
+        (
+            StatusCode::INTERNAL_SERVER_ERROR,
+            Json(json!({ "error": "Database error" })),
+        )
+    })?;
+
+    info!(discord_id, username = %user.discord_username, "user authenticated via OAuth");
+
+    // 5. Create session
+    let session = crate::data::sessions::create_session(
+        &state.db_pool,
+        discord_id,
+        Duration::from_secs(7 * 24 * 3600),
+    )
+    .await
+    .map_err(|e| {
+        error!(error = %e, "failed to create session");
+        (
+            StatusCode::INTERNAL_SERVER_ERROR,
+            Json(json!({ "error": "Failed to create session" })),
+        )
+    })?;
+
+    // 6. Build response with session cookie
+    let secure = redirect_uri.starts_with("https://");
+    let cookie = session_cookie(&session.id, 604800, secure);
+
+    let redirect_to = if user.is_admin { "/admin" } else { "/" };
+
+    Ok((
+        [(header::SET_COOKIE, cookie)],
+        Redirect::temporary(redirect_to),
+    )
+        .into_response())
+}
+
+/// `POST /api/auth/logout` — Destroy the current session.
+pub async fn auth_logout(State(state): State<AppState>, headers: HeaderMap) -> Response {
+    if let Some(token) = extract_session_token(&headers) {
+        if let Err(e) = crate::data::sessions::delete_session(&state.db_pool, &token).await {
+            warn!(error = %e, "failed to delete session from database");
+        }
+        state.session_cache.evict(&token);
+    }
+
+    let cookie = session_cookie("", 0, false);
+
+    (
+        StatusCode::OK,
+        [(header::SET_COOKIE, cookie)],
+        Json(json!({ "ok": true })),
+    )
+        .into_response()
+}
+
+/// `GET /api/auth/me` — Return the current authenticated user's info.
+pub async fn auth_me(
+    State(state): State<AppState>,
+    headers: HeaderMap,
+) -> Result<Json<Value>, StatusCode> {
+    let token = extract_session_token(&headers).ok_or(StatusCode::UNAUTHORIZED)?;
+
+    let user = state
+        .session_cache
+        .get_user(&token)
+        .await
+        .ok_or(StatusCode::UNAUTHORIZED)?;
+
+    Ok(Json(json!({
+        "discordId": user.discord_id.to_string(),
+        "username": user.discord_username,
+        "avatarHash": user.discord_avatar_hash,
+        "isAdmin": user.is_admin,
+    })))
+}
@@ -0,0 +1,136 @@
+//! Web API endpoints for calendar export (ICS download + Google Calendar redirect).
+
+use axum::{
+    extract::{Path, State},
+    http::{StatusCode, header},
+    response::{IntoResponse, Redirect, Response},
+};
+
+use crate::calendar::{CalendarCourse, generate_gcal_url, generate_ics};
+use crate::data::models::DbMeetingTime;
+use crate::state::AppState;
+
+/// Fetch course + meeting times, build a `CalendarCourse`.
+async fn load_calendar_course(
+    state: &AppState,
+    term: &str,
+    crn: &str,
+) -> Result<(CalendarCourse, Vec<DbMeetingTime>), (StatusCode, String)> {
+    let course = crate::data::courses::get_course_by_crn(&state.db_pool, crn, term)
+        .await
+        .map_err(|e| {
+            tracing::error!(error = %e, "Calendar: course lookup failed");
+            (
+                StatusCode::INTERNAL_SERVER_ERROR,
+                "Lookup failed".to_string(),
+            )
+        })?
+        .ok_or_else(|| (StatusCode::NOT_FOUND, "Course not found".to_string()))?;
+
+    let instructors = crate::data::courses::get_course_instructors(&state.db_pool, course.id)
+        .await
+        .unwrap_or_default();
+
+    let primary_instructor = instructors
+        .iter()
+        .find(|i| i.is_primary)
+        .or(instructors.first())
+        .map(|i| i.display_name.clone());
+
+    let meeting_times: Vec<DbMeetingTime> =
+        serde_json::from_value(course.meeting_times.clone()).unwrap_or_default();
+
+    let cal_course = CalendarCourse {
+        crn: course.crn.clone(),
+        subject: course.subject.clone(),
+        course_number: course.course_number.clone(),
+        title: course.title.clone(),
+        sequence_number: course.sequence_number.clone(),
+        primary_instructor,
+    };
+
+    Ok((cal_course, meeting_times))
+}
+
+/// `GET /api/courses/{term}/{crn}/calendar.ics`
+///
+/// Returns an ICS file download for the course.
+pub async fn course_ics(
+    State(state): State<AppState>,
+    Path((term, crn)): Path<(String, String)>,
+) -> Result<Response, (StatusCode, String)> {
+    let (cal_course, meeting_times) = load_calendar_course(&state, &term, &crn).await?;
+
+    if meeting_times.is_empty() {
+        return Err((
+            StatusCode::NOT_FOUND,
+            "No meeting times found for this course".to_string(),
+        ));
+    }
+
+    let result = generate_ics(&cal_course, &meeting_times).map_err(|e| {
+        tracing::error!(error = %e, "ICS generation failed");
+        (
+            StatusCode::INTERNAL_SERVER_ERROR,
+            "Failed to generate ICS file".to_string(),
+        )
+    })?;
+
+    let response = (
+        [
+            (header::CONTENT_TYPE, "text/calendar; charset=utf-8"),
+            (
+                header::CONTENT_DISPOSITION,
+                &format!("attachment; filename=\"{}\"", result.filename),
+            ),
+            (header::CACHE_CONTROL, "no-cache"),
+        ],
+        result.content,
+    )
+        .into_response();
+
+    Ok(response)
+}
+
+/// `GET /api/courses/{term}/{crn}/gcal`
+///
+/// Redirects to Google Calendar with a pre-filled event for the first meeting time.
+/// If multiple meeting times exist, uses the first one with scheduled days/times.
+pub async fn course_gcal(
+    State(state): State<AppState>,
+    Path((term, crn)): Path<(String, String)>,
+) -> Result<Response, (StatusCode, String)> {
+    let (cal_course, meeting_times) = load_calendar_course(&state, &term, &crn).await?;
+
+    if meeting_times.is_empty() {
+        return Err((
+            StatusCode::NOT_FOUND,
+            "No meeting times found for this course".to_string(),
+        ));
+    }
+
+    // Prefer the first meeting time that has actual days/times scheduled
+    let mt = meeting_times
+        .iter()
+        .find(|mt| {
+            mt.begin_time.is_some()
+                && (mt.monday
+                    || mt.tuesday
+                    || mt.wednesday
+                    || mt.thursday
+                    || mt.friday
+                    || mt.saturday
+                    || mt.sunday)
+        })
+        .unwrap_or(&meeting_times[0]);
+
+    let url = generate_gcal_url(&cal_course, mt).map_err(|e| {
+        tracing::error!(error = %e, "Google Calendar URL generation failed");
+        (
+            StatusCode::INTERNAL_SERVER_ERROR,
+            "Failed to generate Google Calendar URL".to_string(),
+        )
+    })?;
+
+    Ok(Redirect::temporary(&url).into_response())
+}
@@ -0,0 +1,196 @@
+//! Content encoding negotiation for pre-compressed asset serving.
+//!
+//! Parses Accept-Encoding headers with quality values and returns
+//! supported encodings in priority order for content negotiation.
+
+use axum::http::{HeaderMap, HeaderValue, header};
+
+/// Minimum size threshold for compression (bytes).
+///
+/// Must match `MIN_SIZE` in `web/scripts/compress-assets.ts`.
+pub const COMPRESSION_MIN_SIZE: usize = 512;
+
+/// Supported content encodings in priority order (best compression first).
+#[derive(Debug, Clone, Copy, PartialEq, Eq, Hash)]
+pub enum ContentEncoding {
+    Zstd,
+    Brotli,
+    Gzip,
+    Identity,
+}
+
+impl ContentEncoding {
+    /// File extension suffix for pre-compressed variant lookup.
+    #[inline]
+    pub fn extension(&self) -> &'static str {
+        match self {
+            Self::Zstd => ".zst",
+            Self::Brotli => ".br",
+            Self::Gzip => ".gz",
+            Self::Identity => "",
+        }
+    }
+
+    /// `Content-Encoding` header value, or `None` for identity.
+    #[inline]
+    pub fn header_value(&self) -> Option<HeaderValue> {
+        match self {
+            Self::Zstd => Some(HeaderValue::from_static("zstd")),
+            Self::Brotli => Some(HeaderValue::from_static("br")),
+            Self::Gzip => Some(HeaderValue::from_static("gzip")),
+            Self::Identity => None,
+        }
+    }
+
+    /// Default priority when quality values are equal (higher = better).
+    #[inline]
+    fn default_priority(&self) -> u8 {
+        match self {
+            Self::Zstd => 4,
+            Self::Brotli => 3,
+            Self::Gzip => 2,
+            Self::Identity => 1,
+        }
+    }
+}
+
+/// Parse `Accept-Encoding` header and return supported encodings in priority order.
+///
+/// Supports quality values: `Accept-Encoding: gzip;q=0.8, br;q=1.0, zstd`
+/// When quality values are equal: zstd > brotli > gzip > identity.
+/// Encodings with `q=0` are excluded.
+pub fn parse_accepted_encodings(headers: &HeaderMap) -> Vec<ContentEncoding> {
+    let Some(accept) = headers
+        .get(header::ACCEPT_ENCODING)
+        .and_then(|v| v.to_str().ok())
+    else {
+        return vec![ContentEncoding::Identity];
+    };
+
+    let mut encodings: Vec<(ContentEncoding, f32)> = Vec::new();
+
+    for part in accept.split(',') {
+        let part = part.trim();
+        if part.is_empty() {
+            continue;
+        }
+
+        let (encoding_str, quality) = if let Some((enc, params)) = part.split_once(';') {
+            let q = params
+                .split(';')
+                .find_map(|p| p.trim().strip_prefix("q="))
+                .and_then(|q| q.parse::<f32>().ok())
+                .unwrap_or(1.0);
+            (enc.trim(), q)
+        } else {
+            (part, 1.0)
+        };
+
+        if quality == 0.0 {
+            continue;
+        }
+
+        let encoding = match encoding_str.to_lowercase().as_str() {
+            "zstd" => ContentEncoding::Zstd,
+            "br" | "brotli" => ContentEncoding::Brotli,
+            "gzip" | "x-gzip" => ContentEncoding::Gzip,
+            "*" => ContentEncoding::Gzip,
+            "identity" => ContentEncoding::Identity,
+            _ => continue,
+        };
+
+        encodings.push((encoding, quality));
+    }
+
+    // Sort by quality (desc), then default priority (desc)
+    encodings.sort_by(|a, b| {
+        b.1.partial_cmp(&a.1)
+            .unwrap_or(std::cmp::Ordering::Equal)
+            .then_with(|| b.0.default_priority().cmp(&a.0.default_priority()))
+    });
+
+    if encodings.is_empty() {
+        vec![ContentEncoding::Identity]
+    } else {
+        encodings.into_iter().map(|(e, _)| e).collect()
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn test_parse_all_encodings() {
+        let mut headers = HeaderMap::new();
+        headers.insert(header::ACCEPT_ENCODING, "gzip, br, zstd".parse().unwrap());
+        let encodings = parse_accepted_encodings(&headers);
+        assert_eq!(encodings[0], ContentEncoding::Zstd);
+        assert_eq!(encodings[1], ContentEncoding::Brotli);
+        assert_eq!(encodings[2], ContentEncoding::Gzip);
+    }
+
+    #[test]
+    fn test_parse_with_quality_values() {
+        let mut headers = HeaderMap::new();
+        headers.insert(
+            header::ACCEPT_ENCODING,
+            "gzip;q=1.0, br;q=0.5, zstd;q=0.8".parse().unwrap(),
+        );
+        let encodings = parse_accepted_encodings(&headers);
+        assert_eq!(encodings[0], ContentEncoding::Gzip);
+        assert_eq!(encodings[1], ContentEncoding::Zstd);
+        assert_eq!(encodings[2], ContentEncoding::Brotli);
+    }
+
+    #[test]
+    fn test_no_header_returns_identity() {
+        let headers = HeaderMap::new();
+        let encodings = parse_accepted_encodings(&headers);
+        assert_eq!(encodings, vec![ContentEncoding::Identity]);
+    }
+
+    #[test]
+    fn test_disabled_encoding_excluded() {
+        let mut headers = HeaderMap::new();
+        headers.insert(
+            header::ACCEPT_ENCODING,
+            "zstd;q=0, br, gzip".parse().unwrap(),
+        );
+        let encodings = parse_accepted_encodings(&headers);
+        assert_eq!(encodings[0], ContentEncoding::Brotli);
+        assert_eq!(encodings[1], ContentEncoding::Gzip);
+        assert!(!encodings.contains(&ContentEncoding::Zstd));
+    }
+
+    #[test]
+    fn test_real_chrome_header() {
+        let mut headers = HeaderMap::new();
+        headers.insert(
+            header::ACCEPT_ENCODING,
+            "gzip, deflate, br, zstd".parse().unwrap(),
+        );
+        assert_eq!(parse_accepted_encodings(&headers)[0], ContentEncoding::Zstd);
+    }
+
+    #[test]
+    fn test_extensions() {
+        assert_eq!(ContentEncoding::Zstd.extension(), ".zst");
+        assert_eq!(ContentEncoding::Brotli.extension(), ".br");
+        assert_eq!(ContentEncoding::Gzip.extension(), ".gz");
+        assert_eq!(ContentEncoding::Identity.extension(), "");
+    }
+
+    #[test]
+    fn test_header_values() {
+        assert_eq!(
+            ContentEncoding::Zstd.header_value().unwrap(),
+            HeaderValue::from_static("zstd")
+        );
+        assert_eq!(
+            ContentEncoding::Brotli.header_value().unwrap(),
+            HeaderValue::from_static("br")
+        );
+        assert!(ContentEncoding::Identity.header_value().is_none());
+    }
+}
@@ -0,0 +1,74 @@
+//! Axum extractors for authentication and authorization.
+
+use axum::extract::FromRequestParts;
+use axum::http::{StatusCode, header};
+use axum::response::Json;
+use http::request::Parts;
+use serde_json::json;
+
+use crate::data::models::User;
+use crate::state::AppState;
+
+/// Extractor that resolves the session cookie to an authenticated [`User`].
+///
+/// Returns 401 if no valid session cookie is present.
+pub struct AuthUser(pub User);
+
+impl FromRequestParts<AppState> for AuthUser {
+    type Rejection = (StatusCode, Json<serde_json::Value>);
+
+    async fn from_request_parts(
+        parts: &mut Parts,
+        state: &AppState,
+    ) -> Result<Self, Self::Rejection> {
+        let token = parts
+            .headers
+            .get(header::COOKIE)
+            .and_then(|v| v.to_str().ok())
+            .and_then(|cookies| {
+                cookies
+                    .split(';')
+                    .find_map(|c| c.trim().strip_prefix("session=").map(|v| v.to_owned()))
+            })
+            .ok_or_else(|| {
+                (
+                    StatusCode::UNAUTHORIZED,
+                    Json(json!({"error": "unauthorized", "message": "No session cookie"})),
+                )
+            })?;
+
+        let user = state.session_cache.get_user(&token).await.ok_or_else(|| {
+            (
+                StatusCode::UNAUTHORIZED,
+                Json(json!({"error": "unauthorized", "message": "Invalid or expired session"})),
+            )
+        })?;
+
+        Ok(AuthUser(user))
+    }
+}
+
+/// Extractor that requires an authenticated admin user.
+///
+/// Returns 401 if not authenticated, 403 if not admin.
+pub struct AdminUser(pub User);
+
+impl FromRequestParts<AppState> for AdminUser {
+    type Rejection = (StatusCode, Json<serde_json::Value>);
+
+    async fn from_request_parts(
+        parts: &mut Parts,
+        state: &AppState,
+    ) -> Result<Self, Self::Rejection> {
+        let AuthUser(user) = AuthUser::from_request_parts(parts, state).await?;
+
+        if !user.is_admin {
+            return Err((
+                StatusCode::FORBIDDEN,
+                Json(json!({"error": "forbidden", "message": "Admin access required"})),
+            ));
+        }
+
+        Ok(AdminUser(user))
+    }
+}
@@ -1,7 +1,19 @@
 //! Web API module for the banner application.

+pub mod admin;
+pub mod admin_rmp;
+pub mod admin_scraper;
 #[cfg(feature = "embed-assets")]
 pub mod assets;
+pub mod auth;
+pub mod calendar;
+#[cfg(feature = "embed-assets")]
+pub mod encoding;
+pub mod extractors;
 pub mod routes;
+pub mod schedule_cache;
+pub mod session_cache;
+pub mod timeline;
+pub mod ws;

 pub use routes::*;
@@ -1,79 +1,119 @@
 //! Web API endpoints for Banner bot monitoring and metrics.

 use axum::{
-    Router,
+    Extension, Router,
    body::Body,
-    extract::{Request, State},
+    extract::{Path, Query, Request, State},
+    http::StatusCode as AxumStatusCode,
    response::{Json, Response},
-    routing::get,
+    routing::{get, post, put},
 };
+
+use crate::web::admin;
+use crate::web::admin_rmp;
+use crate::web::admin_scraper;
+use crate::web::auth::{self, AuthConfig};
+use crate::web::calendar;
+use crate::web::timeline;
+use crate::web::ws;
 #[cfg(feature = "embed-assets")]
 use axum::{
-    http::{HeaderMap, HeaderValue, StatusCode, Uri},
-    response::{Html, IntoResponse},
+    http::{HeaderMap, StatusCode, Uri},
+    response::IntoResponse,
 };
-#[cfg(feature = "embed-assets")]
-use http::header;
-use serde::Serialize;
+use serde::{Deserialize, Serialize};
 use serde_json::{Value, json};
 use std::{collections::BTreeMap, time::Duration};
+use ts_rs::TS;

 use crate::state::AppState;
 use crate::status::ServiceStatus;
 #[cfg(not(feature = "embed-assets"))]
 use tower_http::cors::{Any, CorsLayer};
-use tower_http::{classify::ServerErrorsFailureClass, timeout::TimeoutLayer, trace::TraceLayer};
+use tower_http::{
+    classify::ServerErrorsFailureClass, compression::CompressionLayer, timeout::TimeoutLayer,
+    trace::TraceLayer,
+};
 use tracing::{Span, debug, trace, warn};

 #[cfg(feature = "embed-assets")]
-use crate::web::assets::{WebAssets, get_asset_metadata_cached};
-
-/// Set appropriate caching headers based on asset type
-#[cfg(feature = "embed-assets")]
-fn set_caching_headers(response: &mut Response, path: &str, etag: &str) {
-    let headers = response.headers_mut();
-
-    // Set ETag
-    if let Ok(etag_value) = HeaderValue::from_str(etag) {
-        headers.insert(header::ETAG, etag_value);
-    }
-
-    // Set Cache-Control based on asset type
-    let cache_control = if path.starts_with("assets/") {
-        // Static assets with hashed filenames - long-term cache
-        "public, max-age=31536000, immutable"
-    } else if path == "index.html" {
-        // HTML files - short-term cache
-        "public, max-age=300"
-    } else {
-        match path.split_once('.').map(|(_, extension)| extension) {
-            Some(ext) => match ext {
-                // CSS/JS files - medium-term cache
-                "css" | "js" => "public, max-age=86400",
-                // Images - long-term cache
-                "png" | "jpg" | "jpeg" | "gif" | "svg" | "ico" => "public, max-age=2592000",
-                // Default for other files
-                _ => "public, max-age=3600",
-            },
-            // Default for files without an extension
-            None => "public, max-age=3600",
-        }
-    };
-
-    if let Ok(cache_control_value) = HeaderValue::from_str(cache_control) {
-        headers.insert(header::CACHE_CONTROL, cache_control_value);
-    }
-}
+use crate::web::assets::try_serve_asset_with_encoding;

 /// Creates the web server router
-pub fn create_router(app_state: AppState) -> Router {
+pub fn create_router(app_state: AppState, auth_config: AuthConfig) -> Router {
    let api_router = Router::new()
        .route("/health", get(health))
        .route("/status", get(status))
        .route("/metrics", get(metrics))
+        .route("/courses/search", get(search_courses))
+        .route("/courses/{term}/{crn}", get(get_course))
+        .route(
+            "/courses/{term}/{crn}/calendar.ics",
+            get(calendar::course_ics),
+        )
+        .route("/courses/{term}/{crn}/gcal", get(calendar::course_gcal))
+        .route("/terms", get(get_terms))
+        .route("/subjects", get(get_subjects))
+        .route("/reference/{category}", get(get_reference))
+        .route("/timeline", post(timeline::timeline))
+        .with_state(app_state.clone());
+
+    let auth_router = Router::new()
+        .route("/auth/login", get(auth::auth_login))
+        .route("/auth/callback", get(auth::auth_callback))
+        .route("/auth/logout", post(auth::auth_logout))
+        .route("/auth/me", get(auth::auth_me))
+        .layer(Extension(auth_config))
+        .with_state(app_state.clone());
+
+    let admin_router = Router::new()
+        .route("/admin/status", get(admin::admin_status))
+        .route("/admin/users", get(admin::list_users))
+        .route(
+            "/admin/users/{discord_id}/admin",
+            put(admin::set_user_admin),
+        )
+        .route("/admin/scrape-jobs", get(admin::list_scrape_jobs))
+        .route("/admin/scrape-jobs/ws", get(ws::scrape_jobs_ws))
+        .route("/admin/audit-log", get(admin::list_audit_log))
+        .route("/admin/instructors", get(admin_rmp::list_instructors))
+        .route("/admin/instructors/{id}", get(admin_rmp::get_instructor))
+        .route(
+            "/admin/instructors/{id}/match",
+            post(admin_rmp::match_instructor),
+        )
+        .route(
+            "/admin/instructors/{id}/reject-candidate",
+            post(admin_rmp::reject_candidate),
+        )
+        .route(
+            "/admin/instructors/{id}/reject-all",
+            post(admin_rmp::reject_all),
+        )
+        .route(
+            "/admin/instructors/{id}/unmatch",
+            post(admin_rmp::unmatch_instructor),
+        )
+        .route("/admin/rmp/rescore", post(admin_rmp::rescore))
+        .route("/admin/scraper/stats", get(admin_scraper::scraper_stats))
+        .route(
+            "/admin/scraper/timeseries",
+            get(admin_scraper::scraper_timeseries),
+        )
+        .route(
+            "/admin/scraper/subjects",
+            get(admin_scraper::scraper_subjects),
+        )
+        .route(
+            "/admin/scraper/subjects/{subject}",
+            get(admin_scraper::scraper_subject_detail),
+        )
        .with_state(app_state);

-    let mut router = Router::new().nest("/api", api_router);
+    let mut router = Router::new()
+        .nest("/api", api_router)
+        .nest("/api", auth_router)
+        .nest("/api", admin_router);

    // When embed-assets feature is enabled, serve embedded static assets
    #[cfg(feature = "embed-assets")]
@@ -93,6 +133,13 @@ pub fn create_router(app_state: AppState) -> Router {
    }

    router.layer((
+        // Compress API responses (gzip/brotli/zstd). Pre-compressed static
+        // assets already have Content-Encoding set, so tower-http skips them.
+        CompressionLayer::new()
+            .zstd(true)
+            .br(true)
+            .gzip(true)
+            .quality(tower_http::CompressionLevel::Fastest),
        TraceLayer::new_for_http()
            .make_span_with(|request: &Request<Body>| {
                tracing::debug_span!("request", path = request.uri().path())
@@ -139,71 +186,35 @@ pub fn create_router(app_state: AppState) -> Router {
    ))
 }

-/// Handler that extracts request information for caching
+/// SPA fallback handler with content encoding negotiation.
+///
+/// Serves embedded static assets with pre-compressed variants when available,
+/// falling back to `index.html` for SPA client-side routing.
 #[cfg(feature = "embed-assets")]
-async fn fallback(request: Request) -> Response {
+async fn fallback(request: Request) -> axum::response::Response {
    let uri = request.uri().clone();
    let headers = request.headers().clone();
-    handle_spa_fallback_with_headers(uri, headers).await
+    handle_spa_fallback(uri, headers).await
 }

-/// Handles SPA routing by serving index.html for non-API, non-asset requests
-/// This version includes HTTP caching headers and ETag support
 #[cfg(feature = "embed-assets")]
-async fn handle_spa_fallback_with_headers(uri: Uri, request_headers: HeaderMap) -> Response {
-    let path = uri.path().trim_start_matches('/');
-
-    if let Some(content) = WebAssets::get(path) {
-        // Get asset metadata (MIME type and hash) with caching
-        let metadata = get_asset_metadata_cached(path, &content.data);
-
-        // Check if client has a matching ETag (conditional request)
-        if let Some(etag) = request_headers.get(header::IF_NONE_MATCH)
-            && etag.to_str().is_ok_and(|s| metadata.etag_matches(s))
-        {
-            return StatusCode::NOT_MODIFIED.into_response();
-        }
-
-        // Use cached MIME type, only set Content-Type if we have a valid MIME type
-        let mut response = (
-            [(
-                header::CONTENT_TYPE,
-                // For unknown types, set to application/octet-stream
-                metadata
-                    .mime_type
-                    .unwrap_or("application/octet-stream".to_string()),
-            )],
-            content.data,
-        )
-            .into_response();
-
-        // Set caching headers
-        set_caching_headers(&mut response, path, &metadata.hash.quoted());
+async fn handle_spa_fallback(uri: Uri, request_headers: HeaderMap) -> axum::response::Response {
+    let path = uri.path();

+    // Try serving the exact asset (with encoding negotiation)
+    if let Some(response) = try_serve_asset_with_encoding(path, &request_headers) {
        return response;
-    } else {
-        // Any assets that are not found should be treated as a 404, not falling back to the SPA index.html
-        if path.starts_with("assets/") {
-            return (StatusCode::NOT_FOUND, "Asset not found").into_response();
-        }
    }

-    // Fall back to the SPA index.html
-    match WebAssets::get("index.html") {
-        Some(content) => {
-            let metadata = get_asset_metadata_cached("index.html", &content.data);
+    // SvelteKit assets under _app/ that don't exist are a hard 404
+    let trimmed = path.trim_start_matches('/');
+    if trimmed.starts_with("_app/") || trimmed.starts_with("assets/") {
+        return (StatusCode::NOT_FOUND, "Asset not found").into_response();
+    }

-            // Check if client has a matching ETag for index.html
-            if let Some(etag) = request_headers.get(header::IF_NONE_MATCH)
-                && etag.to_str().is_ok_and(|s| metadata.etag_matches(s))
-            {
-                return StatusCode::NOT_MODIFIED.into_response();
-            }
-
-            let mut response = Html(content.data).into_response();
-            set_caching_headers(&mut response, "index.html", &metadata.hash.quoted());
-            response
-        }
+    // SPA fallback: serve index.html with encoding negotiation
+    match try_serve_asset_with_encoding("/index.html", &request_headers) {
+        Some(response) => response,
        None => (
            StatusCode::INTERNAL_SERVER_ERROR,
            "Failed to load index.html",
@@ -221,14 +232,16 @@ async fn health() -> Json<Value> {
    }))
 }

-#[derive(Serialize)]
-struct ServiceInfo {
+#[derive(Serialize, TS)]
+#[ts(export)]
+pub struct ServiceInfo {
    name: String,
    status: ServiceStatus,
 }

-#[derive(Serialize)]
-struct StatusResponse {
+#[derive(Serialize, TS)]
+#[ts(export)]
+pub struct StatusResponse {
    status: ServiceStatus,
    version: String,
    commit: String,
@@ -249,7 +262,10 @@ async fn status(State(state): State<AppState>) -> Json<StatusResponse> {
        );
    }

-    let overall_status = if services.values().any(|s| matches!(s.status, ServiceStatus::Error)) {
+    let overall_status = if services
+        .values()
+        .any(|s| matches!(s.status, ServiceStatus::Error))
+    {
        ServiceStatus::Error
    } else if !services.is_empty()
        && services
@@ -272,12 +288,441 @@ async fn status(State(state): State<AppState>) -> Json<StatusResponse> {
 }

 /// Metrics endpoint for monitoring
-async fn metrics() -> Json<Value> {
-    // For now, return basic metrics structure
-    Json(json!({
-        "banner_api": {
-            "status": "connected"
+async fn metrics(
+    State(state): State<AppState>,
+    Query(params): Query<MetricsParams>,
+) -> Result<Json<Value>, (AxumStatusCode, String)> {
+    let limit = params.limit.clamp(1, 5000);
+
+    // Parse range shorthand, defaulting to 24h
+    let range_str = params.range.as_deref().unwrap_or("24h");
+    let duration = match range_str {
+        "1h" => chrono::Duration::hours(1),
+        "6h" => chrono::Duration::hours(6),
+        "24h" => chrono::Duration::hours(24),
+        "7d" => chrono::Duration::days(7),
+        "30d" => chrono::Duration::days(30),
+        _ => {
+            return Err((
+                AxumStatusCode::BAD_REQUEST,
+                format!("Invalid range '{range_str}'. Valid: 1h, 6h, 24h, 7d, 30d"),
+            ));
+        }
+    };
+    let since = chrono::Utc::now() - duration;
+
+    // Resolve course_id: explicit param takes priority, then term+crn lookup
+    let course_id = if let Some(id) = params.course_id {
+        Some(id)
+    } else if let (Some(term), Some(crn)) = (params.term.as_deref(), params.crn.as_deref()) {
+        let row: Option<(i32,)> =
+            sqlx::query_as("SELECT id FROM courses WHERE term_code = $1 AND crn = $2")
+                .bind(term)
+                .bind(crn)
+                .fetch_optional(&state.db_pool)
+                .await
+                .map_err(|e| {
+                    tracing::error!(error = %e, "Course lookup for metrics failed");
+                    (
+                        AxumStatusCode::INTERNAL_SERVER_ERROR,
+                        "Course lookup failed".to_string(),
+                    )
+                })?;
+        row.map(|(id,)| id)
+    } else {
+        None
+    };
+
+    // Build query dynamically based on filters
+    let metrics: Vec<(i32, i32, chrono::DateTime<chrono::Utc>, i32, i32, i32)> =
+        if let Some(cid) = course_id {
+            sqlx::query_as(
+                "SELECT id, course_id, timestamp, enrollment, wait_count, seats_available \
+             FROM course_metrics \
+             WHERE course_id = $1 AND timestamp >= $2 \
+             ORDER BY timestamp DESC \
+             LIMIT $3",
+            )
+            .bind(cid)
+            .bind(since)
+            .bind(limit)
+            .fetch_all(&state.db_pool)
+            .await
+        } else {
+            sqlx::query_as(
+                "SELECT id, course_id, timestamp, enrollment, wait_count, seats_available \
+             FROM course_metrics \
+             WHERE timestamp >= $1 \
+             ORDER BY timestamp DESC \
+             LIMIT $2",
+            )
+            .bind(since)
+            .bind(limit)
+            .fetch_all(&state.db_pool)
+            .await
+        }
+        .map_err(|e| {
+            tracing::error!(error = %e, "Metrics query failed");
+            (
+                AxumStatusCode::INTERNAL_SERVER_ERROR,
+                "Metrics query failed".to_string(),
+            )
+        })?;
+
+    let count = metrics.len();
+    let metrics_json: Vec<Value> = metrics
+        .into_iter()
+        .map(
+            |(id, course_id, timestamp, enrollment, wait_count, seats_available)| {
+                json!({
+                    "id": id,
+                    "courseId": course_id,
+                    "timestamp": timestamp.to_rfc3339(),
+                    "enrollment": enrollment,
+                    "waitCount": wait_count,
+                    "seatsAvailable": seats_available,
+                })
+            },
+        )
+        .collect();
+
+    Ok(Json(json!({
+        "metrics": metrics_json,
+        "count": count,
+        "timestamp": chrono::Utc::now().to_rfc3339(),
+    })))
+}
+
+// ============================================================
+// Course search & detail API
+// ============================================================
+
+#[derive(Deserialize)]
+struct MetricsParams {
+    course_id: Option<i32>,
+    term: Option<String>,
+    crn: Option<String>,
+    /// Shorthand durations: "1h", "6h", "24h", "7d", "30d"
+    range: Option<String>,
+    #[serde(default = "default_metrics_limit")]
+    limit: i32,
+}
+
+fn default_metrics_limit() -> i32 {
+    500
+}
+
+#[derive(Deserialize)]
+struct SubjectsParams {
+    term: String,
+}
+
+#[derive(Deserialize)]
+struct SearchParams {
+    term: String,
+    #[serde(default)]
+    subject: Vec<String>,
+    q: Option<String>,
+    course_number_low: Option<i32>,
+    course_number_high: Option<i32>,
+    #[serde(default)]
+    open_only: bool,
+    instructional_method: Option<String>,
+    campus: Option<String>,
+    #[serde(default = "default_limit")]
+    limit: i32,
+    #[serde(default)]
+    offset: i32,
+    sort_by: Option<SortColumn>,
+    sort_dir: Option<SortDirection>,
+}
+
+use crate::data::courses::{SortColumn, SortDirection};
+
+fn default_limit() -> i32 {
+    25
+}
+
+#[derive(Serialize, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export)]
+pub struct CourseResponse {
+    crn: String,
+    subject: String,
+    course_number: String,
+    title: String,
+    term_code: String,
+    sequence_number: Option<String>,
+    instructional_method: Option<String>,
+    campus: Option<String>,
+    enrollment: i32,
+    max_enrollment: i32,
+    wait_count: i32,
+    wait_capacity: i32,
+    credit_hours: Option<i32>,
+    credit_hour_low: Option<i32>,
+    credit_hour_high: Option<i32>,
+    cross_list: Option<String>,
+    cross_list_capacity: Option<i32>,
+    cross_list_count: Option<i32>,
+    link_identifier: Option<String>,
+    is_section_linked: Option<bool>,
+    part_of_term: Option<String>,
+    meeting_times: Vec<crate::data::models::DbMeetingTime>,
+    attributes: Vec<String>,
+    instructors: Vec<InstructorResponse>,
+}
+
+#[derive(Serialize, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export)]
+pub struct InstructorResponse {
+    instructor_id: i32,
+    banner_id: String,
+    display_name: String,
+    email: String,
+    is_primary: bool,
+    rmp_rating: Option<f32>,
+    rmp_num_ratings: Option<i32>,
+    rmp_legacy_id: Option<i32>,
+}
+
+#[derive(Serialize, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export)]
+pub struct SearchResponse {
+    courses: Vec<CourseResponse>,
+    total_count: i32,
+    offset: i32,
+    limit: i32,
+}
+
+#[derive(Serialize, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export)]
+pub struct CodeDescription {
+    code: String,
+    description: String,
+}
+
+/// Build a `CourseResponse` from a DB course with pre-fetched instructor details.
+fn build_course_response(
+    course: &crate::data::models::Course,
+    instructors: Vec<crate::data::models::CourseInstructorDetail>,
+) -> CourseResponse {
+    let instructors = instructors
+        .into_iter()
+        .map(|i| InstructorResponse {
+            instructor_id: i.instructor_id,
+            banner_id: i.banner_id,
+            display_name: i.display_name,
+            email: i.email,
+            is_primary: i.is_primary,
+            rmp_rating: i.avg_rating,
+            rmp_num_ratings: i.num_ratings,
+            rmp_legacy_id: i.rmp_legacy_id,
+        })
+        .collect();
+
+    CourseResponse {
+        crn: course.crn.clone(),
+        subject: course.subject.clone(),
+        course_number: course.course_number.clone(),
+        title: course.title.clone(),
+        term_code: course.term_code.clone(),
+        sequence_number: course.sequence_number.clone(),
+        instructional_method: course.instructional_method.clone(),
+        campus: course.campus.clone(),
+        enrollment: course.enrollment,
+        max_enrollment: course.max_enrollment,
+        wait_count: course.wait_count,
+        wait_capacity: course.wait_capacity,
+        credit_hours: course.credit_hours,
+        credit_hour_low: course.credit_hour_low,
+        credit_hour_high: course.credit_hour_high,
+        cross_list: course.cross_list.clone(),
+        cross_list_capacity: course.cross_list_capacity,
+        cross_list_count: course.cross_list_count,
+        link_identifier: course.link_identifier.clone(),
+        is_section_linked: course.is_section_linked,
+        part_of_term: course.part_of_term.clone(),
+        meeting_times: serde_json::from_value(course.meeting_times.clone()).unwrap_or_default(),
+        attributes: serde_json::from_value(course.attributes.clone()).unwrap_or_default(),
+        instructors,
+    }
+}
+
+/// `GET /api/courses/search`
+async fn search_courses(
+    State(state): State<AppState>,
+    axum_extra::extract::Query(params): axum_extra::extract::Query<SearchParams>,
+) -> Result<Json<SearchResponse>, (AxumStatusCode, String)> {
+    let limit = params.limit.clamp(1, 100);
+    let offset = params.offset.max(0);
+
+    let (courses, total_count) = crate::data::courses::search_courses(
+        &state.db_pool,
+        &params.term,
+        if params.subject.is_empty() {
+            None
+        } else {
+            Some(&params.subject)
        },
-        "timestamp": chrono::Utc::now().to_rfc3339()
+        params.q.as_deref(),
+        params.course_number_low,
+        params.course_number_high,
+        params.open_only,
+        params.instructional_method.as_deref(),
+        params.campus.as_deref(),
+        limit,
+        offset,
+        params.sort_by,
+        params.sort_dir,
+    )
+    .await
+    .map_err(|e| {
+        tracing::error!(error = %e, "Course search failed");
+        (
+            AxumStatusCode::INTERNAL_SERVER_ERROR,
+            "Search failed".to_string(),
+        )
+    })?;
+
+    // Batch-fetch all instructors in a single query instead of N+1
+    let course_ids: Vec<i32> = courses.iter().map(|c| c.id).collect();
+    let mut instructor_map =
+        crate::data::courses::get_instructors_for_courses(&state.db_pool, &course_ids)
+            .await
+            .unwrap_or_default();
+
+    let course_responses: Vec<CourseResponse> = courses
+        .iter()
+        .map(|course| {
+            let instructors = instructor_map.remove(&course.id).unwrap_or_default();
+            build_course_response(course, instructors)
+        })
+        .collect();
+
+    Ok(Json(SearchResponse {
+        courses: course_responses,
+        total_count: total_count as i32,
+        offset,
+        limit,
    }))
 }
+
+/// `GET /api/courses/:term/:crn`
+async fn get_course(
+    State(state): State<AppState>,
+    Path((term, crn)): Path<(String, String)>,
+) -> Result<Json<CourseResponse>, (AxumStatusCode, String)> {
+    let course = crate::data::courses::get_course_by_crn(&state.db_pool, &crn, &term)
+        .await
+        .map_err(|e| {
+            tracing::error!(error = %e, "Course lookup failed");
+            (
+                AxumStatusCode::INTERNAL_SERVER_ERROR,
+                "Lookup failed".to_string(),
+            )
+        })?
+        .ok_or_else(|| (AxumStatusCode::NOT_FOUND, "Course not found".to_string()))?;
+
+    let instructors = crate::data::courses::get_course_instructors(&state.db_pool, course.id)
+        .await
+        .unwrap_or_default();
+    Ok(Json(build_course_response(&course, instructors)))
+}
+
+/// `GET /api/terms`
+async fn get_terms(
+    State(state): State<AppState>,
+) -> Result<Json<Vec<CodeDescription>>, (AxumStatusCode, String)> {
+    let cache = state.reference_cache.read().await;
+    let term_codes = crate::data::courses::get_available_terms(&state.db_pool)
+        .await
+        .map_err(|e| {
+            tracing::error!(error = %e, "Failed to get terms");
+            (
+                AxumStatusCode::INTERNAL_SERVER_ERROR,
+                "Failed to get terms".to_string(),
+            )
+        })?;
+
+    let terms: Vec<CodeDescription> = term_codes
+        .into_iter()
+        .map(|code| {
+            let description = cache
+                .lookup("term", &code)
+                .unwrap_or("Unknown Term")
+                .to_string();
+            CodeDescription { code, description }
+        })
+        .collect();
+
+    Ok(Json(terms))
+}
+
+/// `GET /api/subjects?term=202620`
+async fn get_subjects(
+    State(state): State<AppState>,
+    Query(params): Query<SubjectsParams>,
+) -> Result<Json<Vec<CodeDescription>>, (AxumStatusCode, String)> {
+    let rows = crate::data::courses::get_subjects_by_enrollment(&state.db_pool, &params.term)
+        .await
+        .map_err(|e| {
+            tracing::error!(error = %e, "Failed to get subjects");
+            (
+                AxumStatusCode::INTERNAL_SERVER_ERROR,
+                "Failed to get subjects".to_string(),
+            )
+        })?;
+
+    let subjects: Vec<CodeDescription> = rows
+        .into_iter()
+        .map(|(code, description, _enrollment)| CodeDescription { code, description })
+        .collect();
+
+    Ok(Json(subjects))
+}
+
+/// `GET /api/reference/:category`
+async fn get_reference(
+    State(state): State<AppState>,
+    Path(category): Path<String>,
+) -> Result<Json<Vec<CodeDescription>>, (AxumStatusCode, String)> {
+    let cache = state.reference_cache.read().await;
+    let entries = cache.entries_for_category(&category);
+
+    if entries.is_empty() {
+        // Fall back to DB query in case cache doesn't have this category
+        drop(cache);
+        let rows = crate::data::reference::get_by_category(&category, &state.db_pool)
+            .await
+            .map_err(|e| {
+                tracing::error!(error = %e, category = %category, "Reference lookup failed");
+                (
+                    AxumStatusCode::INTERNAL_SERVER_ERROR,
+                    "Lookup failed".to_string(),
+                )
+            })?;
+
+        return Ok(Json(
+            rows.into_iter()
+                .map(|r| CodeDescription {
+                    code: r.code,
+                    description: r.description,
+                })
+                .collect(),
+        ));
+    }
+
+    Ok(Json(
+        entries
+            .into_iter()
+            .map(|(code, desc)| CodeDescription {
+                code: code.to_string(),
+                description: desc.to_string(),
+            })
+            .collect(),
+    ))
+}
@@ -0,0 +1,444 @@
+//! ISR-style schedule cache for timeline enrollment queries.
+//!
+//! Loads all courses with their meeting times from the database, parses the
+//! JSONB meeting times into a compact in-memory representation, and caches
+//! the result. The cache is refreshed in the background every hour using a
+//! stale-while-revalidate pattern with singleflight deduplication — readers
+//! always get the current cached value instantly, never blocking on a refresh.
+
+use chrono::NaiveDate;
+use serde_json::Value;
+use sqlx::PgPool;
+use std::sync::atomic::{AtomicBool, Ordering};
+use std::sync::Arc;
+use tokio::sync::watch;
+use tracing::{debug, error, info};
+
+/// How often the cache is considered fresh (1 hour).
+const REFRESH_INTERVAL: std::time::Duration = std::time::Duration::from_secs(60 * 60);
+
+// ── Compact schedule representation ─────────────────────────────────
+
+/// A single meeting time block, pre-parsed for fast filtering.
+#[derive(Debug, Clone)]
+pub(crate) struct ParsedSchedule {
+    /// Bitmask of days: bit 0 = Monday, bit 6 = Sunday.
+    days: u8,
+    /// Minutes since midnight for start (e.g. 600 = 10:00).
+    begin_minutes: u16,
+    /// Minutes since midnight for end (e.g. 650 = 10:50).
+    end_minutes: u16,
+    /// First day the meeting pattern is active.
+    start_date: NaiveDate,
+    /// Last day the meeting pattern is active.
+    end_date: NaiveDate,
+}
+
+/// A course with its enrollment and pre-parsed schedule blocks.
+#[derive(Debug, Clone)]
+pub(crate) struct CachedCourse {
+    pub(crate) subject: String,
+    pub(crate) enrollment: i32,
+    pub(crate) schedules: Vec<ParsedSchedule>,
+}
+
+/// The immutable snapshot of all courses, swapped atomically on refresh.
+#[derive(Debug, Clone)]
+pub(crate) struct ScheduleSnapshot {
+    pub(crate) courses: Vec<CachedCourse>,
+    refreshed_at: std::time::Instant,
+}
+
+// ── Cache handle ────────────────────────────────────────────────────
+
+/// Shared schedule cache. Clone-cheap (all `Arc`-wrapped internals).
+#[derive(Clone)]
+pub struct ScheduleCache {
+    /// Current snapshot, updated via `watch` channel for lock-free reads.
+    rx: watch::Receiver<Arc<ScheduleSnapshot>>,
+    /// Sender side, held to push new snapshots.
+    tx: Arc<watch::Sender<Arc<ScheduleSnapshot>>>,
+    /// Singleflight guard — true while a refresh task is in flight.
+    refreshing: Arc<AtomicBool>,
+    /// Database pool for refresh queries.
+    pool: PgPool,
+}
+
+impl ScheduleCache {
+    /// Create a new cache with an empty initial snapshot.
+    pub(crate) fn new(pool: PgPool) -> Self {
+        let empty = Arc::new(ScheduleSnapshot {
+            courses: Vec::new(),
+            refreshed_at: std::time::Instant::now(),
+        });
+        let (tx, rx) = watch::channel(empty);
+        Self {
+            rx,
+            tx: Arc::new(tx),
+            refreshing: Arc::new(AtomicBool::new(false)),
+            pool,
+        }
+    }
+
+    /// Get the current snapshot. Never blocks on refresh.
+    pub(crate) fn snapshot(&self) -> Arc<ScheduleSnapshot> {
+        self.rx.borrow().clone()
+    }
+
+    /// Check freshness and trigger a background refresh if stale.
+    /// Always returns immediately — the caller uses the current snapshot.
+    pub(crate) fn ensure_fresh(&self) {
+        let snap = self.rx.borrow();
+        if snap.refreshed_at.elapsed() < REFRESH_INTERVAL {
+            return;
+        }
+        // Singleflight: only one refresh at a time.
+        if self
+            .refreshing
+            .compare_exchange(false, true, Ordering::AcqRel, Ordering::Acquire)
+            .is_err()
+        {
+            debug!("Schedule cache refresh already in flight, skipping");
+            return;
+        }
+        let cache = self.clone();
+        tokio::spawn(async move {
+            match load_snapshot(&cache.pool).await {
+                Ok(snap) => {
+                    let count = snap.courses.len();
+                    let _ = cache.tx.send(Arc::new(snap));
+                    info!(courses = count, "Schedule cache refreshed");
+                }
+                Err(e) => {
+                    error!(error = %e, "Failed to refresh schedule cache");
+                }
+            }
+            cache.refreshing.store(false, Ordering::Release);
+        });
+    }
+
+    /// Force an initial load (blocking). Call once at startup.
+    pub(crate) async fn load(&self) -> anyhow::Result<()> {
+        let snap = load_snapshot(&self.pool).await?;
+        let count = snap.courses.len();
+        let _ = self.tx.send(Arc::new(snap));
+        info!(courses = count, "Schedule cache initially loaded");
+        Ok(())
+    }
+}
+
+// ── Database loading ────────────────────────────────────────────────
+
+/// Row returned from the lightweight schedule query.
+#[derive(sqlx::FromRow)]
+struct ScheduleRow {
+    subject: String,
+    enrollment: i32,
+    meeting_times: Value,
+}
+
+/// Load all courses and parse their meeting times into a snapshot.
+async fn load_snapshot(pool: &PgPool) -> anyhow::Result<ScheduleSnapshot> {
+    let start = std::time::Instant::now();
+
+    let rows: Vec<ScheduleRow> = sqlx::query_as(
+        "SELECT subject, enrollment, meeting_times FROM courses",
+    )
+    .fetch_all(pool)
+    .await?;
+
+    let courses: Vec<CachedCourse> = rows
+        .into_iter()
+        .map(|row| {
+            let schedules = parse_meeting_times(&row.meeting_times);
+            CachedCourse {
+                subject: row.subject,
+                enrollment: row.enrollment,
+                schedules,
+            }
+        })
+        .collect();
+
+    debug!(
+        courses = courses.len(),
+        elapsed_ms = start.elapsed().as_millis(),
+        "Schedule snapshot built"
+    );
+
+    Ok(ScheduleSnapshot {
+        courses,
+        refreshed_at: std::time::Instant::now(),
+    })
+}
+
+// ── Meeting time parsing ────────────────────────────────────────────
+
+/// Parse the JSONB `meeting_times` array into compact `ParsedSchedule` values.
+fn parse_meeting_times(value: &Value) -> Vec<ParsedSchedule> {
+    let Value::Array(arr) = value else {
+        return Vec::new();
+    };
+
+    arr.iter().filter_map(parse_one_meeting).collect()
+}
+
+fn parse_one_meeting(mt: &Value) -> Option<ParsedSchedule> {
+    let begin_time = mt.get("begin_time")?.as_str()?;
+    let end_time = mt.get("end_time")?.as_str()?;
+
+    let begin_minutes = parse_hhmm(begin_time)?;
+    let end_minutes = parse_hhmm(end_time)?;
+
+    if end_minutes <= begin_minutes {
+        return None;
+    }
+
+    let start_date = parse_date(mt.get("start_date")?.as_str()?)?;
+    let end_date = parse_date(mt.get("end_date")?.as_str()?)?;
+
+    const DAY_KEYS: [&str; 7] = [
+        "monday",
+        "tuesday",
+        "wednesday",
+        "thursday",
+        "friday",
+        "saturday",
+        "sunday",
+    ];
+    let mut days: u8 = 0;
+    for (bit, key) in DAY_KEYS.iter().enumerate() {
+        if mt.get(*key).and_then(Value::as_bool).unwrap_or(false) {
+            days |= 1 << bit;
+        }
+    }
+
+    // Skip meetings with no days (online async, etc.)
+    if days == 0 {
+        return None;
+    }
+
+    Some(ParsedSchedule {
+        days,
+        begin_minutes,
+        end_minutes,
+        start_date,
+        end_date,
+    })
+}
+
+/// Parse "HHMM" → minutes since midnight.
+fn parse_hhmm(s: &str) -> Option<u16> {
+    if s.len() != 4 {
+        return None;
+    }
+    let hours: u16 = s[..2].parse().ok()?;
+    let mins: u16 = s[2..].parse().ok()?;
+    if hours >= 24 || mins >= 60 {
+        return None;
+    }
+    Some(hours * 60 + mins)
+}
+
+/// Parse "MM/DD/YYYY" → NaiveDate.
+fn parse_date(s: &str) -> Option<NaiveDate> {
+    NaiveDate::parse_from_str(s, "%m/%d/%Y").ok()
+}
+
+// ── Slot matching ───────────────────────────────────────────────────
+
+/// Day-of-week as our bitmask index (Monday = 0 .. Sunday = 6).
+/// Chrono's `weekday().num_days_from_monday()` already gives 0=Mon..6=Sun.
+pub(crate) fn weekday_bit(day: chrono::Weekday) -> u8 {
+    1 << day.num_days_from_monday()
+}
+
+impl ParsedSchedule {
+    /// Check if this schedule is active during a given slot.
+    ///
+    /// `slot_date` is the calendar date of the slot.
+    /// `slot_start` / `slot_end` are minutes since midnight for the 15-min window.
+    #[inline]
+    pub(crate) fn active_during(
+        &self,
+        slot_date: NaiveDate,
+        slot_weekday_bit: u8,
+        slot_start_minutes: u16,
+        slot_end_minutes: u16,
+    ) -> bool {
+        // Day-of-week check
+        if self.days & slot_weekday_bit == 0 {
+            return false;
+        }
+        // Date range check
+        if slot_date < self.start_date || slot_date > self.end_date {
+            return false;
+        }
+        // Time overlap: meeting [begin, end) overlaps slot [start, end)
+        self.begin_minutes < slot_end_minutes && self.end_minutes > slot_start_minutes
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use chrono::NaiveDate;
+    use serde_json::json;
+
+    #[test]
+    fn parse_hhmm_valid() {
+        assert_eq!(parse_hhmm("0000"), Some(0));
+        assert_eq!(parse_hhmm("0930"), Some(570));
+        assert_eq!(parse_hhmm("1350"), Some(830));
+        assert_eq!(parse_hhmm("2359"), Some(1439));
+    }
+
+    #[test]
+    fn parse_hhmm_invalid() {
+        assert_eq!(parse_hhmm(""), None);
+        assert_eq!(parse_hhmm("abc"), None);
+        assert_eq!(parse_hhmm("2500"), None);
+        assert_eq!(parse_hhmm("0060"), None);
+    }
+
+    #[test]
+    fn parse_date_valid() {
+        assert_eq!(
+            parse_date("08/26/2025"),
+            Some(NaiveDate::from_ymd_opt(2025, 8, 26).unwrap())
+        );
+    }
+
+    #[test]
+    fn parse_meeting_times_basic() {
+        let json = json!([{
+            "begin_time": "1000",
+            "end_time": "1050",
+            "start_date": "08/26/2025",
+            "end_date": "12/13/2025",
+            "monday": true,
+            "tuesday": false,
+            "wednesday": true,
+            "thursday": false,
+            "friday": true,
+            "saturday": false,
+            "sunday": false,
+            "building": "NPB",
+            "building_description": "North Paseo Building",
+            "room": "1.238",
+            "campus": "11",
+            "meeting_type": "FF",
+            "meeting_schedule_type": "AFF"
+        }]);
+
+        let schedules = parse_meeting_times(&json);
+        assert_eq!(schedules.len(), 1);
+
+        let s = &schedules[0];
+        assert_eq!(s.begin_minutes, 600); // 10:00
+        assert_eq!(s.end_minutes, 650); // 10:50
+        assert_eq!(s.days, 0b0010101); // Mon, Wed, Fri
+    }
+
+    #[test]
+    fn parse_meeting_times_skips_null_times() {
+        let json = json!([{
+            "begin_time": null,
+            "end_time": null,
+            "start_date": "08/26/2025",
+            "end_date": "12/13/2025",
+            "monday": false,
+            "tuesday": false,
+            "wednesday": false,
+            "thursday": false,
+            "friday": false,
+            "saturday": false,
+            "sunday": false,
+            "meeting_type": "OS",
+            "meeting_schedule_type": "AFF"
+        }]);
+
+        let schedules = parse_meeting_times(&json);
+        assert!(schedules.is_empty());
+    }
+
+    #[test]
+    fn active_during_matching_slot() {
+        let sched = ParsedSchedule {
+            days: 0b0000001, // Monday
+            begin_minutes: 600,
+            end_minutes: 650,
+            start_date: NaiveDate::from_ymd_opt(2025, 8, 26).unwrap(),
+            end_date: NaiveDate::from_ymd_opt(2025, 12, 13).unwrap(),
+        };
+
+        // Monday Sept 1 2025, 10:00-10:15 slot
+        let date = NaiveDate::from_ymd_opt(2025, 9, 1).unwrap();
+        assert!(sched.active_during(date, weekday_bit(chrono::Weekday::Mon), 600, 615));
+    }
+
+    #[test]
+    fn active_during_wrong_day() {
+        let sched = ParsedSchedule {
+            days: 0b0000001, // Monday only
+            begin_minutes: 600,
+            end_minutes: 650,
+            start_date: NaiveDate::from_ymd_opt(2025, 8, 26).unwrap(),
+            end_date: NaiveDate::from_ymd_opt(2025, 12, 13).unwrap(),
+        };
+
+        // Tuesday Sept 2 2025
+        let date = NaiveDate::from_ymd_opt(2025, 9, 2).unwrap();
+        assert!(!sched.active_during(date, weekday_bit(chrono::Weekday::Tue), 600, 615));
+    }
+
+    #[test]
+    fn active_during_no_time_overlap() {
+        let sched = ParsedSchedule {
+            days: 0b0000001,
+            begin_minutes: 600, // 10:00
+            end_minutes: 650,   // 10:50
+            start_date: NaiveDate::from_ymd_opt(2025, 8, 26).unwrap(),
+            end_date: NaiveDate::from_ymd_opt(2025, 12, 13).unwrap(),
+        };
+
+        let date = NaiveDate::from_ymd_opt(2025, 9, 1).unwrap(); // Monday
+        // Slot 11:00-11:15 — after the meeting ends
+        assert!(!sched.active_during(date, weekday_bit(chrono::Weekday::Mon), 660, 675));
+        // Slot 9:45-10:00 — just before meeting starts (end=600, begin=600 → no overlap)
+        assert!(!sched.active_during(date, weekday_bit(chrono::Weekday::Mon), 585, 600));
+    }
+
+    #[test]
+    fn active_during_outside_date_range() {
+        let sched = ParsedSchedule {
+            days: 0b0000001,
+            begin_minutes: 600,
+            end_minutes: 650,
+            start_date: NaiveDate::from_ymd_opt(2025, 8, 26).unwrap(),
+            end_date: NaiveDate::from_ymd_opt(2025, 12, 13).unwrap(),
+        };
+
+        // Monday Jan 6 2025 — before semester
+        let date = NaiveDate::from_ymd_opt(2025, 1, 6).unwrap();
+        assert!(!sched.active_during(date, weekday_bit(chrono::Weekday::Mon), 600, 615));
+    }
+
+    #[test]
+    fn active_during_edge_overlap() {
+        let sched = ParsedSchedule {
+            days: 0b0000001,
+            begin_minutes: 600,
+            end_minutes: 650,
+            start_date: NaiveDate::from_ymd_opt(2025, 8, 26).unwrap(),
+            end_date: NaiveDate::from_ymd_opt(2025, 12, 13).unwrap(),
+        };
+
+        let date = NaiveDate::from_ymd_opt(2025, 9, 1).unwrap();
+        // Slot 10:45-11:00 — overlaps last 5 minutes of meeting
+        assert!(sched.active_during(date, weekday_bit(chrono::Weekday::Mon), 645, 660));
+        // Slot 9:45-10:00 — ends exactly when meeting starts, no overlap
+        assert!(!sched.active_during(date, weekday_bit(chrono::Weekday::Mon), 585, 600));
+        // Slot 10:50-11:05 — starts exactly when meeting ends, no overlap
+        assert!(!sched.active_during(date, weekday_bit(chrono::Weekday::Mon), 650, 665));
+    }
+}
@@ -0,0 +1,188 @@
+//! In-memory caches for session resolution and OAuth CSRF state.
+
+use chrono::{DateTime, Utc};
+use dashmap::DashMap;
+use rand::Rng;
+use sqlx::PgPool;
+use std::sync::Arc;
+use std::time::{Duration, Instant};
+
+use crate::data::models::User;
+
+/// Cached session entry with TTL.
+#[derive(Debug, Clone)]
+struct CachedSession {
+    user: User,
+    session_expires_at: DateTime<Utc>,
+    cached_at: Instant,
+}
+
+/// In-memory session cache backed by PostgreSQL.
+///
+/// Provides fast session resolution without a DB round-trip on every request.
+/// Cache entries expire after a configurable TTL (default 5 minutes).
+#[derive(Clone)]
+pub struct SessionCache {
+    cache: Arc<DashMap<String, CachedSession>>,
+    db_pool: PgPool,
+    cache_ttl: Duration,
+}
+
+impl SessionCache {
+    /// Create a new session cache with a 5-minute default TTL.
+    pub fn new(db_pool: PgPool) -> Self {
+        Self {
+            cache: Arc::new(DashMap::new()),
+            db_pool,
+            cache_ttl: Duration::from_secs(5 * 60),
+        }
+    }
+
+    /// Resolve a session token to a [`User`], using the cache when possible.
+    ///
+    /// On cache hit (entry present, not stale, session not expired), returns the
+    /// cached user immediately. On miss or stale entry, queries the database for
+    /// the session and user, populates the cache, and fire-and-forgets a
+    /// `touch_session` call to update `last_active_at`.
+    pub async fn get_user(&self, token: &str) -> Option<User> {
+        // Check cache first
+        if let Some(entry) = self.cache.get(token) {
+            let now_instant = Instant::now();
+            let now_utc = Utc::now();
+
+            let cache_fresh = entry.cached_at + self.cache_ttl > now_instant;
+            let session_valid = entry.session_expires_at > now_utc;
+
+            if cache_fresh && session_valid {
+                return Some(entry.user.clone());
+            }
+
+            // Stale or expired — drop the ref before removing
+            drop(entry);
+            self.cache.remove(token);
+        }
+
+        // Cache miss — query DB
+        let session = crate::data::sessions::get_session(&self.db_pool, token)
+            .await
+            .ok()
+            .flatten()?;
+
+        let user = crate::data::users::get_user(&self.db_pool, session.user_id)
+            .await
+            .ok()
+            .flatten()?;
+
+        self.cache.insert(
+            token.to_owned(),
+            CachedSession {
+                user: user.clone(),
+                session_expires_at: session.expires_at,
+                cached_at: Instant::now(),
+            },
+        );
+
+        // Fire-and-forget touch to update last_active_at
+        let pool = self.db_pool.clone();
+        let token_owned = token.to_owned();
+        tokio::spawn(async move {
+            if let Err(e) = crate::data::sessions::touch_session(&pool, &token_owned).await {
+                tracing::warn!(error = %e, "failed to touch session");
+            }
+        });
+
+        Some(user)
+    }
+
+    /// Remove a single session from the cache (e.g. on logout).
+    pub fn evict(&self, token: &str) {
+        self.cache.remove(token);
+    }
+
+    /// Remove all cached sessions belonging to a user.
+    pub fn evict_user(&self, discord_id: i64) {
+        self.cache
+            .retain(|_, entry| entry.user.discord_id != discord_id);
+    }
+
+    /// Delete expired sessions from the database and sweep the in-memory cache.
+    ///
+    /// Returns the number of sessions deleted from the database.
+    #[allow(dead_code)] // Intended for periodic cleanup task (not yet wired)
+    pub async fn cleanup_expired(&self) -> anyhow::Result<u64> {
+        let deleted = crate::data::sessions::cleanup_expired(&self.db_pool).await?;
+
+        let now = Utc::now();
+        self.cache.retain(|_, entry| entry.session_expires_at > now);
+
+        Ok(deleted)
+    }
+}
+
+/// Data stored alongside each OAuth CSRF state token.
+struct OAuthStateEntry {
+    created_at: Instant,
+    /// The browser origin that initiated the login flow, so the callback
+    /// can reconstruct the exact redirect_uri Discord expects.
+    origin: String,
+}
+
+/// Ephemeral store for OAuth CSRF state tokens.
+///
+/// Tokens are stored with creation time and expire after a configurable TTL.
+/// Each token is single-use: validation consumes it.
+#[derive(Clone)]
+pub struct OAuthStateStore {
+    states: Arc<DashMap<String, OAuthStateEntry>>,
+    ttl: Duration,
+}
+
+impl Default for OAuthStateStore {
+    fn default() -> Self {
+        Self::new()
+    }
+}
+
+impl OAuthStateStore {
+    /// Create a new store with a 10-minute TTL.
+    pub fn new() -> Self {
+        Self {
+            states: Arc::new(DashMap::new()),
+            ttl: Duration::from_secs(10 * 60),
+        }
+    }
+
+    /// Generate a random 16-byte hex CSRF token, store it with the given
+    /// origin, and return the token.
+    pub fn generate(&self, origin: String) -> String {
+        let bytes: [u8; 16] = rand::rng().random();
+        let token: String = bytes.iter().map(|b| format!("{b:02x}")).collect();
+        self.states.insert(
+            token.clone(),
+            OAuthStateEntry {
+                created_at: Instant::now(),
+                origin,
+            },
+        );
+        token
+    }
+
+    /// Validate and consume a CSRF token. Returns the stored origin if the
+    /// token was present and not expired.
+    pub fn validate(&self, state: &str) -> Option<String> {
+        let (_, entry) = self.states.remove(state)?;
+        if entry.created_at.elapsed() < self.ttl {
+            Some(entry.origin)
+        } else {
+            None
+        }
+    }
+
+    /// Remove all expired entries from the store.
+    #[allow(dead_code)] // Intended for periodic cleanup task (not yet wired)
+    pub fn cleanup(&self) {
+        let ttl = self.ttl;
+        self.states
+            .retain(|_, entry| entry.created_at.elapsed() < ttl);
+    }
+}
@@ -0,0 +1,259 @@
+//! Timeline API endpoint for enrollment aggregation by subject over time.
+//!
+//! Accepts multiple time ranges, merges overlaps, aligns to 15-minute
+//! slot boundaries, and returns per-subject enrollment totals for each slot.
+//! Only courses whose meeting times overlap a given slot contribute to that
+//! slot's totals — so the chart reflects the actual class schedule rhythm.
+//!
+//! Course data is served from an ISR-style in-memory cache (see
+//! [`ScheduleCache`]) that refreshes hourly in the background with
+//! stale-while-revalidate semantics.
+
+use axum::{
+    extract::State,
+    http::StatusCode,
+    response::{IntoResponse, Json, Response},
+};
+use chrono::{DateTime, Datelike, Duration, NaiveTime, Timelike, Utc};
+use chrono_tz::US::Central;
+use serde::{Deserialize, Serialize};
+use std::collections::{BTreeMap, BTreeSet};
+use ts_rs::TS;
+
+use crate::state::AppState;
+use crate::web::schedule_cache::weekday_bit;
+
+/// 15 minutes in seconds, matching the frontend `SLOT_INTERVAL_MS`.
+const SLOT_SECONDS: i64 = 15 * 60;
+const SLOT_MINUTES: u16 = 15;
+
+/// Maximum number of ranges in a single request.
+const MAX_RANGES: usize = 20;
+
+/// Maximum span of a single range (72 hours).
+const MAX_RANGE_SPAN: Duration = Duration::hours(72);
+
+/// Maximum total span across all ranges to prevent excessive queries.
+const MAX_TOTAL_SPAN: Duration = Duration::hours(168); // 1 week
+
+// ── Request / Response types ────────────────────────────────────────
+
+#[derive(Debug, Deserialize)]
+pub(crate) struct TimelineRequest {
+    ranges: Vec<TimeRange>,
+}
+
+#[derive(Debug, Deserialize)]
+pub(crate) struct TimeRange {
+    start: DateTime<Utc>,
+    end: DateTime<Utc>,
+}
+
+#[derive(Debug, Serialize, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export)]
+pub struct TimelineResponse {
+    /// 15-minute slots with per-subject enrollment totals, sorted by time.
+    slots: Vec<TimelineSlot>,
+    /// All subject codes present in the returned data.
+    subjects: Vec<String>,
+}
+
+#[derive(Debug, Serialize, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export)]
+pub struct TimelineSlot {
+    /// ISO-8601 timestamp at the start of this 15-minute bucket.
+    time: DateTime<Utc>,
+    /// Subject code → total enrollment in this slot.
+    subjects: BTreeMap<String, i64>,
+}
+
+// ── Error type ──────────────────────────────────────────────────────
+
+pub(crate) struct TimelineError {
+    status: StatusCode,
+    message: String,
+}
+
+impl TimelineError {
+    fn bad_request(msg: impl Into<String>) -> Self {
+        Self {
+            status: StatusCode::BAD_REQUEST,
+            message: msg.into(),
+        }
+    }
+}
+
+impl IntoResponse for TimelineError {
+    fn into_response(self) -> Response {
+        (
+            self.status,
+            Json(serde_json::json!({ "error": self.message })),
+        )
+            .into_response()
+    }
+}
+
+// ── Alignment helpers ───────────────────────────────────────────────
+
+/// Floor a timestamp to the nearest 15-minute boundary.
+fn align_floor(ts: DateTime<Utc>) -> DateTime<Utc> {
+    let secs = ts.timestamp();
+    let aligned = (secs / SLOT_SECONDS) * SLOT_SECONDS;
+    DateTime::from_timestamp(aligned, 0).unwrap_or(ts)
+}
+
+/// Ceil a timestamp to the nearest 15-minute boundary.
+fn align_ceil(ts: DateTime<Utc>) -> DateTime<Utc> {
+    let secs = ts.timestamp();
+    let aligned = ((secs + SLOT_SECONDS - 1) / SLOT_SECONDS) * SLOT_SECONDS;
+    DateTime::from_timestamp(aligned, 0).unwrap_or(ts)
+}
+
+// ── Range merging ───────────────────────────────────────────────────
+
+/// Aligned, validated range.
+#[derive(Debug, Clone, Copy)]
+struct AlignedRange {
+    start: DateTime<Utc>,
+    end: DateTime<Utc>,
+}
+
+/// Merge overlapping/adjacent ranges into a minimal set.
+fn merge_ranges(mut ranges: Vec<AlignedRange>) -> Vec<AlignedRange> {
+    if ranges.is_empty() {
+        return ranges;
+    }
+    ranges.sort_by_key(|r| r.start);
+    let mut merged: Vec<AlignedRange> = vec![ranges[0]];
+    for r in &ranges[1..] {
+        let last = merged.last_mut().unwrap();
+        if r.start <= last.end {
+            last.end = last.end.max(r.end);
+        } else {
+            merged.push(*r);
+        }
+    }
+    merged
+}
+
+/// Generate all aligned slot timestamps within the merged ranges.
+fn generate_slots(merged: &[AlignedRange]) -> BTreeSet<DateTime<Utc>> {
+    let mut slots = BTreeSet::new();
+    for range in merged {
+        let mut t = range.start;
+        while t < range.end {
+            slots.insert(t);
+            t += Duration::seconds(SLOT_SECONDS);
+        }
+    }
+    slots
+}
+
+// ── Handler ─────────────────────────────────────────────────────────
+
+/// `POST /api/timeline`
+///
+/// Accepts a JSON body with multiple time ranges. Returns per-subject
+/// enrollment totals bucketed into 15-minute slots. Only courses whose
+/// meeting schedule overlaps a slot contribute to that slot's count.
+pub(crate) async fn timeline(
+    State(state): State<AppState>,
+    Json(body): Json<TimelineRequest>,
+) -> Result<Json<TimelineResponse>, TimelineError> {
+    // ── Validate ────────────────────────────────────────────────────
+    if body.ranges.is_empty() {
+        return Err(TimelineError::bad_request("At least one range is required"));
+    }
+    if body.ranges.len() > MAX_RANGES {
+        return Err(TimelineError::bad_request(format!(
+            "Too many ranges (max {MAX_RANGES})"
+        )));
+    }
+
+    let mut aligned: Vec<AlignedRange> = Vec::with_capacity(body.ranges.len());
+    for r in &body.ranges {
+        if r.end <= r.start {
+            return Err(TimelineError::bad_request(format!(
+                "Range end ({}) must be after start ({})",
+                r.end, r.start
+            )));
+        }
+        let span = r.end - r.start;
+        if span > MAX_RANGE_SPAN {
+            return Err(TimelineError::bad_request(format!(
+                "Range span ({} hours) exceeds maximum ({} hours)",
+                span.num_hours(),
+                MAX_RANGE_SPAN.num_hours()
+            )));
+        }
+        aligned.push(AlignedRange {
+            start: align_floor(r.start),
+            end: align_ceil(r.end),
+        });
+    }
+
+    let merged = merge_ranges(aligned);
+
+    // Validate total span
+    let total_span: Duration = merged.iter().map(|r| r.end - r.start).sum();
+    if total_span > MAX_TOTAL_SPAN {
+        return Err(TimelineError::bad_request(format!(
+            "Total time span ({} hours) exceeds maximum ({} hours)",
+            total_span.num_hours(),
+            MAX_TOTAL_SPAN.num_hours()
+        )));
+    }
+
+    // ── Get cached schedule data (ISR: stale-while-revalidate) ───────
+    state.schedule_cache.ensure_fresh();
+    let snapshot = state.schedule_cache.snapshot();
+
+    // ── Build per-slot enrollment by filtering on meeting times ──────
+    let slot_times = generate_slots(&merged);
+    let mut all_subjects: BTreeSet<String> = BTreeSet::new();
+
+    let slots: Vec<TimelineSlot> = slot_times
+        .into_iter()
+        .map(|utc_time| {
+            // Convert UTC slot to Central time for local day-of-week and time-of-day
+            let local = utc_time.with_timezone(&Central);
+            let local_date = local.date_naive();
+            let local_time = local.time();
+            let weekday = local.weekday();
+            let wday_bit = weekday_bit(weekday);
+            let slot_start_minutes = time_to_minutes(local_time);
+            let slot_end_minutes = slot_start_minutes + SLOT_MINUTES;
+
+            let mut subject_totals: BTreeMap<String, i64> = BTreeMap::new();
+
+            for course in &snapshot.courses {
+                let active = course.schedules.iter().any(|s| {
+                    s.active_during(local_date, wday_bit, slot_start_minutes, slot_end_minutes)
+                });
+                if active {
+                    *subject_totals
+                        .entry(course.subject.clone())
+                        .or_default() += course.enrollment as i64;
+                }
+            }
+
+            all_subjects.extend(subject_totals.keys().cloned());
+
+            TimelineSlot {
+                time: utc_time,
+                subjects: subject_totals,
+            }
+        })
+        .collect();
+
+    let subjects: Vec<String> = all_subjects.into_iter().collect();
+
+    Ok(Json(TimelineResponse { slots, subjects }))
+}
+
+/// Convert a `NaiveTime` to minutes since midnight.
+fn time_to_minutes(t: NaiveTime) -> u16 {
+    (t.hour() * 60 + t.minute()) as u16
+}
@@ -0,0 +1,205 @@
+//! WebSocket event types and handler for real-time scrape job updates.
+
+use axum::{
+    extract::{
+        State,
+        ws::{Message, WebSocket, WebSocketUpgrade},
+    },
+    response::IntoResponse,
+};
+use futures::{SinkExt, StreamExt};
+use serde::Serialize;
+use sqlx::PgPool;
+use tokio::sync::broadcast;
+use tracing::debug;
+
+use crate::data::models::{ScrapeJob, ScrapeJobStatus};
+use crate::state::AppState;
+use crate::web::extractors::AdminUser;
+
+/// A serializable DTO for `ScrapeJob` with computed `status`.
+#[derive(Debug, Clone, Serialize)]
+#[serde(rename_all = "camelCase")]
+pub struct ScrapeJobDto {
+    pub id: i32,
+    pub target_type: String,
+    pub target_payload: serde_json::Value,
+    pub priority: String,
+    pub execute_at: String,
+    pub created_at: String,
+    pub locked_at: Option<String>,
+    pub retry_count: i32,
+    pub max_retries: i32,
+    pub queued_at: String,
+    pub status: ScrapeJobStatus,
+}
+
+impl From<&ScrapeJob> for ScrapeJobDto {
+    fn from(job: &ScrapeJob) -> Self {
+        Self {
+            id: job.id,
+            target_type: format!("{:?}", job.target_type),
+            target_payload: job.target_payload.clone(),
+            priority: format!("{:?}", job.priority),
+            execute_at: job.execute_at.to_rfc3339(),
+            created_at: job.created_at.to_rfc3339(),
+            locked_at: job.locked_at.map(|t| t.to_rfc3339()),
+            retry_count: job.retry_count,
+            max_retries: job.max_retries,
+            queued_at: job.queued_at.to_rfc3339(),
+            status: job.status(),
+        }
+    }
+}
+
+/// Events broadcast when scrape job state changes.
+#[derive(Debug, Clone, Serialize)]
+#[serde(tag = "type", rename_all = "camelCase")]
+pub enum ScrapeJobEvent {
+    Init {
+        jobs: Vec<ScrapeJobDto>,
+    },
+    JobCreated {
+        job: ScrapeJobDto,
+    },
+    JobLocked {
+        id: i32,
+        locked_at: String,
+        status: ScrapeJobStatus,
+    },
+    JobCompleted {
+        id: i32,
+    },
+    JobRetried {
+        id: i32,
+        retry_count: i32,
+        queued_at: String,
+        status: ScrapeJobStatus,
+    },
+    JobExhausted {
+        id: i32,
+    },
+    JobDeleted {
+        id: i32,
+    },
+}
+
+/// Fetch current scrape jobs from the DB and build an `Init` event.
+async fn build_init_event(db_pool: &PgPool) -> Result<ScrapeJobEvent, sqlx::Error> {
+    let rows = sqlx::query_as::<_, ScrapeJob>(
+        "SELECT * FROM scrape_jobs ORDER BY priority DESC, execute_at ASC LIMIT 100",
+    )
+    .fetch_all(db_pool)
+    .await?;
+
+    let jobs = rows.iter().map(ScrapeJobDto::from).collect();
+    Ok(ScrapeJobEvent::Init { jobs })
+}
+
+/// WebSocket endpoint for real-time scrape job updates.
+///
+/// Auth is checked via `AdminUser` before the upgrade occurs — if rejected,
+/// a 401/403 is returned and the upgrade never happens.
+pub async fn scrape_jobs_ws(
+    ws: WebSocketUpgrade,
+    AdminUser(_user): AdminUser,
+    State(state): State<AppState>,
+) -> impl IntoResponse {
+    ws.on_upgrade(|socket| handle_scrape_jobs_ws(socket, state))
+}
+
+/// Serialize an event and send it over the WebSocket sink.
+/// Returns `true` if the message was sent, `false` if the client disconnected.
+async fn send_event(
+    sink: &mut futures::stream::SplitSink<WebSocket, Message>,
+    event: &ScrapeJobEvent,
+) -> bool {
+    let Ok(json) = serde_json::to_string(event) else {
+        return true; // serialization failed, but connection is still alive
+    };
+    sink.send(Message::Text(json.into())).await.is_ok()
+}
+
+async fn handle_scrape_jobs_ws(socket: WebSocket, state: AppState) {
+    debug!("scrape-jobs WebSocket connected");
+
+    let (mut sink, mut stream) = socket.split();
+
+    // Send initial state
+    let init_event = match build_init_event(&state.db_pool).await {
+        Ok(event) => event,
+        Err(e) => {
+            debug!(error = %e, "failed to build init event, closing WebSocket");
+            return;
+        }
+    };
+    if !send_event(&mut sink, &init_event).await {
+        debug!("client disconnected during init send");
+        return;
+    }
+
+    // Subscribe to broadcast events
+    let mut rx = state.scrape_job_events();
+
+    loop {
+        tokio::select! {
+            result = rx.recv() => {
+                match result {
+                    Ok(ref event) => {
+                        if !send_event(&mut sink, event).await {
+                            debug!("client disconnected during event send");
+                            break;
+                        }
+                    }
+                    Err(broadcast::error::RecvError::Lagged(n)) => {
+                        debug!(missed = n, "broadcast lagged, resyncing");
+                        match build_init_event(&state.db_pool).await {
+                            Ok(ref event) => {
+                                if !send_event(&mut sink, event).await {
+                                    debug!("client disconnected during resync send");
+                                    break;
+                                }
+                            }
+                            Err(e) => {
+                                debug!(error = %e, "failed to build resync init event");
+                            }
+                        }
+                    }
+                    Err(broadcast::error::RecvError::Closed) => {
+                        debug!("broadcast channel closed");
+                        break;
+                    }
+                }
+            }
+            msg = stream.next() => {
+                match msg {
+                    Some(Ok(Message::Text(text))) => {
+                        if let Ok(parsed) = serde_json::from_str::<serde_json::Value>(&text)
+                            && parsed.get("type").and_then(|t| t.as_str()) == Some("resync")
+                        {
+                            debug!("client requested resync");
+                            match build_init_event(&state.db_pool).await {
+                                Ok(ref event) => {
+                                    if !send_event(&mut sink, event).await {
+                                        debug!("client disconnected during resync send");
+                                        break;
+                                    }
+                                }
+                                Err(e) => {
+                                    debug!(error = %e, "failed to build resync init event");
+                                }
+                            }
+                        }
+                    }
+                    Some(Ok(Message::Close(_))) | None => {
+                        debug!("client disconnected");
+                        break;
+                    }
+                    _ => {}
+                }
+            }
+        }
+    }
+
+    debug!("scrape-jobs WebSocket disconnected");
+}
@@ -210,3 +210,116 @@ async fn test_batch_upsert_unique_constraint_crn_term(pool: PgPool) {
    assert_eq!(rows[1].0, "202520");
    assert_eq!(rows[1].1, 10);
 }
+
+#[sqlx::test]
+async fn test_batch_upsert_creates_audit_and_metric_entries(pool: PgPool) {
+    // Insert initial data — should NOT create audits/metrics (it's a fresh insert)
+    let initial = vec![helpers::make_course(
+        "50001",
+        "202510",
+        "CS",
+        "3443",
+        "App Programming",
+        10,
+        35,
+        0,
+        5,
+    )];
+    batch_upsert_courses(&initial, &pool).await.unwrap();
+
+    let (audit_count,): (i64,) = sqlx::query_as("SELECT COUNT(*) FROM course_audits")
+        .fetch_one(&pool)
+        .await
+        .unwrap();
+    assert_eq!(
+        audit_count, 0,
+        "initial insert should not create audit entries"
+    );
+
+    let (metric_count,): (i64,) = sqlx::query_as("SELECT COUNT(*) FROM course_metrics")
+        .fetch_one(&pool)
+        .await
+        .unwrap();
+    assert_eq!(
+        metric_count, 0,
+        "initial insert should not create metric entries"
+    );
+
+    // Update enrollment and wait_count
+    let updated = vec![helpers::make_course(
+        "50001",
+        "202510",
+        "CS",
+        "3443",
+        "App Programming",
+        20,
+        35,
+        2,
+        5,
+    )];
+    batch_upsert_courses(&updated, &pool).await.unwrap();
+
+    // Should have audit entries for enrollment and wait_count changes
+    let (audit_count,): (i64,) = sqlx::query_as("SELECT COUNT(*) FROM course_audits")
+        .fetch_one(&pool)
+        .await
+        .unwrap();
+    assert!(
+        audit_count >= 2,
+        "should have audit entries for enrollment and wait_count changes, got {audit_count}"
+    );
+
+    // Should have exactly 1 metric entry
+    let (metric_count,): (i64,) = sqlx::query_as("SELECT COUNT(*) FROM course_metrics")
+        .fetch_one(&pool)
+        .await
+        .unwrap();
+    assert_eq!(metric_count, 1, "should have 1 metric snapshot");
+
+    // Verify metric values
+    let (enrollment, wait_count, seats): (i32, i32, i32) = sqlx::query_as(
+        "SELECT enrollment, wait_count, seats_available FROM course_metrics LIMIT 1",
+    )
+    .fetch_one(&pool)
+    .await
+    .unwrap();
+    assert_eq!(enrollment, 20);
+    assert_eq!(wait_count, 2);
+    assert_eq!(seats, 15); // 35 - 20
+}
+
+#[sqlx::test]
+async fn test_batch_upsert_no_change_no_audit(pool: PgPool) {
+    // Insert then re-insert identical data — should produce zero audits/metrics
+    let course = vec![helpers::make_course(
+        "60001",
+        "202510",
+        "CS",
+        "1083",
+        "Intro to CS",
+        25,
+        30,
+        0,
+        5,
+    )];
+    batch_upsert_courses(&course, &pool).await.unwrap();
+    batch_upsert_courses(&course, &pool).await.unwrap();
+
+    let (audit_count,): (i64,) = sqlx::query_as("SELECT COUNT(*) FROM course_audits")
+        .fetch_one(&pool)
+        .await
+        .unwrap();
+    assert_eq!(
+        audit_count, 0,
+        "identical re-upsert should not create audit entries"
+    );
+
+    let (metric_count,): (i64,) = sqlx::query_as("SELECT COUNT(*) FROM course_metrics")
+        .fetch_one(&pool)
+        .await
+        .unwrap();
+    assert_eq!(
+        metric_count, 0,
+        "identical re-upsert should not create metric entries"
+    );
+}
@@ -217,10 +217,13 @@ async fn unlock_and_increment_retry_has_retries_remaining(pool: PgPool) {
    )
    .await;

-    let has_retries = scrape_jobs::unlock_and_increment_retry(id, 3, &pool)
+    let result = scrape_jobs::unlock_and_increment_retry(id, 3, &pool)
        .await
        .unwrap();
-    assert!(has_retries, "should have retries remaining (0→1, max=3)");
+    assert!(
+        result.is_some(),
+        "should have retries remaining (0→1, max=3)"
+    );

    // Verify state in DB
    let (retry_count, locked_at): (i32, Option<chrono::DateTime<chrono::Utc>>) =
@@ -241,17 +244,17 @@ async fn unlock_and_increment_retry_exhausted(pool: PgPool) {
        json!({"subject": "CS"}),
        ScrapePriority::Medium,
        true,
-        2, // retry_count
+        3, // retry_count (already used all 3 retries)
        3, // max_retries
    )
    .await;

-    let has_retries = scrape_jobs::unlock_and_increment_retry(id, 3, &pool)
+    let result = scrape_jobs::unlock_and_increment_retry(id, 3, &pool)
        .await
        .unwrap();
    assert!(
-        !has_retries,
-        "should NOT have retries remaining (2→3, max=3)"
+        result.is_none(),
+        "should NOT have retries remaining (3→4, max=3)"
    );

    let (retry_count,): (i32,) =
@@ -260,7 +263,7 @@ async fn unlock_and_increment_retry_exhausted(pool: PgPool) {
            .fetch_one(&pool)
            .await
            .unwrap();
-    assert_eq!(retry_count, 3);
+    assert_eq!(retry_count, 4);
 }

 #[sqlx::test]
@@ -276,11 +279,11 @@ async fn unlock_and_increment_retry_already_exceeded(pool: PgPool) {
    )
    .await;

-    let has_retries = scrape_jobs::unlock_and_increment_retry(id, 3, &pool)
+    let result = scrape_jobs::unlock_and_increment_retry(id, 3, &pool)
        .await
        .unwrap();
    assert!(
-        !has_retries,
+        result.is_none(),
        "should NOT have retries remaining (5→6, max=3)"
    );

@@ -346,7 +349,7 @@ async fn find_existing_payloads_returns_matching(pool: PgPool) {
 }

 #[sqlx::test]
-async fn find_existing_payloads_ignores_locked(pool: PgPool) {
+async fn find_existing_payloads_includes_locked(pool: PgPool) {
    let payload = json!({"subject": "CS"});

    helpers::insert_scrape_job(
@@ -365,7 +368,10 @@ async fn find_existing_payloads_ignores_locked(pool: PgPool) {
        .await
        .unwrap();

-    assert!(existing.is_empty(), "locked jobs should be ignored");
+    assert!(
+        existing.contains(&payload.to_string()),
+        "locked jobs should be included in deduplication"
+    );
 }

 #[sqlx::test]
@@ -5,5 +5,4 @@ dist-ssr
 *.local
 count.txt
 .env
-.nitro
-.tanstack
+.svelte-kit
@@ -7,7 +7,7 @@
  },
  "files": {
    "ignoreUnknown": false,
-    "ignore": ["dist/", "node_modules/", ".tanstack/"]
+    "ignore": ["dist/", "node_modules/", ".svelte-kit/", "src/lib/bindings/"]
  },
  "formatter": {
    "enabled": true,
@@ -1,60 +0,0 @@
-import js from "@eslint/js";
-import tseslint from "typescript-eslint";
-import react from "eslint-plugin-react";
-import reactHooks from "eslint-plugin-react-hooks";
-import reactRefresh from "eslint-plugin-react-refresh";
-
-export default tseslint.config(
-  // Ignore generated files and build outputs
-  {
-    ignores: ["dist", "node_modules", "src/routeTree.gen.ts", "*.config.js"],
-  },
-  // Base configs
-  js.configs.recommended,
-  ...tseslint.configs.recommendedTypeChecked,
-  // React plugin configuration
-  {
-    files: ["**/*.{ts,tsx}"],
-    plugins: {
-      react,
-      "react-hooks": reactHooks,
-      "react-refresh": reactRefresh,
-    },
-    languageOptions: {
-      parserOptions: {
-        project: true,
-        tsconfigRootDir: import.meta.dirname,
-        ecmaFeatures: {
-          jsx: true,
-        },
-      },
-    },
-    settings: {
-      react: {
-        version: "19.0",
-      },
-    },
-    rules: {
-      // React rules
-      ...react.configs.recommended.rules,
-      ...react.configs["jsx-runtime"].rules,
-      ...reactHooks.configs.recommended.rules,
-
-      // React Refresh
-      "react-refresh/only-export-components": ["warn", { allowConstantExport: true }],
-
-      // TypeScript overrides
-      "@typescript-eslint/no-unused-vars": [
-        "error",
-        {
-          argsIgnorePattern: "^_",
-          varsIgnorePattern: "^_",
-        },
-      ],
-      "@typescript-eslint/no-explicit-any": "warn",
-
-      // Disable prop-types since we're using TypeScript
-      "react/prop-types": "off",
-    },
-  }
-);
@@ -1,20 +0,0 @@
-<!doctype html>
-<html lang="en">
-  <head>
-    <meta charset="UTF-8" />
-    <meta name="viewport" content="width=device-width, initial-scale=1.0" />
-    <link rel="icon" href="/favicon.ico" />
-    <meta name="theme-color" content="#000000" />
-    <meta
-      name="description"
-      content="Banner, a Discord bot and web interface for UTSA Course Monitoring"
-    />
-    <link rel="apple-touch-icon" href="/logo192.png" />
-    <link rel="manifest" href="/manifest.json" />
-    <title>Banner</title>
-  </head>
-  <body>
-    <div id="app"></div>
-    <script type="module" src="/src/main.tsx"></script>
-  </body>
-</html>
@@ -3,48 +3,46 @@
  "private": true,
  "type": "module",
  "scripts": {
-    "dev": "vite --port 3000",
-    "start": "vite --port 3000",
-    "build": "vite build && tsc",
-    "serve": "vite preview",
+    "dev": "vite dev --port 3000",
+    "build": "vite build",
+    "preview": "vite preview",
+    "check": "svelte-kit sync && svelte-check --tsconfig ./tsconfig.json",
    "test": "vitest run",
-    "lint": "tsc && eslint . --ext .ts,.tsx",
-    "typecheck": "tsc --noEmit",
    "format": "biome format --write .",
    "format:check": "biome format ."
  },
-  "dependencies": {
-    "@radix-ui/themes": "^3.2.1",
-    "@tanstack/react-devtools": "^0.2.2",
-    "@tanstack/react-router": "^1.157.16",
-    "@tanstack/react-router-devtools": "^1.157.16",
-    "@tanstack/router-plugin": "^1.157.16",
-    "lucide-react": "^0.544.0",
-    "next-themes": "^0.4.6",
-    "react": "^19.2.4",
-    "react-dom": "^19.2.4",
-    "react-timeago": "^8.3.0",
-    "recharts": "^3.7.0"
-  },
  "devDependencies": {
    "@biomejs/biome": "^1.9.4",
-    "@eslint/js": "^9.39.2",
-    "@testing-library/dom": "^10.4.0",
-    "@testing-library/react": "^16.3.2",
-    "@types/node": "^24.10.9",
-    "@types/react": "^19.2.10",
-    "@types/react-dom": "^19.0.3",
-    "@vitejs/plugin-react": "^4.3.4",
-    "baseline-browser-mapping": "^2.9.19",
-    "eslint": "^9.39.2",
-    "eslint-plugin-react": "^7.37.5",
-    "eslint-plugin-react-hooks": "^7.0.1",
-    "eslint-plugin-react-refresh": "^0.4.26",
+    "@fontsource-variable/inter": "^5.2.5",
+    "@lucide/svelte": "^0.563.0",
+    "@sveltejs/adapter-static": "^3.0.8",
+    "@sveltejs/kit": "^2.16.0",
+    "@sveltejs/vite-plugin-svelte": "^5.0.3",
+    "@tailwindcss/vite": "^4.0.0",
+    "@tanstack/table-core": "^8.21.3",
+    "@types/d3-scale": "^4.0.9",
+    "@types/d3-shape": "^3.1.8",
+    "@types/d3-time-format": "^4.0.3",
+    "@types/node": "^25.1.0",
+    "bits-ui": "^1.3.7",
+    "clsx": "^2.1.1",
    "jsdom": "^26.0.0",
+    "svelte": "^5.19.0",
+    "svelte-check": "^4.1.4",
+    "tailwind-merge": "^3.0.1",
+    "tailwindcss": "^4.0.0",
    "typescript": "^5.7.2",
-    "typescript-eslint": "^8.54.0",
    "vite": "^6.3.5",
-    "vitest": "^3.0.5",
-    "web-vitals": "^4.2.4"
+    "vitest": "^3.0.5"
+  },
+  "dependencies": {
+    "@icons-pack/svelte-simple-icons": "^6.5.0",
+    "d3-scale": "^4.0.2",
+    "d3-shape": "^3.2.0",
+    "d3-time-format": "^4.1.0",
+    "date-fns": "^4.1.0",
+    "layerchart": "^1.0.13",
+    "overlayscrollbars": "^2.14.0",
+    "overlayscrollbars-svelte": "^0.5.5"
  }
 }
@@ -0,0 +1,148 @@
+#!/usr/bin/env bun
+/**
+ * Pre-compress static assets with maximum compression levels.
+ * Run after `bun run build`.
+ *
+ * Generates .gz, .br, .zst variants for compressible files ≥ MIN_SIZE bytes.
+ * These are embedded alongside originals by rust-embed and served via
+ * content negotiation in src/web/assets.rs.
+ */
+import { readdir, stat, readFile, writeFile } from "fs/promises";
+import { join, extname } from "path";
+import { gzipSync, brotliCompressSync, constants } from "zlib";
+import { $ } from "bun";
+
+// Must match COMPRESSION_MIN_SIZE in src/web/encoding.rs
+const MIN_SIZE = 512;
+
+const COMPRESSIBLE_EXTENSIONS = new Set([
+  ".js",
+  ".css",
+  ".html",
+  ".json",
+  ".svg",
+  ".txt",
+  ".xml",
+  ".map",
+]);
+
+// Check if zstd CLI is available
+let hasZstd = false;
+try {
+  await $`which zstd`.quiet();
+  hasZstd = true;
+} catch {
+  console.warn("Warning: zstd not found, skipping .zst generation");
+}
+
+async function* walkDir(dir: string): AsyncGenerator<string> {
+  try {
+    const entries = await readdir(dir, { withFileTypes: true });
+    for (const entry of entries) {
+      const path = join(dir, entry.name);
+      if (entry.isDirectory()) {
+        yield* walkDir(path);
+      } else if (entry.isFile()) {
+        yield path;
+      }
+    }
+  } catch {
+    // Directory doesn't exist, skip
+  }
+}
+
+async function compressFile(path: string): Promise<void> {
+  const ext = extname(path);
+
+  if (!COMPRESSIBLE_EXTENSIONS.has(ext)) return;
+  if (path.endsWith(".br") || path.endsWith(".gz") || path.endsWith(".zst")) return;
+
+  const stats = await stat(path);
+  if (stats.size < MIN_SIZE) return;
+
+  // Skip if all compressed variants already exist
+  const variantsExist = await Promise.all([
+    stat(`${path}.br`).then(
+      () => true,
+      () => false
+    ),
+    stat(`${path}.gz`).then(
+      () => true,
+      () => false
+    ),
+    hasZstd
+      ? stat(`${path}.zst`).then(
+          () => true,
+          () => false
+        )
+      : Promise.resolve(false),
+  ]);
+
+  if (variantsExist.every((exists) => exists || !hasZstd)) {
+    return;
+  }
+
+  const content = await readFile(path);
+  const originalSize = content.length;
+
+  // Brotli (maximum quality = 11)
+  const brContent = brotliCompressSync(content, {
+    params: {
+      [constants.BROTLI_PARAM_QUALITY]: 11,
+    },
+  });
+  await writeFile(`${path}.br`, brContent);
+
+  // Gzip (level 9)
+  const gzContent = gzipSync(content, { level: 9 });
+  await writeFile(`${path}.gz`, gzContent);
+
+  // Zstd (level 19 - maximum)
+  if (hasZstd) {
+    try {
+      await $`zstd -19 -q -f -o ${path}.zst ${path}`.quiet();
+    } catch (e) {
+      console.warn(`Warning: Failed to compress ${path} with zstd: ${e}`);
+    }
+  }
+
+  const brRatio = ((brContent.length / originalSize) * 100).toFixed(1);
+  const gzRatio = ((gzContent.length / originalSize) * 100).toFixed(1);
+  console.log(`Compressed: ${path} (br: ${brRatio}%, gz: ${gzRatio}%, ${originalSize} bytes)`);
+}
+
+async function main() {
+  console.log("Pre-compressing static assets...");
+
+  // Banner uses adapter-static with output in dist/
+  const dirs = ["dist"];
+  let scannedFiles = 0;
+  let compressedFiles = 0;
+
+  for (const dir of dirs) {
+    for await (const file of walkDir(dir)) {
+      const ext = extname(file);
+      scannedFiles++;
+
+      if (
+        COMPRESSIBLE_EXTENSIONS.has(ext) &&
+        !file.endsWith(".br") &&
+        !file.endsWith(".gz") &&
+        !file.endsWith(".zst")
+      ) {
+        const stats = await stat(file);
+        if (stats.size >= MIN_SIZE) {
+          await compressFile(file);
+          compressedFiles++;
+        }
+      }
+    }
+  }
+
+  console.log(`Done! Scanned ${scannedFiles} files, compressed ${compressedFiles} files.`);
+}
+
+main().catch((e) => {
+  console.error("Compression failed:", e);
+  process.exit(1);
+});
@@ -1,54 +0,0 @@
-.App {
-  min-height: 100vh;
-  font-family: -apple-system, BlinkMacSystemFont, "Segoe UI", "Roboto", "Oxygen", "Ubuntu",
-    "Cantarell", "Fira Sans", "Droid Sans", "Helvetica Neue", sans-serif;
-  background-color: var(--color-background);
-  color: var(--color-text);
-}
-
-@keyframes pulse {
-  0%,
-  100% {
-    opacity: 0.2;
-  }
-  50% {
-    opacity: 0.4;
-  }
-}
-
-.animate-pulse {
-  animation: pulse 2s ease-in-out infinite;
-}
-
-/* Theme toggle button */
-.theme-toggle {
-  cursor: pointer;
-  background-color: transparent;
-  border: none;
-  margin: 4px;
-  padding: 7px;
-  border-radius: 6px;
-  display: flex;
-  align-items: center;
-  justify-content: center;
-  color: var(--gray-11);
-  transition: background-color 0.2s, color 0.2s;
-  transform: scale(1.25);
-}
-
-.theme-toggle:hover {
-  background-color: var(--gray-4);
-}
-
-/* Screen reader only text */
-.sr-only {
-  position: absolute;
-  width: 1px;
-  height: 1px;
-  padding: 0;
-  margin: -1px;
-  overflow: hidden;
-  clip: rect(0, 0, 0, 0);
-  white-space: nowrap;
-  border: 0;
-}
@@ -0,0 +1,11 @@
+/// <reference types="@sveltejs/kit" />
+
+declare const __APP_VERSION__: string;
+
+declare namespace App {
+  // interface Error {}
+  // interface Locals {}
+  // interface PageData {}
+  // interface PageState {}
+  // interface Platform {}
+}
@@ -0,0 +1,32 @@
+<!doctype html>
+<html lang="en">
+  <head>
+    <meta charset="utf-8" />
+    <meta name="viewport" content="width=device-width, initial-scale=1" />
+    <link rel="icon" href="%sveltekit.assets%/favicon.ico" />
+    <meta name="theme-color" content="#000000" />
+    <meta
+      name="description"
+      content="Banner, a Discord bot and web interface for UTSA Course Monitoring"
+    />
+    <link rel="apple-touch-icon" href="%sveltekit.assets%/logo192.png" />
+    <link rel="manifest" href="%sveltekit.assets%/manifest.json" />
+    <title>Banner</title>
+    <script>
+      (function () {
+        var stored = localStorage.getItem("theme");
+        var isDark =
+          stored === "dark" ||
+          (stored !== "light" &&
+            window.matchMedia("(prefers-color-scheme: dark)").matches);
+        if (isDark) {
+          document.documentElement.classList.add("dark");
+        }
+      })();
+    </script>
+    %sveltekit.head%
+  </head>
+  <body data-sveltekit-preload-data="hover">
+    <div style="display: contents">%sveltekit.body%</div>
+  </body>
+</html>
@@ -1,36 +0,0 @@
-import { Button } from "@radix-ui/themes";
-import { Monitor, Moon, Sun } from "lucide-react";
-import { useTheme } from "next-themes";
-import { useMemo } from "react";
-
-export function ThemeToggle() {
-  const { theme, setTheme } = useTheme();
-
-  const nextTheme = useMemo(() => {
-    switch (theme) {
-      case "light":
-        return "dark";
-      case "dark":
-        return "system";
-      case "system":
-        return "light";
-      default:
-        console.error(`Invalid theme: ${theme}`);
-        return "system";
-    }
-  }, [theme]);
-
-  const icon = useMemo(() => {
-    if (nextTheme === "system") {
-      return <Monitor size={18} />;
-    }
-    return nextTheme === "dark" ? <Moon size={18} /> : <Sun size={18} />;
-  }, [nextTheme]);
-
-  return (
-    <Button variant="ghost" size="3" onClick={() => setTheme(nextTheme)} className="theme-toggle">
-      {icon}
-      <span className="sr-only">Toggle theme</span>
-    </Button>
-  );
-}
@@ -1,7 +1,6 @@
 import { beforeEach, describe, expect, it, vi } from "vitest";
 import { BannerApiClient } from "./api";

-// Mock fetch
 global.fetch = vi.fn();

 describe("BannerApiClient", () => {
@@ -12,23 +11,6 @@ describe("BannerApiClient", () => {
    vi.clearAllMocks();
  });

-  it("should fetch health data", async () => {
-    const mockHealth = {
-      status: "healthy",
-      timestamp: "2024-01-01T00:00:00Z",
-    };
-
-    vi.mocked(fetch).mockResolvedValueOnce({
-      ok: true,
-      json: () => Promise.resolve(mockHealth),
-    } as Response);
-
-    const result = await apiClient.getHealth();
-
-    expect(fetch).toHaveBeenCalledWith("/api/health");
-    expect(result).toEqual(mockHealth);
-  });
-
  it("should fetch status data", async () => {
    const mockStatus = {
      status: "active" as const,
@@ -58,8 +40,105 @@ describe("BannerApiClient", () => {
      statusText: "Internal Server Error",
    } as Response);

-    await expect(apiClient.getHealth()).rejects.toThrow(
+    await expect(apiClient.getStatus()).rejects.toThrow(
      "API request failed: 500 Internal Server Error"
    );
  });
+
+  it("should search courses with all params", async () => {
+    const mockResponse = {
+      courses: [],
+      totalCount: 0,
+      offset: 0,
+      limit: 25,
+    };
+
+    vi.mocked(fetch).mockResolvedValueOnce({
+      ok: true,
+      json: () => Promise.resolve(mockResponse),
+    } as Response);
+
+    const result = await apiClient.searchCourses({
+      term: "202420",
+      subjects: ["CS"],
+      q: "data",
+      open_only: true,
+      limit: 25,
+      offset: 50,
+    });
+
+    expect(fetch).toHaveBeenCalledWith(
+      "/api/courses/search?term=202420&subject=CS&q=data&open_only=true&limit=25&offset=50"
+    );
+    expect(result).toEqual(mockResponse);
+  });
+
+  it("should search courses with minimal params", async () => {
+    const mockResponse = {
+      courses: [],
+      totalCount: 0,
+      offset: 0,
+      limit: 25,
+    };
+
+    vi.mocked(fetch).mockResolvedValueOnce({
+      ok: true,
+      json: () => Promise.resolve(mockResponse),
+    } as Response);
+
+    await apiClient.searchCourses({ term: "202420" });
+
+    expect(fetch).toHaveBeenCalledWith("/api/courses/search?term=202420");
+  });
+
+  it("should fetch terms", async () => {
+    const mockTerms = [
+      { code: "202420", description: "Fall 2024" },
+      { code: "202510", description: "Spring 2025" },
+    ];
+
+    vi.mocked(fetch).mockResolvedValueOnce({
+      ok: true,
+      json: () => Promise.resolve(mockTerms),
+    } as Response);
+
+    const result = await apiClient.getTerms();
+
+    expect(fetch).toHaveBeenCalledWith("/api/terms");
+    expect(result).toEqual(mockTerms);
+  });
+
+  it("should fetch subjects for a term", async () => {
+    const mockSubjects = [
+      { code: "CS", description: "Computer Science" },
+      { code: "MAT", description: "Mathematics" },
+    ];
+
+    vi.mocked(fetch).mockResolvedValueOnce({
+      ok: true,
+      json: () => Promise.resolve(mockSubjects),
+    } as Response);
+
+    const result = await apiClient.getSubjects("202420");
+
+    expect(fetch).toHaveBeenCalledWith("/api/subjects?term=202420");
+    expect(result).toEqual(mockSubjects);
+  });
+
+  it("should fetch reference data", async () => {
+    const mockRef = [
+      { code: "F", description: "Face to Face" },
+      { code: "OL", description: "Online" },
+    ];
+
+    vi.mocked(fetch).mockResolvedValueOnce({
+      ok: true,
+      json: () => Promise.resolve(mockRef),
+    } as Response);
+
+    const result = await apiClient.getReference("instructional_methods");
+
+    expect(fetch).toHaveBeenCalledWith("/api/reference/instructional_methods");
+    expect(result).toEqual(mockRef);
+  });
 });
@@ -1,41 +1,216 @@
-// API client for Banner backend
+import type {
+  CandidateResponse,
+  CodeDescription,
+  CourseResponse,
+  DbMeetingTime,
+  InstructorDetail,
+  InstructorDetailResponse,
+  InstructorListItem,
+  InstructorResponse,
+  InstructorStats,
+  LinkedRmpProfile,
+  ListInstructorsResponse,
+  RescoreResponse,
+  ScraperStatsResponse,
+  SearchResponse as SearchResponseGenerated,
+  ServiceInfo,
+  ServiceStatus,
+  StatusResponse,
+  SubjectDetailResponse,
+  SubjectResultEntry,
+  SubjectSummary,
+  SubjectsResponse,
+  TimeseriesPoint,
+  TimeseriesResponse,
+  TopCandidateResponse,
+  User,
+} from "$lib/bindings";
+
 const API_BASE_URL = "/api";

-export interface HealthResponse {
-  status: string;
+// Re-export generated types under their canonical names
+export type {
+  CandidateResponse,
+  CodeDescription,
+  CourseResponse,
+  DbMeetingTime,
+  InstructorDetail,
+  InstructorDetailResponse,
+  InstructorListItem,
+  InstructorResponse,
+  InstructorStats,
+  LinkedRmpProfile,
+  ListInstructorsResponse,
+  RescoreResponse,
+  ScraperStatsResponse,
+  ServiceInfo,
+  ServiceStatus,
+  StatusResponse,
+  SubjectDetailResponse,
+  SubjectResultEntry,
+  SubjectSummary,
+  SubjectsResponse,
+  TimeseriesPoint,
+  TimeseriesResponse,
+  TopCandidateResponse,
+};
+
+// Semantic aliases — these all share the CodeDescription shape
+export type Term = CodeDescription;
+export type Subject = CodeDescription;
+export type ReferenceEntry = CodeDescription;
+
+// SearchResponse re-exported (aliased to strip the "Generated" suffix)
+export type SearchResponse = SearchResponseGenerated;
+
+export type ScraperPeriod = "1h" | "6h" | "24h" | "7d" | "30d";
+
+// Client-side only — not generated from Rust
+export type SortColumn = "course_code" | "title" | "instructor" | "time" | "seats";
+export type SortDirection = "asc" | "desc";
+
+export interface AdminStatus {
+  userCount: number;
+  sessionCount: number;
+  courseCount: number;
+  scrapeJobCount: number;
+  services: { name: string; status: string }[];
+}
+
+export interface ScrapeJob {
+  id: number;
+  targetType: string;
+  targetPayload: unknown;
+  priority: string;
+  executeAt: string;
+  createdAt: string;
+  lockedAt: string | null;
+  retryCount: number;
+  maxRetries: number;
+  queuedAt: string;
+  status: "processing" | "staleLock" | "exhausted" | "scheduled" | "pending";
+}
+
+export interface ScrapeJobsResponse {
+  jobs: ScrapeJob[];
+}
+
+export interface AuditLogEntry {
+  id: number;
+  courseId: number;
  timestamp: string;
+  fieldChanged: string;
+  oldValue: string;
+  newValue: string;
+  subject: string | null;
+  courseNumber: string | null;
+  crn: string | null;
+  courseTitle: string | null;
 }

-export type Status = "starting" | "active" | "connected" | "disabled" | "error";
-
-export interface ServiceInfo {
-  name: string;
-  status: Status;
+export interface AuditLogResponse {
+  entries: AuditLogEntry[];
 }

-export interface StatusResponse {
-  status: Status;
-  version: string;
-  commit: string;
-  services: Record<string, ServiceInfo>;
+export interface MetricEntry {
+  id: number;
+  courseId: number;
+  timestamp: string;
+  enrollment: number;
+  waitCount: number;
+  seatsAvailable: number;
 }

 export interface MetricsResponse {
-  banner_api: {
-    status: string;
-  };
+  metrics: MetricEntry[];
+  count: number;
  timestamp: string;
 }

+export interface MetricsParams {
+  course_id?: number;
+  term?: string;
+  crn?: string;
+  range?: "1h" | "6h" | "24h" | "7d" | "30d";
+  limit?: number;
+}
+
+/** A time range for timeline queries (ISO-8601 strings). */
+export interface TimelineRange {
+  start: string;
+  end: string;
+}
+
+/** Request body for POST /api/timeline. */
+export interface TimelineRequest {
+  ranges: TimelineRange[];
+}
+
+/** A single 15-minute slot returned by the timeline API. */
+export interface TimelineSlot {
+  time: string;
+  subjects: Record<string, number>;
+}
+
+/** Response from POST /api/timeline. */
+export interface TimelineResponse {
+  slots: TimelineSlot[];
+  subjects: string[];
+}
+
+export interface SearchParams {
+  term: string;
+  subjects?: string[];
+  q?: string;
+  open_only?: boolean;
+  limit?: number;
+  offset?: number;
+  sort_by?: SortColumn;
+  sort_dir?: SortDirection;
+}
+
+// Admin instructor query params (client-only, not generated)
+export interface AdminInstructorListParams {
+  status?: string;
+  search?: string;
+  page?: number;
+  per_page?: number;
+  sort?: string;
+}
+
 export class BannerApiClient {
  private baseUrl: string;
+  private fetchFn: typeof fetch;

-  constructor(baseUrl: string = API_BASE_URL) {
+  constructor(baseUrl: string = API_BASE_URL, fetchFn: typeof fetch = fetch) {
    this.baseUrl = baseUrl;
+    this.fetchFn = fetchFn;
  }

-  private async request<T>(endpoint: string): Promise<T> {
-    const response = await fetch(`${this.baseUrl}${endpoint}`);
+  private buildInit(options?: { method?: string; body?: unknown }): RequestInit | undefined {
+    if (!options) return undefined;
+    const init: RequestInit = {};
+    if (options.method) {
+      init.method = options.method;
+    }
+    if (options.body !== undefined) {
+      init.headers = { "Content-Type": "application/json" };
+      init.body = JSON.stringify(options.body);
+    } else if (options.method) {
+      init.headers = { "Content-Type": "application/json" };
+    }
+    return Object.keys(init).length > 0 ? init : undefined;
+  }
+
+  private async request<T>(
+    endpoint: string,
+    options?: { method?: string; body?: unknown }
+  ): Promise<T> {
+    const init = this.buildInit(options);
+    const args: [string, RequestInit?] = [`${this.baseUrl}${endpoint}`];
+    if (init) args.push(init);
+
+    const response = await this.fetchFn(...args);

    if (!response.ok) {
      throw new Error(`API request failed: ${response.status} ${response.statusText}`);
@@ -44,18 +219,200 @@ export class BannerApiClient {
    return (await response.json()) as T;
  }

-  async getHealth(): Promise<HealthResponse> {
-    return this.request<HealthResponse>("/health");
+  private async requestVoid(
+    endpoint: string,
+    options?: { method?: string; body?: unknown }
+  ): Promise<void> {
+    const init = this.buildInit(options);
+    const args: [string, RequestInit?] = [`${this.baseUrl}${endpoint}`];
+    if (init) args.push(init);
+
+    const response = await this.fetchFn(...args);
+
+    if (!response.ok) {
+      throw new Error(`API request failed: ${response.status} ${response.statusText}`);
+    }
  }

  async getStatus(): Promise<StatusResponse> {
    return this.request<StatusResponse>("/status");
  }

-  async getMetrics(): Promise<MetricsResponse> {
-    return this.request<MetricsResponse>("/metrics");
+  async searchCourses(params: SearchParams): Promise<SearchResponse> {
+    const query = new URLSearchParams();
+    query.set("term", params.term);
+    if (params.subjects) {
+      for (const s of params.subjects) {
+        query.append("subject", s);
+      }
+    }
+    if (params.q) query.set("q", params.q);
+    if (params.open_only) query.set("open_only", "true");
+    if (params.limit !== undefined) query.set("limit", String(params.limit));
+    if (params.offset !== undefined) query.set("offset", String(params.offset));
+    if (params.sort_by) query.set("sort_by", params.sort_by);
+    if (params.sort_dir) query.set("sort_dir", params.sort_dir);
+    return this.request<SearchResponse>(`/courses/search?${query.toString()}`);
+  }
+
+  async getTerms(): Promise<Term[]> {
+    return this.request<Term[]>("/terms");
+  }
+
+  async getSubjects(termCode: string): Promise<Subject[]> {
+    return this.request<Subject[]>(`/subjects?term=${encodeURIComponent(termCode)}`);
+  }
+
+  async getReference(category: string): Promise<ReferenceEntry[]> {
+    return this.request<ReferenceEntry[]>(`/reference/${encodeURIComponent(category)}`);
+  }
+
+  // Admin endpoints
+  async getAdminStatus(): Promise<AdminStatus> {
+    return this.request<AdminStatus>("/admin/status");
+  }
+
+  async getAdminUsers(): Promise<User[]> {
+    return this.request<User[]>("/admin/users");
+  }
+
+  async setUserAdmin(discordId: string, isAdmin: boolean): Promise<User> {
+    return this.request<User>(`/admin/users/${discordId}/admin`, {
+      method: "PUT",
+      body: { is_admin: isAdmin },
+    });
+  }
+
+  async getAdminScrapeJobs(): Promise<ScrapeJobsResponse> {
+    return this.request<ScrapeJobsResponse>("/admin/scrape-jobs");
+  }
+
+  /**
+   * Fetch the audit log with conditional request support.
+   *
+   * Returns `null` when the server responds 304 (data unchanged).
+   * Stores and sends `Last-Modified` / `If-Modified-Since` automatically.
+   */
+  async getAdminAuditLog(): Promise<AuditLogResponse | null> {
+    const headers: Record<string, string> = {};
+    if (this._auditLastModified) {
+      headers["If-Modified-Since"] = this._auditLastModified;
+    }
+
+    const response = await this.fetchFn(`${this.baseUrl}/admin/audit-log`, { headers });
+
+    if (response.status === 304) {
+      return null;
+    }
+
+    if (!response.ok) {
+      throw new Error(`API request failed: ${response.status} ${response.statusText}`);
+    }
+
+    const lastMod = response.headers.get("Last-Modified");
+    if (lastMod) {
+      this._auditLastModified = lastMod;
+    }
+
+    return (await response.json()) as AuditLogResponse;
+  }
+
+  /** Stored `Last-Modified` value for audit log conditional requests. */
+  private _auditLastModified: string | null = null;
+
+  async getTimeline(ranges: TimelineRange[]): Promise<TimelineResponse> {
+    return this.request<TimelineResponse>("/timeline", {
+      method: "POST",
+      body: { ranges } satisfies TimelineRequest,
+    });
+  }
+
+  async getMetrics(params?: MetricsParams): Promise<MetricsResponse> {
+    const query = new URLSearchParams();
+    if (params?.course_id !== undefined) query.set("course_id", String(params.course_id));
+    if (params?.term) query.set("term", params.term);
+    if (params?.crn) query.set("crn", params.crn);
+    if (params?.range) query.set("range", params.range);
+    if (params?.limit !== undefined) query.set("limit", String(params.limit));
+    const qs = query.toString();
+    return this.request<MetricsResponse>(`/metrics${qs ? `?${qs}` : ""}`);
+  }
+
+  // Admin instructor endpoints
+
+  async getAdminInstructors(params?: AdminInstructorListParams): Promise<ListInstructorsResponse> {
+    const query = new URLSearchParams();
+    if (params?.status) query.set("status", params.status);
+    if (params?.search) query.set("search", params.search);
+    if (params?.page !== undefined) query.set("page", String(params.page));
+    if (params?.per_page !== undefined) query.set("per_page", String(params.per_page));
+    if (params?.sort) query.set("sort", params.sort);
+    const qs = query.toString();
+    return this.request<ListInstructorsResponse>(`/admin/instructors${qs ? `?${qs}` : ""}`);
+  }
+
+  async getAdminInstructor(id: number): Promise<InstructorDetailResponse> {
+    return this.request<InstructorDetailResponse>(`/admin/instructors/${id}`);
+  }
+
+  async matchInstructor(id: number, rmpLegacyId: number): Promise<InstructorDetailResponse> {
+    return this.request<InstructorDetailResponse>(`/admin/instructors/${id}/match`, {
+      method: "POST",
+      body: { rmpLegacyId },
+    });
+  }
+
+  async rejectCandidate(id: number, rmpLegacyId: number): Promise<void> {
+    return this.requestVoid(`/admin/instructors/${id}/reject-candidate`, {
+      method: "POST",
+      body: { rmpLegacyId },
+    });
+  }
+
+  async rejectAllCandidates(id: number): Promise<void> {
+    return this.requestVoid(`/admin/instructors/${id}/reject-all`, {
+      method: "POST",
+    });
+  }
+
+  async unmatchInstructor(id: number, rmpLegacyId?: number): Promise<void> {
+    return this.requestVoid(`/admin/instructors/${id}/unmatch`, {
+      method: "POST",
+      ...(rmpLegacyId !== undefined ? { body: { rmpLegacyId } } : {}),
+    });
+  }
+
+  async rescoreInstructors(): Promise<RescoreResponse> {
+    return this.request<RescoreResponse>("/admin/rmp/rescore", {
+      method: "POST",
+    });
+  }
+
+  // Scraper analytics endpoints
+
+  async getScraperStats(period?: ScraperPeriod): Promise<ScraperStatsResponse> {
+    const qs = period ? `?period=${period}` : "";
+    return this.request<ScraperStatsResponse>(`/admin/scraper/stats${qs}`);
+  }
+
+  async getScraperTimeseries(period?: ScraperPeriod, bucket?: string): Promise<TimeseriesResponse> {
+    const query = new URLSearchParams();
+    if (period) query.set("period", period);
+    if (bucket) query.set("bucket", bucket);
+    const qs = query.toString();
+    return this.request<TimeseriesResponse>(`/admin/scraper/timeseries${qs ? `?${qs}` : ""}`);
+  }
+
+  async getScraperSubjects(): Promise<SubjectsResponse> {
+    return this.request<SubjectsResponse>("/admin/scraper/subjects");
+  }
+
+  async getScraperSubjectDetail(subject: string, limit?: number): Promise<SubjectDetailResponse> {
+    const qs = limit !== undefined ? `?limit=${limit}` : "";
+    return this.request<SubjectDetailResponse>(
+      `/admin/scraper/subjects/${encodeURIComponent(subject)}${qs}`
+    );
  }
 }

-// Export a default instance
 export const client = new BannerApiClient();
@@ -0,0 +1,77 @@
+import type { User } from "$lib/bindings";
+
+type AuthState =
+  | { mode: "loading" }
+  | { mode: "authenticated"; user: User }
+  | { mode: "unauthenticated" };
+
+class AuthStore {
+  state = $state<AuthState>({ mode: "loading" });
+
+  get user(): User | null {
+    return this.state.mode === "authenticated" ? this.state.user : null;
+  }
+
+  get isAdmin(): boolean {
+    return this.user?.isAdmin ?? false;
+  }
+
+  get isLoading(): boolean {
+    return this.state.mode === "loading";
+  }
+
+  get isAuthenticated(): boolean {
+    return this.state.mode === "authenticated";
+  }
+
+  /**
+   * Attempt to load the current user session from the backend.
+   * Only transitions to "unauthenticated" on a definitive 401/403.
+   * Retries indefinitely on transient failures (network errors, 5xx)
+   * so that a slow backend startup doesn't kick the user to login.
+   */
+  async init() {
+    const MAX_DELAY_MS = 7_000;
+    let delayMs = 500;
+
+    for (;;) {
+      try {
+        const response = await fetch("/api/auth/me");
+
+        if (response.ok) {
+          const user: User = await response.json();
+          this.state = { mode: "authenticated", user };
+          return;
+        }
+
+        // Definitive rejection — no session or not authorized
+        if (response.status === 401 || response.status === 403) {
+          this.state = { mode: "unauthenticated" };
+          return;
+        }
+
+        // Server error (5xx) or unexpected status — retry
+      } catch {
+        // Network error (backend not up yet) — retry
+      }
+
+      await new Promise((r) => setTimeout(r, delayMs));
+      delayMs = Math.min(delayMs * 2, MAX_DELAY_MS);
+    }
+  }
+
+  login() {
+    window.location.href = "/api/auth/login";
+  }
+
+  async logout() {
+    try {
+      await fetch("/api/auth/logout", { method: "POST" });
+    } finally {
+      this.state = { mode: "unauthenticated" };
+      window.location.href = "/";
+    }
+  }
+}
+
+export const authStore = new AuthStore();
@@ -0,0 +1,26 @@
+export type { CandidateResponse } from "./CandidateResponse";
+export type { CodeDescription } from "./CodeDescription";
+export type { CourseResponse } from "./CourseResponse";
+export type { DbMeetingTime } from "./DbMeetingTime";
+export type { InstructorDetail } from "./InstructorDetail";
+export type { InstructorDetailResponse } from "./InstructorDetailResponse";
+export type { InstructorListItem } from "./InstructorListItem";
+export type { InstructorResponse } from "./InstructorResponse";
+export type { InstructorStats } from "./InstructorStats";
+export type { LinkedRmpProfile } from "./LinkedRmpProfile";
+export type { ListInstructorsResponse } from "./ListInstructorsResponse";
+export type { OkResponse } from "./OkResponse";
+export type { RescoreResponse } from "./RescoreResponse";
+export type { ScraperStatsResponse } from "./ScraperStatsResponse";
+export type { SearchResponse } from "./SearchResponse";
+export type { ServiceInfo } from "./ServiceInfo";
+export type { ServiceStatus } from "./ServiceStatus";
+export type { StatusResponse } from "./StatusResponse";
+export type { SubjectDetailResponse } from "./SubjectDetailResponse";
+export type { SubjectResultEntry } from "./SubjectResultEntry";
+export type { SubjectSummary } from "./SubjectSummary";
+export type { SubjectsResponse } from "./SubjectsResponse";
+export type { TimeseriesPoint } from "./TimeseriesPoint";
+export type { TimeseriesResponse } from "./TimeseriesResponse";
+export type { TopCandidateResponse } from "./TopCandidateResponse";
+export type { User } from "./User";
@@ -0,0 +1,352 @@
+<script lang="ts">
+import type { CourseResponse } from "$lib/api";
+import {
+  formatTime,
+  formatCreditHours,
+  formatDate,
+  formatMeetingDaysLong,
+  isMeetingTimeTBA,
+  isTimeTBA,
+  ratingStyle,
+  rmpUrl,
+  RMP_CONFIDENCE_THRESHOLD,
+} from "$lib/course";
+import { themeStore } from "$lib/stores/theme.svelte";
+import { useClipboard } from "$lib/composables/useClipboard.svelte";
+import { cn, tooltipContentClass, formatNumber } from "$lib/utils";
+import { Tooltip } from "bits-ui";
+import SimpleTooltip from "./SimpleTooltip.svelte";
+import {
+  Info,
+  Copy,
+  Check,
+  Star,
+  Triangle,
+  ExternalLink,
+  Calendar,
+  Download,
+} from "@lucide/svelte";
+
+let { course }: { course: CourseResponse } = $props();
+
+const clipboard = useClipboard();
+</script>
+
+<div class="bg-muted/60 p-5 text-sm border-b border-border">
+    <div class="grid grid-cols-1 sm:grid-cols-2 gap-5">
+        <!-- Instructors -->
+        <div>
+            <h4 class="text-sm text-foreground mb-2">Instructors</h4>
+            {#if course.instructors.length > 0}
+                <div class="flex flex-wrap gap-1.5">
+                    {#each course.instructors as instructor}
+                        <Tooltip.Root delayDuration={200}>
+                            <Tooltip.Trigger>
+                                <span
+                                    class="inline-flex items-center gap-1.5 text-sm font-medium bg-card border border-border rounded-md px-2.5 py-1 text-foreground hover:border-foreground/20 hover:bg-card/80 transition-colors"
+                                >
+                                    {instructor.displayName}
+                                    {#if instructor.rmpRating != null}
+                                        {@const rating = instructor.rmpRating}
+                                        {@const lowConfidence =
+                                            (instructor.rmpNumRatings ?? 0) <
+                                            RMP_CONFIDENCE_THRESHOLD}
+                                        <span
+                                            class="text-[10px] font-semibold inline-flex items-center gap-0.5"
+                                            style={ratingStyle(
+                                                rating,
+                                                themeStore.isDark,
+                                            )}
+                                        >
+                                            {rating.toFixed(1)}
+                                            {#if lowConfidence}
+                                                <Triangle
+                                                    class="size-2 fill-current"
+                                                />
+                                            {:else}
+                                                <Star
+                                                    class="size-2.5 fill-current"
+                                                />
+                                            {/if}
+                                        </span>
+                                    {/if}
+                                </span>
+                            </Tooltip.Trigger>
+                            <Tooltip.Content
+                                sideOffset={6}
+                                class={cn(tooltipContentClass, "px-3 py-2")}
+                            >
+                                <div class="space-y-1.5">
+                                    <div class="font-medium">
+                                        {instructor.displayName}
+                                    </div>
+                                    {#if instructor.isPrimary}
+                                        <div class="text-muted-foreground">
+                                            Primary instructor
+                                        </div>
+                                    {/if}
+                                    {#if instructor.rmpRating != null}
+                                        <div class="text-muted-foreground">
+                                            {instructor.rmpRating.toFixed(1)}/5
+                                            · {instructor.rmpNumRatings ?? 0} ratings
+                                            {#if (instructor.rmpNumRatings ?? 0) < RMP_CONFIDENCE_THRESHOLD}
+                                                (low)
+                                            {/if}
+                                        </div>
+                                    {/if}
+                                    {#if instructor.rmpLegacyId != null}
+                                        <a
+                                            href={rmpUrl(
+                                                instructor.rmpLegacyId,
+                                            )}
+                                            target="_blank"
+                                            rel="noopener"
+                                            class="inline-flex items-center gap-1 text-muted-foreground hover:text-foreground transition-colors"
+                                        >
+                                            <ExternalLink class="size-3" />
+                                            <span>View on RMP</span>
+                                        </a>
+                                    {/if}
+                                    {#if instructor.email}
+                                        <button
+                                            onclick={(e) =>
+                                                clipboard.copy(
+                                                    instructor.email!,
+                                                    e,
+                                                )}
+                                            class="inline-flex items-center gap-1 text-muted-foreground hover:text-foreground transition-colors cursor-pointer"
+                                        >
+                                            {#if clipboard.copiedValue === instructor.email}
+                                                <Check class="size-3" />
+                                                <span>Copied!</span>
+                                            {:else}
+                                                <Copy class="size-3" />
+                                                <span>{instructor.email}</span>
+                                            {/if}
+                                        </button>
+                                    {/if}
+                                </div>
+                            </Tooltip.Content>
+                        </Tooltip.Root>
+                    {/each}
+                </div>
+            {:else}
+                <span class="text-muted-foreground italic">Staff</span>
+            {/if}
+        </div>
+
+        <!-- Meeting Times -->
+        <div>
+            <h4 class="text-sm text-foreground mb-2">Meeting Times</h4>
+            {#if course.meetingTimes.length > 0}
+                <ul class="space-y-2">
+                    {#each course.meetingTimes as mt}
+                        <li>
+                            {#if isMeetingTimeTBA(mt) && isTimeTBA(mt)}
+                                <span class="italic text-muted-foreground"
+                                    >TBA</span
+                                >
+                            {:else}
+                                <div class="flex items-baseline gap-1.5">
+                                    {#if !isMeetingTimeTBA(mt)}
+                                        <span
+                                            class="font-medium text-foreground"
+                                        >
+                                            {formatMeetingDaysLong(mt)}
+                                        </span>
+                                    {/if}
+                                    {#if !isTimeTBA(mt)}
+                                        <span class="text-muted-foreground">
+                                            {formatTime(
+                                                mt.begin_time,
+                                            )}&ndash;{formatTime(mt.end_time)}
+                                        </span>
+                                    {:else}
+                                        <span
+                                            class="italic text-muted-foreground"
+                                            >Time TBA</span
+                                        >
+                                    {/if}
+                                </div>
+                            {/if}
+                            {#if mt.building || mt.room}
+                                <div
+                                    class="text-xs text-muted-foreground mt-0.5"
+                                >
+                                    {mt.building_description ??
+                                        mt.building}{mt.room
+                                        ? ` ${mt.room}`
+                                        : ""}
+                                </div>
+                            {/if}
+                            <div
+                                class="text-xs text-muted-foreground/70 mt-0.5"
+                            >
+                                {formatDate(mt.start_date)} &ndash; {formatDate(
+                                    mt.end_date,
+                                )}
+                            </div>
+                        </li>
+                    {/each}
+                </ul>
+            {:else}
+                <span class="italic text-muted-foreground">TBA</span>
+            {/if}
+        </div>
+
+        <!-- Delivery -->
+        <div>
+            <h4 class="text-sm text-foreground mb-2">
+                <span class="inline-flex items-center gap-1">
+                    Delivery
+                    <SimpleTooltip
+                        text="How the course is taught: in-person, online, hybrid, etc."
+                        delay={150}
+                        passthrough
+                    >
+                        <Info class="size-3 text-muted-foreground/50" />
+                    </SimpleTooltip>
+                </span>
+            </h4>
+            <span class="text-foreground">
+                {course.instructionalMethod ?? "—"}
+                {#if course.campus}
+                    <span class="text-muted-foreground">
+                        · {course.campus}
+                    </span>
+                {/if}
+            </span>
+        </div>
+
+        <!-- Credits -->
+        <div>
+            <h4 class="text-sm text-foreground mb-2">Credits</h4>
+            <span class="text-foreground">{formatCreditHours(course)}</span>
+        </div>
+
+        <!-- Attributes -->
+        {#if course.attributes.length > 0}
+            <div>
+                <h4 class="text-sm text-foreground mb-2">
+                    <span class="inline-flex items-center gap-1">
+                        Attributes
+                        <SimpleTooltip
+                            text="Course flags for degree requirements, core curriculum, or special designations"
+                            delay={150}
+                            passthrough
+                        >
+                            <Info class="size-3 text-muted-foreground/50" />
+                        </SimpleTooltip>
+                    </span>
+                </h4>
+                <div class="flex flex-wrap gap-1.5">
+                    {#each course.attributes as attr}
+                        <SimpleTooltip
+                            text="Course attribute code"
+                            delay={150}
+                            passthrough
+                        >
+                            <span
+                                class="inline-flex text-xs font-medium bg-card border border-border rounded-md px-2 py-0.5 text-muted-foreground hover:text-foreground hover:border-foreground/20 transition-colors"
+                            >
+                                {attr}
+                            </span>
+                        </SimpleTooltip>
+                    {/each}
+                </div>
+            </div>
+        {/if}
+
+        <!-- Cross-list -->
+        {#if course.crossList}
+            <div>
+                <h4 class="text-sm text-foreground mb-2">
+                    <span class="inline-flex items-center gap-1">
+                        Cross-list
+                        <SimpleTooltip
+                            text="Cross-listed sections share enrollment across multiple course numbers. Students in any linked section attend the same class."
+                            delay={150}
+                            passthrough
+                        >
+                            <Info class="size-3 text-muted-foreground/50" />
+                        </SimpleTooltip>
+                    </span>
+                </h4>
+                <Tooltip.Root delayDuration={150} disableHoverableContent>
+                    <Tooltip.Trigger>
+                        <span
+                            class="inline-flex items-center gap-1.5 text-foreground font-mono"
+                        >
+                            <span
+                                class="bg-card border border-border rounded-md px-2 py-0.5 text-xs font-medium"
+                            >
+                                {course.crossList}
+                            </span>
+                            {#if course.crossListCount != null && course.crossListCapacity != null}
+                                <span class="text-muted-foreground text-xs">
+                                    {formatNumber(course.crossListCount)}/{formatNumber(course.crossListCapacity)}
+                                </span>
+                            {/if}
+                        </span>
+                    </Tooltip.Trigger>
+                    <Tooltip.Content sideOffset={6} class={tooltipContentClass}>
+                        Group <span class="font-mono font-medium"
+                            >{course.crossList}</span
+                        >
+                        {#if course.crossListCount != null && course.crossListCapacity != null}
+                            — {formatNumber(course.crossListCount)} enrolled across {formatNumber(course.crossListCapacity)}
+                            shared seats
+                        {/if}
+                    </Tooltip.Content>
+                </Tooltip.Root>
+            </div>
+        {/if}
+
+        <!-- Waitlist -->
+        {#if course.waitCapacity > 0}
+            <div>
+                <h4 class="text-sm text-foreground mb-2">Waitlist</h4>
+                <span class="text-2foreground"
+                    >{formatNumber(course.waitCount)} / {formatNumber(course.waitCapacity)}</span
+                >
+            </div>
+        {/if}
+
+        <!-- Calendar Export -->
+        {#if course.meetingTimes.length > 0}
+            <div>
+                <h4 class="text-sm text-foreground mb-2">
+                    <span class="inline-flex items-center gap-1">
+                        Calendar
+                        <SimpleTooltip
+                            text="Export this course schedule to your calendar app"
+                            delay={150}
+                            passthrough
+                        >
+                            <Info class="size-3 text-muted-foreground/50" />
+                        </SimpleTooltip>
+                    </span>
+                </h4>
+                <div class="flex flex-wrap gap-1.5">
+                    <a
+                        href="/api/courses/{course.termCode}/{course.crn}/calendar.ics"
+                        download
+                        class="inline-flex items-center gap-1.5 text-sm font-medium bg-card border border-border rounded-md px-2.5 py-1 text-foreground hover:border-foreground/20 hover:bg-card/80 transition-colors"
+                    >
+                        <Download class="size-3.5" />
+                        ICS File
+                    </a>
+                    <a
+                        href="/api/courses/{course.termCode}/{course.crn}/gcal"
+                        target="_blank"
+                        rel="noopener"
+                        class="inline-flex items-center gap-1.5 text-sm font-medium bg-card border border-border rounded-md px-2.5 py-1 text-foreground hover:border-foreground/20 hover:bg-card/80 transition-colors"
+                    >
+                        <Calendar class="size-3.5" />
+                        Google Calendar
+                    </a>
+                </div>
+            </div>
+        {/if}
+    </div>
+</div>
@@ -0,0 +1,787 @@
+<script lang="ts">
+import type { CourseResponse } from "$lib/api";
+import {
+  abbreviateInstructor,
+  concernAccentColor,
+  formatLocationDisplay,
+  formatLocationTooltip,
+  formatMeetingDays,
+  formatMeetingTimesTooltip,
+  formatTimeRange,
+  getDeliveryConcern,
+  getPrimaryInstructor,
+  isMeetingTimeTBA,
+  isTimeTBA,
+  openSeats,
+  seatsColor,
+  seatsDotColor,
+  ratingStyle,
+  rmpUrl,
+  RMP_CONFIDENCE_THRESHOLD,
+} from "$lib/course";
+import { themeStore } from "$lib/stores/theme.svelte";
+import { useClipboard } from "$lib/composables/useClipboard.svelte";
+import { useOverlayScrollbars } from "$lib/composables/useOverlayScrollbars.svelte";
+import CourseDetail from "./CourseDetail.svelte";
+import { fade, fly, slide } from "svelte/transition";
+import { flip } from "svelte/animate";
+import { createSvelteTable, FlexRender } from "$lib/components/ui/data-table/index.js";
+import {
+  getCoreRowModel,
+  getSortedRowModel,
+  type ColumnDef,
+  type SortingState,
+  type VisibilityState,
+  type Updater,
+} from "@tanstack/table-core";
+import {
+  ArrowUp,
+  ArrowDown,
+  ArrowUpDown,
+  Columns3,
+  Check,
+  RotateCcw,
+  Star,
+  Triangle,
+  ExternalLink,
+} from "@lucide/svelte";
+import { DropdownMenu, ContextMenu, Tooltip } from "bits-ui";
+import { cn, tooltipContentClass, formatNumber } from "$lib/utils";
+import SimpleTooltip from "./SimpleTooltip.svelte";
+
+let {
+  courses,
+  loading,
+  sorting = [],
+  onSortingChange,
+  manualSorting = false,
+  subjectMap = {},
+}: {
+  courses: CourseResponse[];
+  loading: boolean;
+  sorting?: SortingState;
+  onSortingChange?: (sorting: SortingState) => void;
+  manualSorting?: boolean;
+  subjectMap?: Record<string, string>;
+} = $props();
+
+let expandedCrn: string | null = $state(null);
+let tableWrapper: HTMLDivElement = undefined!;
+const clipboard = useClipboard(1000);
+
+// Collapse expanded row when the dataset changes to avoid stale detail rows
+// and FLIP position calculation glitches from lingering expanded content
+$effect(() => {
+  courses; // track dependency
+  expandedCrn = null;
+});
+
+useOverlayScrollbars(() => tableWrapper, {
+  overflow: { x: "scroll", y: "hidden" },
+  scrollbars: { autoHide: "never" },
+});
+
+// Column visibility state
+let columnVisibility: VisibilityState = $state({});
+
+function resetColumnVisibility() {
+  columnVisibility = {};
+}
+
+function handleVisibilityChange(updater: Updater<VisibilityState>) {
+  const newVisibility = typeof updater === "function" ? updater(columnVisibility) : updater;
+  columnVisibility = newVisibility;
+}
+
+// visibleColumnIds and hasCustomVisibility derived after column definitions below
+
+function toggleRow(crn: string) {
+  expandedCrn = expandedCrn === crn ? null : crn;
+}
+
+function primaryInstructorDisplay(course: CourseResponse): string {
+  const primary = getPrimaryInstructor(course.instructors);
+  if (!primary) return "Staff";
+  return abbreviateInstructor(primary.displayName);
+}
+
+function primaryRating(
+  course: CourseResponse
+): { rating: number; count: number; legacyId: number | null } | null {
+  const primary = getPrimaryInstructor(course.instructors);
+  if (!primary?.rmpRating) return null;
+  return {
+    rating: primary.rmpRating,
+    count: primary.rmpNumRatings ?? 0,
+    legacyId: primary.rmpLegacyId ?? null,
+  };
+}
+
+function timeIsTBA(course: CourseResponse): boolean {
+  if (course.meetingTimes.length === 0) return true;
+  const mt = course.meetingTimes[0];
+  return isMeetingTimeTBA(mt) && isTimeTBA(mt);
+}
+
+// Column definitions
+const columns: ColumnDef<CourseResponse, unknown>[] = [
+  {
+    id: "crn",
+    accessorKey: "crn",
+    header: "CRN",
+    enableSorting: false,
+  },
+  {
+    id: "course_code",
+    accessorFn: (row) => `${row.subject} ${row.courseNumber}`,
+    header: "Course",
+    enableSorting: true,
+  },
+  {
+    id: "title",
+    accessorKey: "title",
+    header: "Title",
+    enableSorting: true,
+  },
+  {
+    id: "instructor",
+    accessorFn: (row) => primaryInstructorDisplay(row),
+    header: "Instructor",
+    enableSorting: true,
+  },
+  {
+    id: "time",
+    accessorFn: (row) => {
+      if (row.meetingTimes.length === 0) return "";
+      const mt = row.meetingTimes[0];
+      return `${formatMeetingDays(mt)} ${formatTimeRange(mt.begin_time, mt.end_time)}`;
+    },
+    header: "Time",
+    enableSorting: true,
+  },
+  {
+    id: "location",
+    accessorFn: (row) => formatLocationDisplay(row) ?? "",
+    header: "Location",
+    enableSorting: false,
+  },
+  {
+    id: "seats",
+    accessorFn: (row) => openSeats(row),
+    header: "Seats",
+    enableSorting: true,
+  },
+];
+
+/** Column IDs that are currently visible */
+let visibleColumnIds = $derived(
+  columns.map((c) => c.id!).filter((id) => columnVisibility[id] !== false)
+);
+
+let hasCustomVisibility = $derived(Object.values(columnVisibility).some((v) => v === false));
+
+function handleSortingChange(updater: Updater<SortingState>) {
+  const newSorting = typeof updater === "function" ? updater(sorting) : updater;
+  onSortingChange?.(newSorting);
+}
+
+const table = createSvelteTable({
+  get data() {
+    return courses;
+  },
+  getRowId: (row) => String(row.crn),
+  columns,
+  state: {
+    get sorting() {
+      return sorting;
+    },
+    get columnVisibility() {
+      return columnVisibility;
+    },
+  },
+  onSortingChange: handleSortingChange,
+  onColumnVisibilityChange: handleVisibilityChange,
+  getCoreRowModel: getCoreRowModel(),
+  get getSortedRowModel() {
+    return manualSorting ? undefined : getSortedRowModel<CourseResponse>();
+  },
+  get manualSorting() {
+    return manualSorting;
+  },
+  enableSortingRemoval: true,
+});
+</script>
+
+{#snippet columnVisibilityGroup(
+    Group: typeof DropdownMenu.Group,
+    GroupHeading: typeof DropdownMenu.GroupHeading,
+    CheckboxItem: typeof DropdownMenu.CheckboxItem,
+    Separator: typeof DropdownMenu.Separator,
+    Item: typeof DropdownMenu.Item,
+)}
+    <Group>
+        <GroupHeading
+            class="px-2 py-1.5 text-xs font-medium text-muted-foreground"
+        >
+            Toggle columns
+        </GroupHeading>
+        {#each columns as col}
+            {@const id = col.id!}
+            {@const label = typeof col.header === "string" ? col.header : id}
+            <CheckboxItem
+                checked={columnVisibility[id] !== false}
+                closeOnSelect={false}
+                onCheckedChange={(checked) => {
+                    columnVisibility = {
+                        ...columnVisibility,
+                        [id]: checked,
+                    };
+                }}
+                class="relative flex items-center gap-2 rounded-sm px-2 py-1.5 text-sm cursor-pointer select-none outline-none data-highlighted:bg-accent data-highlighted:text-accent-foreground"
+            >
+                {#snippet children({ checked })}
+                    <span
+                        class="flex size-4 items-center justify-center rounded-sm border border-border"
+                    >
+                        {#if checked}
+                            <Check class="size-3" />
+                        {/if}
+                    </span>
+                    {label}
+                {/snippet}
+            </CheckboxItem>
+        {/each}
+    </Group>
+    {#if hasCustomVisibility}
+        <Separator class="mx-1 my-1 h-px bg-border" />
+        <Item
+            class="flex items-center gap-2 rounded-sm px-2 py-1.5 text-sm cursor-pointer select-none outline-none data-highlighted:bg-accent data-highlighted:text-accent-foreground"
+            onSelect={resetColumnVisibility}
+        >
+            <RotateCcw class="size-3.5" />
+            Reset to default
+        </Item>
+    {/if}
+{/snippet}
+
+<!-- Toolbar: View columns button -->
+<div class="flex items-center justify-end pb-2">
+    <DropdownMenu.Root>
+        <DropdownMenu.Trigger
+            class="inline-flex items-center gap-1.5 rounded-md border border-border bg-background px-2.5 py-1.5 text-xs font-medium text-muted-foreground hover:bg-accent hover:text-accent-foreground transition-colors cursor-pointer"
+        >
+            <Columns3 class="size-3.5" />
+            View
+        </DropdownMenu.Trigger>
+        <DropdownMenu.Portal>
+            <DropdownMenu.Content
+                class="z-50 min-w-40 rounded-md border border-border bg-card p-1 text-card-foreground shadow-lg"
+                align="end"
+                sideOffset={4}
+                forceMount
+            >
+                {#snippet child({ wrapperProps, props, open })}
+                    {#if open}
+                        <div {...wrapperProps}>
+                            <div
+                                {...props}
+                                transition:fly={{ duration: 150, y: -10 }}
+                            >
+                                {@render columnVisibilityGroup(
+                                    DropdownMenu.Group,
+                                    DropdownMenu.GroupHeading,
+                                    DropdownMenu.CheckboxItem,
+                                    DropdownMenu.Separator,
+                                    DropdownMenu.Item,
+                                )}
+                            </div>
+                        </div>
+                    {/if}
+                {/snippet}
+            </DropdownMenu.Content>
+        </DropdownMenu.Portal>
+    </DropdownMenu.Root>
+</div>
+
+<!-- Table with context menu on header -->
+<div bind:this={tableWrapper} class="overflow-x-auto">
+    <ContextMenu.Root>
+        <ContextMenu.Trigger class="contents">
+            <table class="w-full min-w-160 border-collapse text-sm">
+                <thead>
+                    {#each table.getHeaderGroups() as headerGroup}
+                        <tr
+                            class="border-b border-border text-left text-muted-foreground"
+                        >
+                            {#each headerGroup.headers as header}
+                                {#if header.column.getIsVisible()}
+                                    <th
+                                        class="py-2 px-2 font-medium {header.id ===
+                                        'seats'
+                                            ? 'text-right'
+                                            : ''}"
+                                        class:cursor-pointer={header.column.getCanSort()}
+                                        class:select-none={header.column.getCanSort()}
+                                        onclick={header.column.getToggleSortingHandler()}
+                                    >
+                                        {#if header.column.getCanSort()}
+                                            <span
+                                                class="inline-flex items-center gap-1"
+                                            >
+                                                {#if typeof header.column.columnDef.header === "string"}
+                                                    {header.column.columnDef
+                                                        .header}
+                                                {:else}
+                                                    <FlexRender
+                                                        content={header.column
+                                                            .columnDef.header}
+                                                        context={header.getContext()}
+                                                    />
+                                                {/if}
+                                                {#if header.column.getIsSorted() === "asc"}
+                                                    <ArrowUp class="size-3.5" />
+                                                {:else if header.column.getIsSorted() === "desc"}
+                                                    <ArrowDown
+                                                        class="size-3.5"
+                                                    />
+                                                {:else}
+                                                    <ArrowUpDown
+                                                        class="size-3.5 text-muted-foreground/40"
+                                                    />
+                                                {/if}
+                                            </span>
+                                        {:else if typeof header.column.columnDef.header === "string"}
+                                            {header.column.columnDef.header}
+                                        {:else}
+                                            <FlexRender
+                                                content={header.column.columnDef
+                                                    .header}
+                                                context={header.getContext()}
+                                            />
+                                        {/if}
+                                    </th>
+                                {/if}
+                            {/each}
+                        </tr>
+                    {/each}
+                </thead>
+                {#if loading && courses.length === 0}
+                    <tbody>
+                        {#each Array(5) as _}
+                            <tr class="border-b border-border">
+                                {#each table.getVisibleLeafColumns() as col}
+                                    <td class="py-2.5 px-2">
+                                        <div
+                                            class="h-4 bg-muted rounded animate-pulse {col.id ===
+                                            'seats'
+                                                ? 'w-14 ml-auto'
+                                                : col.id === 'title'
+                                                  ? 'w-40'
+                                                  : col.id === 'crn'
+                                                    ? 'w-10'
+                                                    : 'w-20'}"
+                                        ></div>
+                                    </td>
+                                {/each}
+                            </tr>
+                        {/each}
+                    </tbody>
+                {:else if courses.length === 0}
+                    <tbody>
+                        <tr>
+                            <td
+                                colspan={visibleColumnIds.length}
+                                class="py-12 text-center text-muted-foreground"
+                            >
+                                No courses found. Try adjusting your filters.
+                            </td>
+                        </tr>
+                    </tbody>
+                {:else}
+                    <!-- No out: transition — Svelte outros break table layout (tbody loses positioning and overlaps) -->
+                    {#each table.getRowModel().rows as row, i (row.id)}
+                        {@const course = row.original}
+                        <tbody
+                            animate:flip={{ duration: 300 }}
+                            in:fade={{
+                                duration: 200,
+                                delay: Math.min(i * 20, 400),
+                            }}
+                        >
+                            <tr
+                                class="border-b border-border cursor-pointer hover:bg-muted/50 transition-colors whitespace-nowrap {expandedCrn ===
+                                course.crn
+                                    ? 'bg-muted/30'
+                                    : ''}"
+                                onclick={() => toggleRow(course.crn)}
+                            >
+                                {#each row.getVisibleCells() as cell (cell.id)}
+                                    {@const colId = cell.column.id}
+                                    {#if colId === "crn"}
+                                        <td class="py-2 px-2 relative">
+                                            <button
+                                                class="relative inline-flex items-center rounded-full px-2 py-0.5 border border-border/50 bg-muted/20 hover:bg-muted/40 hover:border-foreground/30 transition-colors duration-150 cursor-copy focus-visible:outline-2 focus-visible:outline-offset-1 focus-visible:outline-ring font-mono text-xs text-muted-foreground/70"
+                                                onclick={(e) =>
+                                                    clipboard.copy(
+                                                        course.crn,
+                                                        e,
+                                                    )}
+                                                onkeydown={(e) => {
+                                                    if (
+                                                        e.key === "Enter" ||
+                                                        e.key === " "
+                                                    ) {
+                                                        e.preventDefault();
+                                                        clipboard.copy(
+                                                            course.crn,
+                                                            e,
+                                                        );
+                                                    }
+                                                }}
+                                                aria-label="Copy CRN {course.crn} to clipboard"
+                                            >
+                                                {course.crn}
+                                                {#if clipboard.copiedValue === course.crn}
+                                                    <span
+                                                        class="absolute -top-8 left-1/2 -translate-x-1/2 whitespace-nowrap text-xs px-2 py-1 rounded-md bg-green-500/10 border border-green-500/20 text-green-700 dark:text-green-300 pointer-events-none z-10"
+                                                        in:fade={{
+                                                            duration: 100,
+                                                        }}
+                                                        out:fade={{
+                                                            duration: 200,
+                                                        }}
+                                                    >
+                                                        Copied!
+                                                    </span>
+                                                {/if}
+                                            </button>
+                                        </td>
+                                    {:else if colId === "course_code"}
+                                        {@const subjectDesc =
+                                            subjectMap[course.subject]}
+                                        <td class="py-2 px-2 whitespace-nowrap">
+                                            <SimpleTooltip
+                                                text={subjectDesc
+                                                    ? `${subjectDesc} ${course.courseNumber}`
+                                                    : `${course.subject} ${course.courseNumber}`}
+                                                delay={200}
+                                                side="bottom"
+                                                passthrough
+                                            >
+                                                <span class="font-semibold"
+                                                    >{course.subject}
+                                                    {course.courseNumber}</span
+                                                >{#if course.sequenceNumber}<span
+                                                        class="text-muted-foreground"
+                                                        >-{course.sequenceNumber}</span
+                                                    >{/if}
+                                            </SimpleTooltip>
+                                        </td>
+                                    {:else if colId === "title"}
+                                        <td
+                                            class="py-2 px-2 font-medium max-w-50 truncate"
+                                        >
+                                            <SimpleTooltip
+                                                text={course.title}
+                                                delay={200}
+                                                side="bottom"
+                                                passthrough
+                                            >
+                                                <span class="block truncate"
+                                                    >{course.title}</span
+                                                >
+                                            </SimpleTooltip>
+                                        </td>
+                                    {:else if colId === "instructor"}
+                                        {@const primary = getPrimaryInstructor(
+                                            course.instructors,
+                                        )}
+                                        {@const display =
+                                            primaryInstructorDisplay(course)}
+                                        {@const commaIdx =
+                                            display.indexOf(", ")}
+                                        {@const ratingData =
+                                            primaryRating(course)}
+                                        <td class="py-2 px-2 whitespace-nowrap">
+                                            {#if display === "Staff"}
+                                                <span
+                                                    class="text-xs text-muted-foreground/60 uppercase"
+                                                    >Staff</span
+                                                >
+                                            {:else}
+                                                <SimpleTooltip
+                                                    text={primary?.displayName ??
+                                                        "Staff"}
+                                                    delay={200}
+                                                    side="bottom"
+                                                    passthrough
+                                                >
+                                                    {#if commaIdx !== -1}
+                                                        <span
+                                                            >{display.slice(
+                                                                0,
+                                                                commaIdx,
+                                                            )},
+                                                            <span
+                                                                class="text-muted-foreground"
+                                                                >{display.slice(
+                                                                    commaIdx +
+                                                                        1,
+                                                                )}</span
+                                                            ></span
+                                                        >
+                                                    {:else}
+                                                        <span>{display}</span>
+                                                    {/if}
+                                                </SimpleTooltip>
+                                            {/if}
+                                            {#if ratingData}
+                                                {@const lowConfidence =
+                                                    ratingData.count <
+                                                    RMP_CONFIDENCE_THRESHOLD}
+                                                <Tooltip.Root
+                                                    delayDuration={150}
+                                                >
+                                                    <Tooltip.Trigger>
+                                                        <span
+                                                            class="ml-1 text-xs font-medium inline-flex items-center gap-0.5"
+                                                            style={ratingStyle(
+                                                                ratingData.rating,
+                                                                themeStore.isDark,
+                                                            )}
+                                                        >
+                                                            {ratingData.rating.toFixed(
+                                                                1,
+                                                            )}
+                                                            {#if lowConfidence}
+                                                                <Triangle
+                                                                    class="size-2 fill-current"
+                                                                />
+                                                            {:else}
+                                                                <Star
+                                                                    class="size-2.5 fill-current"
+                                                                />
+                                                            {/if}
+                                                        </span>
+                                                    </Tooltip.Trigger>
+                                                    <Tooltip.Content
+                                                        side="bottom"
+                                                        sideOffset={6}
+                                                        class={cn(
+                                                            tooltipContentClass,
+                                                            "px-2.5 py-1.5",
+                                                        )}
+                                                    >
+                                                        <span
+                                                            class="inline-flex items-center gap-1.5 text-xs"
+                                                        >
+                                                            {ratingData.rating.toFixed(
+                                                                1,
+                                                            )}/5 · {formatNumber(ratingData.count)}
+                                                            ratings
+                                                            {#if (ratingData.count ?? 0) < RMP_CONFIDENCE_THRESHOLD}
+                                                                (low)
+                                                            {/if}
+                                                            {#if ratingData.legacyId != null}
+                                                                ·
+                                                                <a
+                                                                    href={rmpUrl(
+                                                                        ratingData.legacyId,
+                                                                    )}
+                                                                    target="_blank"
+                                                                    rel="noopener"
+                                                                    class="inline-flex items-center gap-0.5 text-muted-foreground hover:text-foreground transition-colors"
+                                                                >
+                                                                    RMP
+                                                                    <ExternalLink
+                                                                        class="size-3"
+                                                                    />
+                                                                </a>
+                                                            {/if}
+                                                        </span>
+                                                    </Tooltip.Content>
+                                                </Tooltip.Root>
+                                            {/if}
+                                        </td>
+                                    {:else if colId === "time"}
+                                        <td class="py-2 px-2 whitespace-nowrap">
+                                            <SimpleTooltip
+                                                text={formatMeetingTimesTooltip(
+                                                    course.meetingTimes,
+                                                )}
+                                                passthrough
+                                            >
+                                                {#if timeIsTBA(course)}
+                                                    <span
+                                                        class="text-xs text-muted-foreground/60"
+                                                        >TBA</span
+                                                    >
+                                                {:else}
+                                                    {@const mt =
+                                                        course.meetingTimes[0]}
+                                                    <span>
+                                                        {#if !isMeetingTimeTBA(mt)}
+                                                            <span
+                                                                class="font-mono font-medium"
+                                                                >{formatMeetingDays(
+                                                                    mt,
+                                                                )}</span
+                                                            >
+                                                            {" "}
+                                                        {/if}
+                                                        {#if !isTimeTBA(mt)}
+                                                            <span
+                                                                class="text-muted-foreground"
+                                                                >{formatTimeRange(
+                                                                    mt.begin_time,
+                                                                    mt.end_time,
+                                                                )}</span
+                                                            >
+                                                        {:else}
+                                                            <span
+                                                                class="text-xs text-muted-foreground/60"
+                                                                >TBA</span
+                                                            >
+                                                        {/if}
+                                                        {#if course.meetingTimes.length > 1}
+                                                            <span
+                                                                class="ml-1 text-xs text-muted-foreground/70 font-medium"
+                                                                >+{course
+                                                                    .meetingTimes
+                                                                    .length -
+                                                                    1}</span
+                                                            >
+                                                        {/if}
+                                                    </span>
+                                                {/if}
+                                            </SimpleTooltip>
+                                        </td>
+                                    {:else if colId === "location"}
+                                        {@const concern =
+                                            getDeliveryConcern(course)}
+                                        {@const accentColor =
+                                            concernAccentColor(concern)}
+                                        {@const locTooltip =
+                                            formatLocationTooltip(course)}
+                                        {@const locDisplay =
+                                            formatLocationDisplay(course)}
+                                        <td class="py-2 px-2 whitespace-nowrap">
+                                            {#if locTooltip}
+                                                <SimpleTooltip
+                                                    text={locTooltip}
+                                                    delay={200}
+                                                    passthrough
+                                                >
+                                                    <span
+                                                        class="text-muted-foreground"
+                                                        class:pl-2={accentColor !==
+                                                            null}
+                                                        style:border-left={accentColor
+                                                            ? `2px solid ${accentColor}`
+                                                            : undefined}
+                                                    >
+                                                        {locDisplay ?? "—"}
+                                                    </span>
+                                                </SimpleTooltip>
+                                            {:else if locDisplay}
+                                                <span
+                                                    class="text-muted-foreground"
+                                                >
+                                                    {locDisplay}
+                                                </span>
+                                            {:else}
+                                                <span
+                                                    class="text-xs text-muted-foreground/50"
+                                                    >—</span
+                                                >
+                                            {/if}
+                                        </td>
+                                    {:else if colId === "seats"}
+                                        <td
+                                            class="py-2 px-2 text-right whitespace-nowrap"
+                                        >
+                                            <SimpleTooltip
+                                                text="{formatNumber(openSeats(
+                                                    course,
+                                                ))} of {formatNumber(course.maxEnrollment)} seats open, {formatNumber(course.enrollment)} enrolled{course.waitCount >
+                                                0
+                                                    ? `, ${formatNumber(course.waitCount)} waitlisted`
+                                                    : ''}"
+                                                delay={200}
+                                                side="left"
+                                                passthrough
+                                            >
+                                                <span
+                                                    class="inline-flex items-center gap-1.5"
+                                                >
+                                                    <span
+                                                        class="size-1.5 rounded-full {seatsDotColor(
+                                                            course,
+                                                        )} shrink-0"
+                                                    ></span>
+                                                    <span
+                                                        class="{seatsColor(
+                                                            course,
+                                                        )} font-medium tabular-nums"
+                                                        >{#if openSeats(course) === 0}Full{:else}{openSeats(
+                                                                course,
+                                                            )} open{/if}</span
+                                                    >
+                                                    <span
+                                                        class="text-muted-foreground/60 tabular-nums"
+                                                        >{formatNumber(course.enrollment)}/{formatNumber(course.maxEnrollment)}{#if course.waitCount > 0}
+                                                            · WL {formatNumber(course.waitCount)}/{formatNumber(course.waitCapacity)}{/if}</span
+                                                    >
+                                                </span>
+                                            </SimpleTooltip>
+                                        </td>
+                                    {/if}
+                                {/each}
+                            </tr>
+                            {#if expandedCrn === course.crn}
+                                <tr>
+                                    <td
+                                        colspan={visibleColumnIds.length}
+                                        class="p-0"
+                                    >
+                                        <div
+                                            transition:slide={{ duration: 200 }}
+                                        >
+                                            <CourseDetail {course} />
+                                        </div>
+                                    </td>
+                                </tr>
+                            {/if}
+                        </tbody>
+                    {/each}
+                {/if}
+            </table>
+        </ContextMenu.Trigger>
+        <ContextMenu.Portal>
+            <ContextMenu.Content
+                class="z-50 min-w-40 rounded-md border border-border bg-card p-1 text-card-foreground shadow-lg"
+                forceMount
+            >
+                {#snippet child({ wrapperProps, props, open })}
+                    {#if open}
+                        <div {...wrapperProps}>
+                            <div
+                                {...props}
+                                in:fade={{ duration: 100 }}
+                                out:fade={{ duration: 100 }}
+                            >
+                                {@render columnVisibilityGroup(
+                                    ContextMenu.Group,
+                                    ContextMenu.GroupHeading,
+                                    ContextMenu.CheckboxItem,
+                                    ContextMenu.Separator,
+                                    ContextMenu.Item,
+                                )}
+                            </div>
+                        </div>
+                    {/if}
+                {/snippet}
+            </ContextMenu.Content>
+        </ContextMenu.Portal>
+    </ContextMenu.Root>
+</div>
@@ -0,0 +1,56 @@
+<script lang="ts">
+import { page } from "$app/state";
+import { TriangleAlert, RotateCcw } from "@lucide/svelte";
+
+interface Props {
+  /** Heading shown in the error card */
+  title?: string;
+  /** The error value from svelte:boundary */
+  error: unknown;
+  /** Reset callback from svelte:boundary */
+  reset: () => void;
+}
+
+let { title = "Something went wrong", error, reset }: Props = $props();
+
+let errorName = $derived(error instanceof Error ? error.constructor.name : "Error");
+let errorMessage = $derived(error instanceof Error ? error.message : String(error));
+let errorStack = $derived(error instanceof Error ? error.stack : null);
+</script>
+
+<div class="flex items-center justify-center py-16 px-4">
+  <div class="w-full max-w-lg rounded-lg border border-status-red/25 bg-status-red/5 overflow-hidden text-sm">
+    <div class="px-4 py-2.5 border-b border-status-red/15 flex items-center justify-between gap-4">
+      <div class="flex items-center gap-2 text-status-red">
+        <TriangleAlert size={16} strokeWidth={2.25} />
+        <span class="font-semibold">{title}</span>
+      </div>
+      <span class="text-xs text-muted-foreground font-mono">{page.url.pathname}</span>
+    </div>
+
+    <div class="px-4 py-3 border-b border-status-red/15">
+      <span class="text-xs text-muted-foreground/70 font-mono">{errorName}</span>
+      <pre class="mt-1 text-xs text-foreground/80 overflow-auto whitespace-pre-wrap break-words">{errorMessage}</pre>
+    </div>
+
+    {#if errorStack}
+      <details class="border-b border-status-red/15">
+        <summary class="px-4 py-2 text-xs text-muted-foreground/70 cursor-pointer hover:text-muted-foreground select-none">
+          Stack trace
+        </summary>
+        <pre class="px-4 py-3 text-xs text-muted-foreground/60 overflow-auto whitespace-pre-wrap break-words max-h-48">{errorStack}</pre>
+      </details>
+    {/if}
+
+    <div class="px-4 py-2.5 flex items-center justify-end gap-3">
+      <span class="text-xs text-muted-foreground/60">Retries this section, not the full page</span>
+      <button
+        class="shrink-0 cursor-pointer inline-flex items-center gap-1.5 rounded-md bg-status-red px-3 py-1.5 text-sm font-medium text-white hover:brightness-110 transition-all"
+        onclick={reset}
+      >
+        <RotateCcw size={14} strokeWidth={2.25} />
+        Try again
+      </button>
+    </div>
+  </div>
+</div>
@@ -0,0 +1,36 @@
+<script lang="ts">
+import { cn } from "$lib/utils";
+
+let {
+  commitHash,
+  showStatusLink = true,
+  class: className,
+}: {
+  commitHash?: string | null;
+  showStatusLink?: boolean;
+  class?: string;
+} = $props();
+</script>
+
+<div class={cn("flex justify-center items-center gap-2 mt-auto pt-6 pb-4", className)}>
+  {#if __APP_VERSION__}
+    <span class="text-xs text-muted-foreground">v{__APP_VERSION__}</span>
+    <div class="w-px h-3 bg-muted-foreground opacity-30"></div>
+  {/if}
+  <a
+    href={commitHash
+      ? `https://github.com/Xevion/banner/commit/${commitHash}`
+      : "https://github.com/Xevion/banner"}
+    target="_blank"
+    rel="noopener noreferrer"
+    class="text-xs text-muted-foreground no-underline hover:underline"
+  >
+    GitHub
+  </a>
+  {#if showStatusLink}
+    <div class="w-px h-3 bg-muted-foreground opacity-30"></div>
+    <a href="/health" class="text-xs text-muted-foreground no-underline hover:underline">
+      Status
+    </a>
+  {/if}
+</div>
@@ -0,0 +1,58 @@
+<script lang="ts">
+import { page } from "$app/state";
+import { Search, User, Clock } from "@lucide/svelte";
+import { authStore } from "$lib/auth.svelte";
+import ThemeToggle from "./ThemeToggle.svelte";
+
+const staticTabs = [
+  { href: "/", label: "Search", icon: Search },
+  { href: "/timeline", label: "Timeline", icon: Clock },
+] as const;
+
+const APP_PREFIXES = ["/profile", "/settings", "/admin"];
+
+let profileTab = $derived({
+  href: authStore.isAuthenticated ? "/profile" : "/login",
+  label: authStore.isAuthenticated ? "Account" : "Login",
+  icon: User,
+});
+
+function isActive(tabHref: string): boolean {
+  if (tabHref === "/") return page.url.pathname === "/";
+  if (tabHref === "/profile") {
+    return APP_PREFIXES.some((p) => page.url.pathname.startsWith(p));
+  }
+  return page.url.pathname.startsWith(tabHref);
+}
+</script>
+
+<nav class="w-full flex justify-center pt-5 px-5">
+  <div class="w-full max-w-6xl flex items-center justify-between">
+    <!-- pointer-events-auto: root layout wraps nav in pointer-events-none overlay -->
+    <div class="flex items-center gap-1 rounded-lg bg-muted p-1 pointer-events-auto">
+      {#each staticTabs as tab}
+        <a
+          href={tab.href}
+          class="flex items-center gap-1.5 rounded-md px-3 py-1.5 text-sm font-medium transition-colors no-underline
+            {isActive(tab.href)
+              ? 'bg-background text-foreground shadow-sm'
+              : 'text-muted-foreground hover:text-foreground hover:bg-background/50'}"
+        >
+          <tab.icon size={15} strokeWidth={2} />
+          {tab.label}
+        </a>
+      {/each}
+      <a
+        href={profileTab.href}
+        class="flex items-center gap-1.5 rounded-md px-3 py-1.5 text-sm font-medium transition-colors no-underline
+          {isActive(profileTab.href)
+            ? 'bg-background text-foreground shadow-sm'
+            : 'text-muted-foreground hover:text-foreground hover:bg-background/50'}"
+      >
+        <User size={15} strokeWidth={2} />
+        {profileTab.label}
+      </a>
+      <ThemeToggle />
+    </div>
+  </div>
+</nav>
@@ -0,0 +1,76 @@
+<script lang="ts">
+import { navigationStore } from "$lib/stores/navigation.svelte";
+import type { Snippet } from "svelte";
+import { cubicOut } from "svelte/easing";
+import type { TransitionConfig } from "svelte/transition";
+
+type Axis = "horizontal" | "vertical";
+
+let {
+  key,
+  children,
+  axis = "horizontal",
+  inDelay = 0,
+  outDelay = 0,
+}: {
+  key: string;
+  children: Snippet;
+  axis?: Axis;
+  inDelay?: number;
+  outDelay?: number;
+} = $props();
+
+const DURATION = 400;
+const OFFSET = 40;
+
+function translate(axis: Axis, value: number): string {
+  return axis === "vertical" ? `translateY(${value}px)` : `translateX(${value}px)`;
+}
+
+function inTransition(_node: HTMLElement): TransitionConfig {
+  const dir = navigationStore.direction;
+  if (dir === "fade") {
+    return {
+      duration: DURATION,
+      delay: inDelay,
+      easing: cubicOut,
+      css: (t: number) => `opacity: ${t}`,
+    };
+  }
+  const offset = dir === "right" ? OFFSET : -OFFSET;
+  return {
+    duration: DURATION,
+    delay: inDelay,
+    easing: cubicOut,
+    css: (t: number) => `opacity: ${t}; transform: ${translate(axis, (1 - t) * offset)}`,
+  };
+}
+
+function outTransition(_node: HTMLElement): TransitionConfig {
+  const dir = navigationStore.direction;
+  const base = "position: absolute; top: 0; left: 0; width: 100%; height: 100%";
+  if (dir === "fade") {
+    return {
+      duration: DURATION,
+      delay: outDelay,
+      easing: cubicOut,
+      css: (t: number) => `${base}; opacity: ${t}`,
+    };
+  }
+  const offset = dir === "right" ? -OFFSET : OFFSET;
+  return {
+    duration: DURATION,
+    delay: outDelay,
+    easing: cubicOut,
+    css: (t: number) => `${base}; opacity: ${t}; transform: ${translate(axis, (1 - t) * offset)}`,
+  };
+}
+</script>
+
+<div class="relative flex flex-1 flex-col overflow-hidden">
+  {#key key}
+    <div in:inTransition out:outTransition class="flex flex-1 flex-col">
+      {@render children()}
+    </div>
+  {/key}
+</div>
@@ -0,0 +1,167 @@
+<script lang="ts">
+import { Select } from "bits-ui";
+import { ChevronUp, ChevronDown } from "@lucide/svelte";
+import type { Action } from "svelte/action";
+import { formatNumber } from "$lib/utils";
+
+const slideIn: Action<HTMLElement, number> = (node, direction) => {
+  if (direction !== 0) {
+    node.animate(
+      [
+        { transform: `translateX(${direction * 20}px)`, opacity: 0 },
+        { transform: "translateX(0)", opacity: 1 },
+      ],
+      { duration: 200, easing: "ease-out" }
+    );
+  }
+};
+
+let {
+  totalCount,
+  offset,
+  limit,
+  onPageChange,
+}: {
+  totalCount: number;
+  offset: number;
+  limit: number;
+  onPageChange: (newOffset: number) => void;
+} = $props();
+
+const currentPage = $derived(Math.floor(offset / limit) + 1);
+const totalPages = $derived(Math.ceil(totalCount / limit));
+const start = $derived(offset + 1);
+const end = $derived(Math.min(offset + limit, totalCount));
+
+// Track direction for slide animation
+let direction = $state(0);
+
+// 5 page slots: current-2, current-1, current, current+1, current+2
+const pageSlots = $derived([-2, -1, 0, 1, 2].map((delta) => currentPage + delta));
+
+function isSlotVisible(page: number): boolean {
+  return page >= 1 && page <= totalPages;
+}
+
+function goToPage(page: number) {
+  direction = page > currentPage ? 1 : -1;
+  onPageChange((page - 1) * limit);
+}
+
+// Build items array for the Select dropdown
+const pageItems = $derived(
+  Array.from({ length: totalPages }, (_, i) => ({
+    value: String(i + 1),
+    label: String(i + 1),
+  }))
+);
+
+const selectValue = $derived(String(currentPage));
+</script>
+
+{#if totalCount > 0 && totalPages > 1}
+  <div class="flex items-start text-xs -mt-3 pl-2">
+    <!-- Left zone: result count -->
+    <div class="flex-1">
+      <span class="text-muted-foreground">
+        Showing {formatNumber(start)}&ndash;{formatNumber(end)} of {formatNumber(totalCount)} courses
+      </span>
+    </div>
+
+    <!-- Center zone: page buttons -->
+    <div class="flex items-center gap-1">
+      {#key currentPage}
+        {#each pageSlots as page, i (i)}
+          {#if i === 2}
+            <!-- Center slot: current page with dropdown trigger -->
+            <Select.Root
+              type="single"
+              value={selectValue}
+              onValueChange={(v) => {
+                if (v) goToPage(Number(v));
+              }}
+              items={pageItems}
+            >
+              <Select.Trigger
+                class="inline-flex items-center justify-center gap-1 w-auto min-w-9 h-9 px-2.5
+                       rounded-md text-sm font-medium tabular-nums
+                       border border-border bg-card text-foreground
+                       hover:bg-muted/50 active:bg-muted transition-colors
+                       cursor-pointer select-none outline-none
+                       focus-visible:ring-2 focus-visible:ring-ring focus-visible:ring-offset-2 focus-visible:ring-offset-background"
+                aria-label="Page {currentPage} of {totalPages}, click to select page"
+              >
+                <span use:slideIn={direction}>{currentPage}</span>
+                <ChevronUp class="size-3 text-muted-foreground" />
+              </Select.Trigger>
+              <Select.Portal>
+                <Select.Content
+                  class="border border-border bg-card shadow-md outline-hidden z-50
+                         max-h-72 min-w-16 w-auto
+                         select-none rounded-md p-1
+                         data-[state=open]:animate-in data-[state=closed]:animate-out
+                         data-[state=closed]:fade-out-0 data-[state=open]:fade-in-0
+                         data-[state=closed]:zoom-out-95 data-[state=open]:zoom-in-95
+                         data-[side=top]:slide-in-from-bottom-2
+                         data-[side=bottom]:slide-in-from-top-2"
+                  side="top"
+                  sideOffset={6}
+                >
+                  <Select.ScrollUpButton class="flex w-full items-center justify-center py-0.5">
+                    <ChevronUp class="size-3.5 text-muted-foreground" />
+                  </Select.ScrollUpButton>
+                  <Select.Viewport class="p-0.5">
+                    {#each pageItems as item (item.value)}
+                      <Select.Item
+                        class="rounded-sm outline-hidden flex h-8 w-full select-none items-center
+                               justify-center px-3 text-sm tabular-nums
+                               data-[highlighted]:bg-accent data-[highlighted]:text-accent-foreground
+                               data-[selected]:font-semibold"
+                        value={item.value}
+                        label={item.label}
+                      >
+                        {item.label}
+                      </Select.Item>
+                    {/each}
+                  </Select.Viewport>
+                  <Select.ScrollDownButton class="flex w-full items-center justify-center py-0.5">
+                    <ChevronDown class="size-3.5 text-muted-foreground" />
+                  </Select.ScrollDownButton>
+                </Select.Content>
+              </Select.Portal>
+            </Select.Root>
+          {:else}
+            <!-- Side slot: navigable page button or invisible placeholder -->
+            <button
+              class="inline-flex items-center justify-center w-9 h-9
+                     rounded-md text-sm tabular-nums
+                     text-muted-foreground
+                     hover:bg-muted/50 hover:text-foreground active:bg-muted transition-colors
+                     cursor-pointer select-none
+                     focus-visible:ring-2 focus-visible:ring-ring focus-visible:ring-offset-2 focus-visible:ring-offset-background
+                     {isSlotVisible(page) ? '' : 'invisible pointer-events-none'}"
+              onclick={() => goToPage(page)}
+              aria-label="Go to page {page}"
+              aria-hidden={!isSlotVisible(page)}
+              tabindex={isSlotVisible(page) ? 0 : -1}
+              disabled={!isSlotVisible(page)}
+            use:slideIn={direction}
+          >
+              {page}
+            </button>
+          {/if}
+        {/each}
+      {/key}
+    </div>
+
+    <!-- Right zone: spacer for centering -->
+    <div class="flex-1"></div>
+  </div>
+{:else if totalCount > 0}
+  <!-- Single page: just show the count, no pagination controls -->
+  <div class="flex items-start text-xs -mt-3 pl-2">
+    <span class="text-muted-foreground">
+      Showing {formatNumber(start)}&ndash;{formatNumber(end)} of {formatNumber(totalCount)} courses
+    </span>
+  </div>
+{/if}
@@ -0,0 +1,45 @@
+<script lang="ts">
+import type { Term, Subject } from "$lib/api";
+import SimpleTooltip from "./SimpleTooltip.svelte";
+import TermCombobox from "./TermCombobox.svelte";
+import SubjectCombobox from "./SubjectCombobox.svelte";
+
+let {
+  terms,
+  subjects,
+  selectedTerm = $bindable(),
+  selectedSubjects = $bindable(),
+  query = $bindable(),
+  openOnly = $bindable(),
+}: {
+  terms: Term[];
+  subjects: Subject[];
+  selectedTerm: string;
+  selectedSubjects: string[];
+  query: string;
+  openOnly: boolean;
+} = $props();
+</script>
+
+<div class="flex flex-wrap gap-3 items-start">
+  <TermCombobox {terms} bind:value={selectedTerm} />
+
+  <SubjectCombobox {subjects} bind:value={selectedSubjects} />
+
+  <input
+    type="text"
+    placeholder="Search courses..."
+    aria-label="Search courses"
+    bind:value={query}
+    class="h-9 border border-border bg-card text-foreground rounded-md px-3 text-sm flex-1 min-w-[200px]
+           focus-visible:outline-none focus-visible:ring-2 focus-visible:ring-ring focus-visible:ring-offset-2 focus-visible:ring-offset-background
+           transition-colors"
+  />
+
+  <SimpleTooltip text="Show only courses with available seats" delay={200} passthrough>
+    <label class="flex items-center gap-1.5 h-9 text-sm text-muted-foreground cursor-pointer">
+      <input type="checkbox" bind:checked={openOnly} />
+      Open only
+    </label>
+  </SimpleTooltip>
+</div>
@@ -0,0 +1,68 @@
+<script lang="ts">
+import { onMount } from "svelte";
+import SimpleTooltip from "$lib/components/SimpleTooltip.svelte";
+import { relativeTime } from "$lib/time";
+import { formatNumber } from "$lib/utils";
+
+export interface SearchMeta {
+  totalCount: number;
+  durationMs: number;
+  timestamp: Date;
+}
+
+let { meta }: { meta: SearchMeta | null } = $props();
+
+let now = $state(new Date());
+
+let formattedTime = $derived(
+  meta
+    ? meta.timestamp.toLocaleTimeString(undefined, {
+        hour: "2-digit",
+        minute: "2-digit",
+        second: "2-digit",
+      })
+    : ""
+);
+
+let relativeTimeResult = $derived(meta ? relativeTime(meta.timestamp, now) : null);
+let relativeTimeText = $derived(relativeTimeResult?.text ?? "");
+
+let countLabel = $derived(meta ? formatNumber(meta.totalCount) : "");
+let resultNoun = $derived(meta ? (meta.totalCount !== 1 ? "results" : "result") : "");
+let durationLabel = $derived(meta ? `${Math.round(meta.durationMs)}ms` : "");
+
+let tooltipText = $derived(meta ? `${relativeTimeText} · ${formattedTime}` : "");
+
+onMount(() => {
+  let nowTimeoutId: ReturnType<typeof setTimeout> | null = null;
+
+  function scheduleNowTick() {
+    const delay = relativeTimeResult?.nextUpdateMs ?? 1000;
+    nowTimeoutId = setTimeout(() => {
+      now = new Date();
+      scheduleNowTick();
+    }, delay);
+  }
+  scheduleNowTick();
+
+  return () => {
+    if (nowTimeoutId) clearTimeout(nowTimeoutId);
+  };
+});
+</script>
+
+<SimpleTooltip
+    text={tooltipText}
+    contentClass="whitespace-nowrap text-[12px] px-2 py-1"
+    triggerClass="self-start"
+    sideOffset={0}
+>
+    <span
+        class="pl-1 text-xs transition-opacity duration-200"
+        style:opacity={meta ? 1 : 0}
+    >
+        <span class="text-muted-foreground/70">{countLabel}</span>
+        <span class="text-muted-foreground/35">{resultNoun} in</span>
+        <span class="text-muted-foreground/70">{durationLabel}</span>
+    </span>
+</SimpleTooltip>
@@ -0,0 +1,42 @@
+<script lang="ts">
+import { Tooltip } from "bits-ui";
+import type { Snippet } from "svelte";
+import { cn } from "$lib/utils";
+
+let {
+  text,
+  delay = 150,
+  side = "top",
+  passthrough = false,
+  triggerClass = "",
+  contentClass = "",
+  sideOffset = 6,
+  children,
+}: {
+  text: string;
+  delay?: number;
+  side?: "top" | "bottom" | "left" | "right";
+  passthrough?: boolean;
+  triggerClass?: string;
+  contentClass?: string;
+  sideOffset?: number;
+  children: Snippet;
+} = $props();
+</script>
+
+<Tooltip.Root delayDuration={delay} disableHoverableContent={passthrough}>
+  <Tooltip.Trigger>
+    {#snippet child({ props })}
+      <span class={triggerClass} {...props}>
+        {@render children()}
+      </span>
+    {/snippet}
+  </Tooltip.Trigger>
+  <Tooltip.Content
+    {side}
+    {sideOffset}
+    class={cn("z-50 bg-card text-card-foreground text-xs border border-border rounded-md px-2.5 py-1.5 shadow-sm whitespace-pre-line max-w-max text-left", contentClass)}
+  >
+    {text}
+  </Tooltip.Content>
+</Tooltip.Root>
@@ -0,0 +1,162 @@
+<script lang="ts">
+import { Combobox } from "bits-ui";
+import { Check, ChevronsUpDown } from "@lucide/svelte";
+import { fly } from "svelte/transition";
+import type { Subject } from "$lib/api";
+import { formatNumber } from "$lib/utils";
+
+let {
+  subjects,
+  value = $bindable(),
+}: {
+  subjects: Subject[];
+  value: string[];
+} = $props();
+
+let open = $state(false);
+let searchValue = $state("");
+let containerEl = $state<HTMLDivElement>(null!);
+
+const filteredSubjects = $derived.by(() => {
+  const query = searchValue.toLowerCase().trim();
+  if (query === "") return subjects;
+
+  const exactCode: Subject[] = [];
+  const codeStartsWith: Subject[] = [];
+  const descriptionMatch: Subject[] = [];
+
+  for (const s of subjects) {
+    const codeLower = s.code.toLowerCase();
+    const descLower = s.description.toLowerCase();
+
+    if (codeLower === query) {
+      exactCode.push(s);
+    } else if (codeLower.startsWith(query)) {
+      codeStartsWith.push(s);
+    } else if (descLower.includes(query) || codeLower.includes(query)) {
+      descriptionMatch.push(s);
+    }
+  }
+
+  return [...exactCode, ...codeStartsWith, ...descriptionMatch];
+});
+
+const MAX_VISIBLE_CHIPS = 3;
+const visibleChips = $derived(value.slice(0, MAX_VISIBLE_CHIPS));
+const overflowCount = $derived(Math.max(0, value.length - MAX_VISIBLE_CHIPS));
+
+function removeSubject(code: string) {
+  value = value.filter((v) => v !== code);
+}
+
+// bits-ui sets the input text to the last selected item's label — clear it
+$effect(() => {
+  value;
+  const input = containerEl?.querySelector("input");
+  if (input) {
+    input.value = "";
+    searchValue = "";
+  }
+});
+</script>
+
+<Combobox.Root
+  type="multiple"
+  bind:value
+  bind:open
+  onOpenChange={(o: boolean) => {
+    if (!o) searchValue = "";
+  }}
+>
+  <!-- svelte-ignore a11y_click_events_have_key_events -->
+  <!-- svelte-ignore a11y_no_static_element_interactions -->
+  <div
+    class="relative h-9 rounded-md border border-border bg-card
+           flex flex-nowrap items-center gap-1 w-56 pr-9 overflow-hidden cursor-pointer
+           has-[:focus-visible]:ring-2 has-[:focus-visible]:ring-ring has-[:focus-visible]:ring-offset-2 has-[:focus-visible]:ring-offset-background"
+    bind:this={containerEl}
+    onclick={() => { containerEl?.querySelector('input')?.focus(); }}
+  >
+    {#if value.length > 0}
+      {#each (open ? value : visibleChips) as code (code)}
+        <span
+          role="button"
+          tabindex="-1"
+          onmousedown={(e) => { e.preventDefault(); e.stopPropagation(); }}
+          onclick={(e) => { e.stopPropagation(); removeSubject(code); }}
+          onkeydown={(e) => { if (e.key === "Enter" || e.key === " ") { e.stopPropagation(); removeSubject(code); } }}
+          class="inline-flex items-center rounded bg-muted px-1.5 py-0.5 text-xs font-mono shrink-0
+                 text-muted-foreground hover:outline hover:outline-1 hover:outline-ring
+                 cursor-pointer transition-[outline] duration-100 first:ml-2"
+        >
+          {code}
+        </span>
+      {/each}
+      {#if !open && overflowCount > 0}
+        <span class="text-xs text-muted-foreground shrink-0">+{formatNumber(overflowCount)}</span>
+      {/if}
+    {/if}
+    <Combobox.Input
+      
+      oninput={(e) => (searchValue = e.currentTarget.value)}
+      onfocus={() => { open = true; }}
+      class="h-full min-w-0 flex-1 bg-transparent text-muted-foreground text-sm
+             placeholder:text-muted-foreground outline-none border-none
+             {value.length > 0 ? 'pl-1' : 'pl-3'}"
+      placeholder={value.length > 0 ? "Filter..." : "All Subjects"}
+      aria-label="Search subjects"
+      autocomplete="off"
+      autocorrect="off"
+      spellcheck={false}
+    />
+    <span class="absolute end-2 top-1/2 -translate-y-1/2 text-muted-foreground pointer-events-none">
+      <ChevronsUpDown class="size-4" />
+    </span>
+  </div>
+  <Combobox.Portal>
+    <Combobox.Content
+      customAnchor={containerEl}
+      class="border border-border bg-card shadow-md
+             outline-hidden z-50
+             max-h-72 min-w-[var(--bits-combobox-anchor-width)] w-max max-w-96
+             select-none rounded-md p-1
+             data-[side=bottom]:translate-y-1 data-[side=top]:-translate-y-1"
+      sideOffset={4}
+      forceMount
+    >
+      {#snippet child({ wrapperProps, props, open: isOpen })}
+        {#if isOpen}
+          <div {...wrapperProps}>
+            <div {...props} transition:fly={{ duration: 150, y: -4 }}>
+              <Combobox.Viewport class="p-0.5">
+                {#each filteredSubjects as subject (subject.code)}
+                  <Combobox.Item
+                    class="rounded-sm outline-hidden flex h-8 w-full select-none items-center gap-2 px-2 text-sm whitespace-nowrap
+                           data-[highlighted]:bg-accent data-[highlighted]:text-accent-foreground"
+                    value={subject.code}
+                    label={subject.description}
+                  >
+                    {#snippet children({ selected })}
+                      <span class="inline-flex items-center justify-center rounded bg-muted px-1 py-0.5
+                                   text-xs font-mono text-muted-foreground w-10 shrink-0 text-center">
+                        {subject.code}
+                      </span>
+                      <span class="flex-1">{subject.description}</span>
+                      {#if selected}
+                        <Check class="ml-auto size-4 shrink-0" />
+                      {/if}
+                    {/snippet}
+                  </Combobox.Item>
+                {:else}
+                  <span class="block px-2 py-2 text-sm text-muted-foreground">
+                    No subjects found.
+                  </span>
+                {/each}
+              </Combobox.Viewport>
+            </div>
+          </div>
+        {/if}
+      {/snippet}
+    </Combobox.Content>
+  </Combobox.Portal>
+</Combobox.Root>
@@ -0,0 +1,139 @@
+<script lang="ts">
+import { Combobox } from "bits-ui";
+import { Check, ChevronsUpDown } from "@lucide/svelte";
+import { fly } from "svelte/transition";
+import type { Term } from "$lib/api";
+
+let {
+  terms,
+  value = $bindable(),
+}: {
+  terms: Term[];
+  value: string;
+} = $props();
+
+let open = $state(false);
+let searchValue = $state("");
+let containerEl = $state<HTMLDivElement>(null!);
+
+const currentTermCode = $derived(
+  terms.find((t) => !t.description.includes("(View Only)"))?.code ?? ""
+);
+
+const selectedLabel = $derived(
+  terms.find((t) => t.code === value)?.description ?? "Select term..."
+);
+
+const filteredTerms = $derived.by(() => {
+  const query = searchValue.toLowerCase();
+  const matched =
+    query === "" ? terms : terms.filter((t) => t.description.toLowerCase().includes(query));
+
+  const current = matched.find((t) => t.code === currentTermCode);
+  const rest = matched.filter((t) => t.code !== currentTermCode);
+  return current ? [current, ...rest] : rest;
+});
+
+// Manage DOM input text: clear when open for searching, restore label when closed
+$effect(() => {
+  const _open = open;
+  void value; // track selection changes
+  const _label = selectedLabel;
+  const input = containerEl?.querySelector("input");
+  if (!input) return;
+  if (_open) {
+    input.value = "";
+    searchValue = "";
+  } else {
+    input.value = _label;
+  }
+});
+</script>
+
+<Combobox.Root
+  type="single"
+  bind:value={() => value, (v) => { if (v) value = v; }}
+  bind:open
+  onOpenChange={(o: boolean) => {
+    if (!o) searchValue = "";
+  }}
+>
+  <!-- svelte-ignore a11y_no_static_element_interactions -->
+  <div
+    class="relative h-9 rounded-md border border-border bg-card
+           flex items-center w-40 cursor-pointer
+           has-[:focus-visible]:ring-2 has-[:focus-visible]:ring-ring has-[:focus-visible]:ring-offset-2 has-[:focus-visible]:ring-offset-background"
+    role="presentation"
+    bind:this={containerEl}
+    onclick={() => { containerEl?.querySelector('input')?.focus(); }}
+    onkeydown={() => { containerEl?.querySelector('input')?.focus(); }}
+  >
+    <Combobox.Input
+      oninput={(e) => (searchValue = e.currentTarget.value)}
+      onfocus={() => { open = true; }}
+      class="h-full w-full bg-transparent text-muted-foreground text-sm
+             placeholder:text-muted-foreground outline-none border-none
+             pl-3 pr-9 truncate"
+      placeholder="Select term..."
+      aria-label="Select term"
+      autocomplete="off"
+      autocorrect="off"
+      spellcheck={false}
+    />
+    <span class="absolute end-2 top-1/2 -translate-y-1/2 text-muted-foreground pointer-events-none">
+      <ChevronsUpDown class="size-4" />
+    </span>
+  </div>
+  <Combobox.Portal>
+    <Combobox.Content
+      customAnchor={containerEl}
+      class="border border-border bg-card shadow-md
+             outline-hidden z-50
+             max-h-72 min-w-[var(--bits-combobox-anchor-width)]
+             select-none rounded-md p-1
+             data-[side=bottom]:translate-y-1 data-[side=top]:-translate-y-1"
+      sideOffset={4}
+      forceMount
+    >
+      {#snippet child({ wrapperProps, props, open: isOpen })}
+        {#if isOpen}
+          <div {...wrapperProps}>
+            <div {...props} transition:fly={{ duration: 150, y: -4 }}>
+              <Combobox.Viewport class="p-0.5">
+                {#each filteredTerms as term, i (term.code)}
+                  {#if i === 1 && term.code !== currentTermCode && filteredTerms[0]?.code === currentTermCode}
+                    <div class="mx-2 my-1 h-px bg-border"></div>
+                  {/if}
+                  <Combobox.Item
+                    class="rounded-sm outline-hidden flex h-8 w-full select-none items-center px-2 text-sm
+                           data-[highlighted]:bg-accent data-[highlighted]:text-accent-foreground
+                           {term.code === value ? 'cursor-default' : 'cursor-pointer'}
+                           {term.code === currentTermCode ? 'font-medium text-foreground' : 'text-foreground'}"
+                    value={term.code}
+                    label={term.description}
+                  >
+                    {#snippet children({ selected })}
+                      <span class="flex-1 truncate">
+                        {term.description}
+                        {#if term.code === currentTermCode}
+                          <span class="ml-1.5 text-xs text-muted-foreground font-normal">current</span>
+                        {/if}
+                      </span>
+                      {#if selected}
+                        <Check class="ml-2 size-4 shrink-0" />
+                      {/if}
+                    {/snippet}
+                  </Combobox.Item>
+                {:else}
+                  <span class="block px-2 py-2 text-sm text-muted-foreground">
+                    No terms found.
+                  </span>
+                {/each}
+              </Combobox.Viewport>
+            </div>
+          </div>
+        {/if}
+      {/snippet}
+    </Combobox.Content>
+  </Combobox.Portal>
+</Combobox.Root>
@@ -0,0 +1,69 @@
+<script lang="ts">
+import { tick } from "svelte";
+import { Moon, Sun } from "@lucide/svelte";
+import { themeStore } from "$lib/stores/theme.svelte";
+import SimpleTooltip from "./SimpleTooltip.svelte";
+
+/**
+ * Theme toggle with View Transitions API circular reveal animation.
+ * The clip-path circle expands from the click point to cover the viewport.
+ */
+async function handleToggle(event: MouseEvent) {
+  const supportsViewTransition =
+    typeof document !== "undefined" &&
+    "startViewTransition" in document &&
+    !window.matchMedia("(prefers-reduced-motion: reduce)").matches;
+
+  if (!supportsViewTransition) {
+    themeStore.toggle();
+    return;
+  }
+
+  const x = event.clientX;
+  const y = event.clientY;
+  const endRadius = Math.hypot(Math.max(x, innerWidth - x), Math.max(y, innerHeight - y));
+
+  const transition = document.startViewTransition(async () => {
+    themeStore.toggle();
+    await tick();
+  });
+
+  transition.ready.then(() => {
+    document.documentElement.animate(
+      {
+        clipPath: [`circle(0px at ${x}px ${y}px)`, `circle(${endRadius}px at ${x}px ${y}px)`],
+      },
+      {
+        duration: 500,
+        easing: "cubic-bezier(0.4, 0, 0.2, 1)",
+        pseudoElement: "::view-transition-new(root)",
+      }
+    );
+  });
+}
+</script>
+
+<SimpleTooltip text={themeStore.isDark ? "Switch to light mode" : "Switch to dark mode"} delay={200} side="bottom" passthrough>
+  <button
+    type="button"
+    onclick={(e) => handleToggle(e)}
+    aria-label={themeStore.isDark ? "Switch to light mode" : "Switch to dark mode"}
+    class="cursor-pointer border-none rounded-md flex items-center justify-center p-1.5
+      text-muted-foreground hover:text-foreground hover:bg-background/50 bg-transparent transition-colors"
+  >
+    <div class="relative size-[18px]">
+      <Sun
+        size={18}
+        class="absolute inset-0 transition-all duration-300 {themeStore.isDark
+          ? 'rotate-90 scale-0 opacity-0'
+          : 'rotate-0 scale-100 opacity-100'}"
+      />
+      <Moon
+        size={18}
+        class="absolute inset-0 transition-all duration-300 {themeStore.isDark
+          ? 'rotate-0 scale-100 opacity-100'
+          : '-rotate-90 scale-0 opacity-0'}"
+      />
+    </div>
+  </button>
+</SimpleTooltip>
@@ -0,0 +1,676 @@
+<script lang="ts">
+import { onMount } from "svelte";
+import { scaleTime, scaleLinear } from "d3-scale";
+
+import type { TimeSlot, ChartContext } from "$lib/timeline/types";
+import {
+  PADDING,
+  DEFAULT_AXIS_RATIO,
+  CHART_HEIGHT_RATIO,
+  MIN_SPAN_MS,
+  MAX_SPAN_MS,
+  DEFAULT_SPAN_MS,
+  ZOOM_FACTOR,
+  ZOOM_KEY_FACTOR,
+  ZOOM_EASE,
+  ZOOM_SETTLE_THRESHOLD,
+  PAN_FRICTION,
+  PAN_STOP_THRESHOLD,
+  PAN_STOP_THRESHOLD_Y,
+  VELOCITY_SAMPLE_WINDOW,
+  VELOCITY_MIN_DT,
+  PAN_STEP_RATIO,
+  PAN_STEP_CTRL_RATIO,
+  PAN_EASE,
+  PAN_SETTLE_THRESHOLD_PX,
+  YRATIO_STEP,
+  YRATIO_MIN,
+  YRATIO_MAX,
+  YRATIO_SETTLE_THRESHOLD,
+  FOLLOW_EASE,
+  MIN_MAXY,
+  MAX_DT,
+  DEFAULT_DT,
+  TAP_MAX_DURATION_MS,
+  TAP_MAX_DISTANCE_PX,
+} from "$lib/timeline/constants";
+import { createTimelineStore } from "$lib/timeline/store.svelte";
+import {
+  createAnimMap,
+  syncAnimTargets,
+  stepAnimations,
+  pruneAnimMap,
+} from "$lib/timeline/animation";
+import {
+  getVisibleSlots,
+  findSlotByTime,
+  snapToSlot,
+  enabledTotalClasses,
+} from "$lib/timeline/viewport";
+import {
+  drawGrid,
+  drawHoverColumn,
+  drawStackedArea,
+  drawNowLine,
+  drawTimeAxis,
+  stackVisibleSlots,
+} from "$lib/timeline/renderer";
+import TimelineDrawer from "./TimelineDrawer.svelte";
+import TimelineTooltip from "./TimelineTooltip.svelte";
+
+// ── Reactive DOM state ──────────────────────────────────────────────
+let canvasEl: HTMLCanvasElement | undefined = $state();
+let containerEl: HTMLDivElement | undefined = $state();
+let width = $state(800);
+let height = $state(400);
+let dpr = $state(1);
+
+// ── View window ─────────────────────────────────────────────────────
+let viewCenter = $state(Date.now());
+let viewSpan = $state(DEFAULT_SPAN_MS);
+let viewYRatio = $state(DEFAULT_AXIS_RATIO);
+
+// ── Interaction state ───────────────────────────────────────────────
+let isDragging = $state(false);
+let dragStartX = $state(0);
+let dragStartY = $state(0);
+let dragStartCenter = $state(0);
+let dragStartYRatio = $state(0);
+let followEnabled = $state(true);
+let ctrlHeld = $state(false);
+
+// ── Animation state (intentionally non-reactive — updated in rAF) ──
+let panVelocity = 0;
+let panVelocityY = 0;
+let pointerSamples: { time: number; x: number; y: number }[] = [];
+
+// ── Multi-touch / pinch state ────────────────────────────────────────
+let activePointers = new Map<number, { x: number; y: number }>();
+let isPinching = false;
+let pinchStartDist = 0;
+let pinchStartSpan = 0;
+let pinchAnchorTime = 0;
+let pinchAnchorRatio = 0.5;
+
+// ── Tap detection ────────────────────────────────────────────────────
+let pointerDownTime = 0;
+let pointerDownPos = { x: 0, y: 0 };
+
+let targetSpan = DEFAULT_SPAN_MS;
+let zoomAnchorTime = 0;
+let zoomAnchorRatio = 0.5;
+let isZoomAnimating = false;
+
+let targetCenter = Date.now();
+let isPanAnimating = false;
+let targetYRatio = DEFAULT_AXIS_RATIO;
+let isYPanAnimating = false;
+
+let animationFrameId = 0;
+let lastFrameTime = 0;
+let animatedMaxY = MIN_MAXY;
+
+const animMap = createAnimMap();
+
+// ── Tooltip + hover ─────────────────────────────────────────────────
+let tooltipVisible = $state(false);
+let tooltipX = $state(0);
+let tooltipY = $state(0);
+let tooltipSlot: TimeSlot | null = $state(null);
+let hoverSlotTime: number | null = $state(null);
+let lastPointerClientX = 0;
+let lastPointerClientY = 0;
+let pointerOverCanvas = false;
+
+// ── Drawer ──────────────────────────────────────────────────────────
+let drawerOpen = $state(false);
+// Start with an empty set — subjects are populated dynamically from the API.
+let enabledSubjects: Set<string> = $state(new Set());
+
+// ── Data store ──────────────────────────────────────────────────────
+const store = createTimelineStore();
+let data: TimeSlot[] = $derived(store.data);
+let allSubjects: string[] = $derived(store.subjects);
+
+// Auto-enable newly discovered subjects.
+$effect(() => {
+  const storeSubjects = store.subjects;
+  const next = new Set(enabledSubjects);
+  let changed = false;
+  for (const s of storeSubjects) {
+    if (!next.has(s)) {
+      next.add(s);
+      changed = true;
+    }
+  }
+  if (changed) {
+    enabledSubjects = next;
+  }
+});
+
+let activeSubjects = $derived(allSubjects.filter((s) => enabledSubjects.has(s)));
+
+// ── Derived layout ──────────────────────────────────────────────────
+let viewStart = $derived(viewCenter - viewSpan / 2);
+let viewEnd = $derived(viewCenter + viewSpan / 2);
+let chartHeight = $derived(height * CHART_HEIGHT_RATIO);
+let chartBottom = $derived(height * viewYRatio);
+let chartTop = $derived(chartBottom - chartHeight);
+
+let xScale = $derived(
+  scaleTime()
+    .domain([new Date(viewStart), new Date(viewEnd)])
+    .range([PADDING.left, width - PADDING.right])
+);
+
+// Reused across frames — domain/range updated imperatively in render().
+let yScale = scaleLinear()
+  .domain([0, MIN_MAXY * 1.1])
+  .range([0, 1]);
+
+// ── Subject toggling ────────────────────────────────────────────────
+function toggleSubject(subject: string) {
+  const next = new Set(enabledSubjects);
+  if (next.has(subject)) next.delete(subject);
+  else next.add(subject);
+  enabledSubjects = next;
+}
+
+function enableAll() {
+  enabledSubjects = new Set(allSubjects);
+}
+
+function disableAll() {
+  enabledSubjects = new Set();
+}
+
+// ── Rendering ───────────────────────────────────────────────────────
+function render() {
+  if (!canvasEl) return;
+  const ctx = canvasEl.getContext("2d");
+  if (!ctx) return;
+
+  // Update yScale in-place (no allocation per frame).
+  yScale.domain([0, animatedMaxY * 1.1]).range([chartBottom, chartTop]);
+
+  ctx.save();
+  ctx.scale(dpr, dpr);
+  ctx.clearRect(0, 0, width, height);
+
+  const chart: ChartContext = {
+    ctx,
+    xScale,
+    yScale,
+    width,
+    chartTop,
+    chartBottom,
+    viewSpan,
+    viewStart,
+    viewEnd,
+  };
+
+  const visible = getVisibleSlots(data, viewStart, viewEnd);
+  const visibleStack = stackVisibleSlots(visible, allSubjects, enabledSubjects, animMap);
+
+  drawGrid(chart);
+  drawHoverColumn(chart, visibleStack, hoverSlotTime);
+  drawStackedArea(chart, visibleStack);
+  drawNowLine(chart);
+  drawTimeAxis(chart);
+
+  ctx.restore();
+}
+
+// ── Hover logic ─────────────────────────────────────────────────────
+function updateHover() {
+  if (!pointerOverCanvas || isDragging || !canvasEl) return;
+
+  const rect = canvasEl.getBoundingClientRect();
+  const x = lastPointerClientX - rect.left;
+  const y = lastPointerClientY - rect.top;
+
+  if (y < chartTop || y > chartBottom) {
+    tooltipVisible = false;
+    hoverSlotTime = null;
+    return;
+  }
+
+  const time = xScale.invert(x);
+  const snappedTime = snapToSlot(time.getTime());
+  const slot = findSlotByTime(data, snappedTime);
+  if (!slot) {
+    tooltipVisible = false;
+    hoverSlotTime = null;
+    return;
+  }
+
+  const total = enabledTotalClasses(slot, activeSubjects);
+  if (total <= 0) {
+    tooltipVisible = false;
+    hoverSlotTime = null;
+    return;
+  }
+
+  // Bypass area overlap check when CTRL is held.
+  if (!ctrlHeld) {
+    const stackTopY = yScale(total);
+    if (y < stackTopY) {
+      tooltipVisible = false;
+      hoverSlotTime = null;
+      return;
+    }
+  }
+
+  tooltipSlot = slot;
+  tooltipX = lastPointerClientX;
+  tooltipY = lastPointerClientY;
+  tooltipVisible = true;
+  hoverSlotTime = snappedTime;
+}
+
+// ── Interaction helpers ───────────────────────────────────────────────
+function pinchDistance(): number {
+  const pts = [...activePointers.values()];
+  if (pts.length < 2) return 0;
+  const dx = pts[1].x - pts[0].x;
+  const dy = pts[1].y - pts[0].y;
+  return Math.hypot(dx, dy);
+}
+
+function pinchMidpoint(): { x: number; y: number } {
+  const pts = [...activePointers.values()];
+  if (pts.length < 2) return { x: 0, y: 0 };
+  return { x: (pts[0].x + pts[1].x) / 2, y: (pts[0].y + pts[1].y) / 2 };
+}
+
+// ── Interaction handlers ────────────────────────────────────────────
+function onPointerDown(e: PointerEvent) {
+  if (e.button !== 0) return;
+  (e.currentTarget as HTMLElement).setPointerCapture(e.pointerId);
+  activePointers.set(e.pointerId, { x: e.clientX, y: e.clientY });
+
+  // Two fingers down → start pinch-to-zoom
+  if (activePointers.size === 2) {
+    isDragging = false;
+    isPinching = true;
+    pinchStartDist = pinchDistance();
+    pinchStartSpan = viewSpan;
+
+    const mid = pinchMidpoint();
+    const rect = canvasEl?.getBoundingClientRect();
+    const midX = rect ? mid.x - rect.left : mid.x;
+    const chartWidth = width - PADDING.left - PADDING.right;
+
+    pinchAnchorTime = xScale.invert(midX).getTime();
+    pinchAnchorRatio = (midX - PADDING.left) / chartWidth;
+    return;
+  }
+
+  // Single finger / mouse → start drag
+  isDragging = true;
+  dragStartX = e.clientX;
+  dragStartY = e.clientY;
+  dragStartCenter = viewCenter;
+  dragStartYRatio = viewYRatio;
+  followEnabled = false;
+  panVelocity = 0;
+  panVelocityY = 0;
+  isZoomAnimating = false;
+  isPanAnimating = false;
+  isYPanAnimating = false;
+  targetSpan = viewSpan;
+  tooltipVisible = false;
+  hoverSlotTime = null;
+  pointerDownTime = performance.now();
+  pointerDownPos = { x: e.clientX, y: e.clientY };
+  pointerSamples = [{ time: performance.now(), x: e.clientX, y: e.clientY }];
+}
+
+function onPointerMove(e: PointerEvent) {
+  ctrlHeld = e.ctrlKey || e.metaKey;
+  lastPointerClientX = e.clientX;
+  lastPointerClientY = e.clientY;
+  pointerOverCanvas = true;
+  activePointers.set(e.pointerId, { x: e.clientX, y: e.clientY });
+
+  // Pinch-to-zoom (two-finger gesture)
+  if (isPinching && activePointers.size >= 2) {
+    const dist = pinchDistance();
+    if (pinchStartDist > 0) {
+      const scale = pinchStartDist / dist; // fingers apart = zoom in
+      const newSpan = Math.min(MAX_SPAN_MS, Math.max(MIN_SPAN_MS, pinchStartSpan * scale));
+      viewSpan = newSpan;
+      targetSpan = newSpan;
+      viewCenter = pinchAnchorTime + (0.5 - pinchAnchorRatio) * viewSpan;
+    }
+    return;
+  }
+
+  if (isDragging) {
+    const dx = e.clientX - dragStartX;
+    const dy = e.clientY - dragStartY;
+    const msPerPx = viewSpan / (width - PADDING.left - PADDING.right);
+    viewCenter = dragStartCenter - dx * msPerPx;
+    viewYRatio = dragStartYRatio + dy / height;
+
+    const now = performance.now();
+    pointerSamples.push({ time: now, x: e.clientX, y: e.clientY });
+    const cutoff = now - VELOCITY_SAMPLE_WINDOW;
+    pointerSamples = pointerSamples.filter((s) => s.time >= cutoff);
+  } else {
+    updateHover();
+  }
+}
+
+function onPointerUp(e: PointerEvent) {
+  (e.currentTarget as HTMLElement).releasePointerCapture(e.pointerId);
+  activePointers.delete(e.pointerId);
+
+  // End pinch when fewer than 2 fingers remain
+  if (isPinching) {
+    if (activePointers.size < 2) {
+      isPinching = false;
+      // If one finger remains, reset drag origin to that finger's position
+      if (activePointers.size === 1) {
+        const remaining = [...activePointers.values()][0];
+        isDragging = true;
+        dragStartX = remaining.x;
+        dragStartY = remaining.y;
+        dragStartCenter = viewCenter;
+        dragStartYRatio = viewYRatio;
+        pointerSamples = [{ time: performance.now(), x: remaining.x, y: remaining.y }];
+      }
+    }
+    return;
+  }
+
+  isDragging = false;
+
+  // Tap detection: short duration + minimal movement → show tooltip
+  const elapsed = performance.now() - pointerDownTime;
+  const dist = Math.hypot(e.clientX - pointerDownPos.x, e.clientY - pointerDownPos.y);
+  if (elapsed < TAP_MAX_DURATION_MS && dist < TAP_MAX_DISTANCE_PX) {
+    lastPointerClientX = e.clientX;
+    lastPointerClientY = e.clientY;
+    pointerOverCanvas = true;
+    // Bypass the isDragging guard in updateHover since we just cleared it
+    updateHover();
+    pointerSamples = [];
+    return;
+  }
+
+  // Momentum from drag
+  if (pointerSamples.length >= 2) {
+    const first = pointerSamples[0];
+    const last = pointerSamples[pointerSamples.length - 1];
+    const dt = last.time - first.time;
+    if (dt > VELOCITY_MIN_DT) {
+      const pxPerMsX = (last.x - first.x) / dt;
+      const msPerPx = viewSpan / (width - PADDING.left - PADDING.right);
+      panVelocity = -pxPerMsX * msPerPx;
+      panVelocityY = (last.y - first.y) / dt;
+    }
+  }
+  pointerSamples = [];
+}
+
+function onPointerLeave() {
+  pointerOverCanvas = false;
+  tooltipVisible = false;
+  hoverSlotTime = null;
+}
+
+function onPointerCancel(e: PointerEvent) {
+  activePointers.delete(e.pointerId);
+  if (activePointers.size < 2) isPinching = false;
+  if (activePointers.size === 0) isDragging = false;
+}
+
+function onWheel(e: WheelEvent) {
+  e.preventDefault();
+  if (!canvasEl) return;
+  followEnabled = false;
+  panVelocity = 0;
+  panVelocityY = 0;
+
+  const rect = canvasEl.getBoundingClientRect();
+  const mouseX = e.clientX - rect.left;
+  const chartWidth = width - PADDING.left - PADDING.right;
+
+  zoomAnchorTime = xScale.invert(mouseX).getTime();
+  zoomAnchorRatio = (mouseX - PADDING.left) / chartWidth;
+
+  const zoomIn = e.deltaY < 0;
+  const factor = zoomIn ? 1 / ZOOM_FACTOR : ZOOM_FACTOR;
+  targetSpan = Math.min(MAX_SPAN_MS, Math.max(MIN_SPAN_MS, targetSpan * factor));
+  isZoomAnimating = true;
+}
+
+function onKeyDown(e: KeyboardEvent) {
+  const wasCtrl = ctrlHeld;
+  ctrlHeld = e.ctrlKey || e.metaKey;
+  if (ctrlHeld !== wasCtrl) updateHover();
+
+  switch (e.key) {
+    case "ArrowLeft":
+    case "ArrowRight": {
+      e.preventDefault();
+      followEnabled = false;
+      panVelocity = 0;
+      const ratio = e.ctrlKey ? PAN_STEP_CTRL_RATIO : PAN_STEP_RATIO;
+      const step = viewSpan * ratio;
+      if (!isPanAnimating) targetCenter = viewCenter;
+      targetCenter += e.key === "ArrowRight" ? step : -step;
+      isPanAnimating = true;
+      break;
+    }
+    case "ArrowUp":
+    case "ArrowDown": {
+      e.preventDefault();
+      if (e.ctrlKey || e.metaKey) {
+        const direction = e.key === "ArrowUp" ? -1 : 1;
+        if (!isYPanAnimating) targetYRatio = viewYRatio;
+        targetYRatio = Math.max(
+          YRATIO_MIN,
+          Math.min(YRATIO_MAX, targetYRatio + direction * YRATIO_STEP)
+        );
+        isYPanAnimating = true;
+      } else {
+        const factor = e.key === "ArrowUp" ? 1 / ZOOM_KEY_FACTOR : ZOOM_KEY_FACTOR;
+        followEnabled = false;
+        panVelocity = 0;
+        zoomAnchorTime = isPanAnimating ? targetCenter : viewCenter;
+        zoomAnchorRatio = 0.5;
+        targetSpan = Math.min(MAX_SPAN_MS, Math.max(MIN_SPAN_MS, targetSpan * factor));
+        isZoomAnimating = true;
+      }
+      break;
+    }
+  }
+}
+
+function onKeyUp(e: KeyboardEvent) {
+  const wasCtrl = ctrlHeld;
+  ctrlHeld = e.ctrlKey || e.metaKey;
+  if (ctrlHeld !== wasCtrl) updateHover();
+}
+
+function onWindowBlur() {
+  const wasCtrl = ctrlHeld;
+  ctrlHeld = false;
+  if (wasCtrl) updateHover();
+}
+
+function resumeFollow() {
+  panVelocity = 0;
+  panVelocityY = 0;
+  isPanAnimating = false;
+  isYPanAnimating = false;
+  targetYRatio = DEFAULT_AXIS_RATIO;
+  viewYRatio = DEFAULT_AXIS_RATIO;
+  targetSpan = DEFAULT_SPAN_MS;
+  isZoomAnimating = true;
+  zoomAnchorTime = Date.now();
+  zoomAnchorRatio = 0.5;
+  followEnabled = true;
+}
+
+// ── Resize ──────────────────────────────────────────────────────────
+function updateSize() {
+  if (!containerEl) return;
+  const rect = containerEl.getBoundingClientRect();
+  width = rect.width;
+  height = rect.height;
+  dpr = window.devicePixelRatio || 1;
+  if (canvasEl) {
+    canvasEl.width = width * dpr;
+    canvasEl.height = height * dpr;
+  }
+}
+
+// ── Animation loop ──────────────────────────────────────────────────
+function tick(timestamp: number) {
+  const dt = lastFrameTime > 0 ? Math.min(timestamp - lastFrameTime, MAX_DT) : DEFAULT_DT;
+  lastFrameTime = timestamp;
+
+  const friction = Math.pow(PAN_FRICTION, dt / 16);
+
+  // Momentum panning
+  if (
+    !isDragging &&
+    (Math.abs(panVelocity) > PAN_STOP_THRESHOLD || Math.abs(panVelocityY) > PAN_STOP_THRESHOLD_Y)
+  ) {
+    viewCenter += panVelocity * dt;
+    viewYRatio += (panVelocityY * dt) / height;
+    panVelocity *= friction;
+    panVelocityY *= friction;
+    if (Math.abs(panVelocity) < PAN_STOP_THRESHOLD) panVelocity = 0;
+    if (Math.abs(panVelocityY) < PAN_STOP_THRESHOLD_Y) panVelocityY = 0;
+  }
+
+  // Smooth zoom
+  if (isZoomAnimating && !isDragging) {
+    const spanDiff = targetSpan - viewSpan;
+    if (Math.abs(spanDiff) < ZOOM_SETTLE_THRESHOLD) {
+      viewSpan = targetSpan;
+      viewCenter = zoomAnchorTime + (0.5 - zoomAnchorRatio) * viewSpan;
+      isZoomAnimating = false;
+    } else {
+      const zf = 1 - Math.pow(1 - ZOOM_EASE, dt / 16);
+      viewSpan += spanDiff * zf;
+      viewCenter = zoomAnchorTime + (0.5 - zoomAnchorRatio) * viewSpan;
+    }
+  }
+
+  // Keyboard pan
+  if (isPanAnimating && !isDragging) {
+    const panDiff = targetCenter - viewCenter;
+    const msPerPx = viewSpan / (width - PADDING.left - PADDING.right);
+    if (Math.abs(panDiff) < msPerPx * PAN_SETTLE_THRESHOLD_PX) {
+      viewCenter = targetCenter;
+      isPanAnimating = false;
+    } else {
+      viewCenter += panDiff * (1 - Math.pow(1 - PAN_EASE, dt / 16));
+    }
+  }
+
+  // Y-axis pan
+  if (isYPanAnimating && !isDragging) {
+    const yDiff = targetYRatio - viewYRatio;
+    if (Math.abs(yDiff) < YRATIO_SETTLE_THRESHOLD) {
+      viewYRatio = targetYRatio;
+      isYPanAnimating = false;
+    } else {
+      viewYRatio += yDiff * (1 - Math.pow(1 - PAN_EASE, dt / 16));
+    }
+  }
+
+  // Follow mode
+  if (followEnabled && !isDragging) {
+    const target = Date.now();
+    viewCenter += (target - viewCenter) * (1 - Math.pow(1 - FOLLOW_EASE, dt / 16));
+  }
+
+  // Step value animations & prune offscreen entries
+  const result = stepAnimations(animMap, dt, animatedMaxY);
+  animatedMaxY = result.maxY;
+  pruneAnimMap(animMap, viewStart, viewEnd, viewSpan);
+
+  render();
+  animationFrameId = requestAnimationFrame(tick);
+}
+
+// ── Animation sync ──────────────────────────────────────────────────
+$effect(() => {
+  const slots = data;
+  const subs = allSubjects;
+  const enabled = enabledSubjects;
+  syncAnimTargets(animMap, slots, subs, enabled);
+});
+
+// Request data whenever the visible window changes.
+$effect(() => {
+  store.requestRange(viewStart, viewEnd);
+});
+
+// ── Lifecycle ───────────────────────────────────────────────────────
+onMount(() => {
+  updateSize();
+
+  const ro = new ResizeObserver(updateSize);
+  if (containerEl) ro.observe(containerEl);
+
+  window.addEventListener("blur", onWindowBlur);
+
+  viewCenter = Date.now();
+  targetCenter = viewCenter;
+  targetSpan = viewSpan;
+  canvasEl?.focus();
+  animationFrameId = requestAnimationFrame(tick);
+
+  return () => {
+    cancelAnimationFrame(animationFrameId);
+    ro.disconnect();
+    window.removeEventListener("blur", onWindowBlur);
+    store.dispose();
+  };
+});
+</script>
+
+<div class="absolute inset-0 select-none" bind:this={containerEl}>
+  <canvas
+    bind:this={canvasEl}
+    class="w-full h-full cursor-grab outline-none"
+    class:cursor-grabbing={isDragging}
+    style="display: block; touch-action: none;"
+    tabindex="0"
+    aria-label="Interactive enrollment timeline chart"
+    onpointerdown={(e) => { canvasEl?.focus(); onPointerDown(e); }}
+    onpointermove={onPointerMove}
+    onpointerup={onPointerUp}
+    onpointerleave={onPointerLeave}
+    onpointercancel={onPointerCancel}
+    onwheel={onWheel}
+    onkeydown={onKeyDown}
+    onkeyup={onKeyUp}
+  ></canvas>
+
+  <TimelineDrawer
+    bind:open={drawerOpen}
+    subjects={allSubjects}
+    {enabledSubjects}
+    {followEnabled}
+    onToggleSubject={toggleSubject}
+    onEnableAll={enableAll}
+    onDisableAll={disableAll}
+    onResumeFollow={resumeFollow}
+  />
+
+  <TimelineTooltip
+    visible={tooltipVisible}
+    x={tooltipX}
+    y={tooltipY}
+    slot={tooltipSlot}
+    {activeSubjects}
+  />
+</div>
@@ -0,0 +1,138 @@
+<script lang="ts">
+import { Filter, X } from "@lucide/svelte";
+import { getSubjectColor } from "$lib/timeline/data";
+import { DRAWER_WIDTH } from "$lib/timeline/constants";
+
+interface Props {
+  open: boolean;
+  subjects: readonly string[];
+  enabledSubjects: Set<string>;
+  followEnabled: boolean;
+  onToggleSubject: (subject: string) => void;
+  onEnableAll: () => void;
+  onDisableAll: () => void;
+  onResumeFollow: () => void;
+}
+
+let {
+  open = $bindable(),
+  subjects,
+  enabledSubjects,
+  followEnabled,
+  onToggleSubject,
+  onEnableAll,
+  onDisableAll,
+  onResumeFollow,
+}: Props = $props();
+
+function onKeyDown(e: KeyboardEvent) {
+  if (e.key === "Escape" && open) {
+    open = false;
+  }
+}
+</script>
+
+<svelte:window onkeydown={onKeyDown} />
+
+<!-- Filter toggle button — slides out when drawer opens -->
+<button
+  class="absolute right-3 z-50 p-2 rounded-md
+    bg-black text-white dark:bg-white dark:text-black
+    hover:bg-neutral-800 dark:hover:bg-neutral-200
+    border border-black/20 dark:border-white/20
+    shadow-md transition-all duration-200 ease-in-out cursor-pointer
+    {open ? 'opacity-0 pointer-events-none' : 'opacity-100'}"
+  style="top: 20%; transform: translateX({open ? '60px' : '0'});"
+  onclick={() => (open = true)}
+  aria-label="Open filters"
+>
+  <Filter size={18} strokeWidth={2} />
+</button>
+
+<!-- Drawer panel -->
+<div
+  class="absolute right-0 z-40 rounded-l-lg shadow-xl transition-transform duration-200 ease-in-out {open ? '' : 'pointer-events-none'}"
+  style="top: 20%; width: {DRAWER_WIDTH}px; height: 60%; transform: translateX({open
+    ? 0
+    : DRAWER_WIDTH}px);"
+>
+  <div
+    class="h-full flex flex-col bg-background/90 backdrop-blur-md border border-border/40 rounded-l-lg overflow-hidden"
+    style="width: {DRAWER_WIDTH}px;"
+  >
+    <!-- Header -->
+    <div class="flex items-center justify-between px-3 py-2.5 border-b border-border/40">
+      <span class="text-xs font-semibold text-foreground">Filters</span>
+      <button
+        class="p-0.5 rounded text-muted-foreground hover:text-foreground transition-colors cursor-pointer"
+        onclick={() => (open = false)}
+        aria-label="Close filters"
+      >
+        <X size={14} strokeWidth={2} />
+      </button>
+    </div>
+
+    <!-- Follow status -->
+    <div class="px-3 py-2 border-b border-border/40">
+      {#if followEnabled}
+        <div
+          class="px-2 py-1 rounded-md text-[10px] font-medium text-center
+            bg-green-500/10 text-green-600 dark:text-green-400 border border-green-500/20"
+        >
+          FOLLOWING
+        </div>
+      {:else}
+        <button
+          class="w-full px-2 py-1 rounded-md text-[10px] font-medium text-center
+            bg-muted/80 text-muted-foreground hover:text-foreground
+            border border-border/50 transition-colors cursor-pointer"
+          onclick={onResumeFollow}
+          aria-label="Resume following current time"
+        >
+          FOLLOW
+        </button>
+      {/if}
+    </div>
+
+    <!-- Subject toggles -->
+    <div class="flex-1 overflow-y-auto px-3 py-2">
+      <div class="flex items-center justify-between mb-2 text-[10px] text-muted-foreground">
+        <span class="uppercase tracking-wider font-medium">Subjects</span>
+        <div class="flex gap-1.5">
+          <button
+            class="hover:text-foreground transition-colors cursor-pointer"
+            onclick={onEnableAll}>All</button
+          >
+          <span class="opacity-40">|</span>
+          <button
+            class="hover:text-foreground transition-colors cursor-pointer"
+            onclick={onDisableAll}>None</button
+          >
+        </div>
+      </div>
+      <div class="space-y-0.5">
+        {#each subjects as subject}
+          {@const enabled = enabledSubjects.has(subject)}
+          {@const color = getSubjectColor(subject)}
+          <button
+            class="flex items-center gap-2 w-full px-1.5 py-1 rounded text-xs
+              hover:bg-muted/50 transition-colors cursor-pointer text-left"
+            onclick={() => onToggleSubject(subject)}
+          >
+            <span
+              class="inline-block w-3 h-3 rounded-sm shrink-0 transition-opacity"
+              style="background: {color}; opacity: {enabled ? 1 : 0.2};"
+            ></span>
+            <span
+              class="transition-opacity {enabled
+                ? 'text-foreground'
+                : 'text-muted-foreground/50'}"
+            >
+              {subject}
+            </span>
+          </button>
+        {/each}
+      </div>
+    </div>
+  </div>
+</div>
@@ -0,0 +1,52 @@
+<script lang="ts">
+import { timeFormat } from "d3-time-format";
+import { getSubjectColor } from "$lib/timeline/data";
+import type { TimeSlot } from "$lib/timeline/types";
+import { enabledTotalClasses } from "$lib/timeline/viewport";
+
+interface Props {
+  visible: boolean;
+  x: number;
+  y: number;
+  slot: TimeSlot | null;
+  activeSubjects: readonly string[];
+}
+
+let { visible, x, y, slot, activeSubjects }: Props = $props();
+
+const fmtTime = timeFormat("%-I:%M %p");
+</script>
+
+{#if visible && slot}
+  {@const total = enabledTotalClasses(slot, activeSubjects)}
+  <div
+    class="pointer-events-none fixed z-50 rounded-lg border border-border/60 bg-background/95
+      backdrop-blur-sm shadow-lg px-3 py-2 text-xs min-w-[140px]"
+    style="left: {x + 12}px; top: {y - 10}px; transform: translateY(-100%);"
+  >
+    <div class="font-semibold text-foreground mb-1.5">
+      {fmtTime(slot.time)}
+    </div>
+    <div class="space-y-0.5">
+      {#each activeSubjects as subject}
+        {@const count = slot.subjects[subject] || 0}
+        {#if count > 0}
+          <div class="flex items-center justify-between gap-3">
+            <div class="flex items-center gap-1.5">
+              <span
+                class="inline-block w-2 h-2 rounded-sm"
+                style="background: {getSubjectColor(subject)}"
+              ></span>
+              <span class="text-muted-foreground">{subject}</span>
+            </div>
+            <span class="font-medium tabular-nums">{count}</span>
+          </div>
+        {/if}
+      {/each}
+    </div>
+    <div class="mt-1.5 pt-1.5 border-t border-border/40 flex justify-between font-medium">
+      <span>Total</span>
+      <span class="tabular-nums">{total}</span>
+    </div>
+  </div>
+{/if}
@@ -0,0 +1,118 @@
+import {
+  type RowData,
+  type TableOptions,
+  type TableOptionsResolved,
+  type TableState,
+  createTable,
+} from "@tanstack/table-core";
+
+/**
+ * Creates a reactive TanStack table for Svelte 5 using runes.
+ *
+ * Adapted from shadcn-svelte's data-table wrapper — uses `$state` and
+ * `$effect.pre` instead of Svelte stores for reactivity.
+ */
+export function createSvelteTable<TData extends RowData>(options: TableOptions<TData>) {
+  const resolvedOptions: TableOptionsResolved<TData> = mergeObjects(
+    {
+      state: {},
+      onStateChange() {},
+      renderFallbackValue: null,
+      mergeOptions: (
+        defaultOptions: TableOptions<TData>,
+        options: Partial<TableOptions<TData>>
+      ) => {
+        return mergeObjects(defaultOptions, options);
+      },
+    },
+    options
+  );
+
+  const table = createTable(resolvedOptions);
+  let state = $state<Partial<TableState>>(table.initialState);
+
+  function updateOptions() {
+    table.setOptions((prev) => {
+      return mergeObjects(prev, options, {
+        state: mergeObjects(state, options.state || {}),
+
+        // eslint-disable-next-line @typescript-eslint/no-explicit-any
+        onStateChange: (updater: any) => {
+          if (updater instanceof Function) state = updater(state);
+          else state = { ...state, ...(updater as Partial<TableState>) };
+
+          options.onStateChange?.(updater);
+        },
+      });
+    });
+  }
+
+  updateOptions();
+
+  $effect.pre(() => {
+    updateOptions();
+  });
+
+  return table;
+}
+
+type MaybeThunk<T extends object> = T | (() => T | null | undefined);
+type Intersection<T extends readonly unknown[]> = (T extends [infer H, ...infer R]
+  ? H & Intersection<R>
+  : unknown) & {};
+
+/**
+ * Lazily merges several objects (or thunks) while preserving
+ * getter semantics from every source. Proxy-based.
+ */
+// eslint-disable-next-line @typescript-eslint/no-explicit-any
+export function mergeObjects<Sources extends readonly MaybeThunk<any>[]>(
+  ...sources: Sources
+): Intersection<{ [K in keyof Sources]: Sources[K] }> {
+  const resolve = <T extends object>(src: MaybeThunk<T>): T | undefined =>
+    typeof src === "function" ? (src() ?? undefined) : src;
+
+  const findSourceWithKey = (key: PropertyKey) => {
+    for (let i = sources.length - 1; i >= 0; i--) {
+      const obj = resolve(sources[i]);
+      if (obj && key in obj) return obj;
+    }
+    return undefined;
+  };
+
+  return new Proxy(Object.create(null), {
+    get(_, key) {
+      const src = findSourceWithKey(key);
+      return src?.[key as never];
+    },
+
+    has(_, key) {
+      return !!findSourceWithKey(key);
+    },
+
+    ownKeys(): (string | symbol)[] {
+      const all = new Set<string | symbol>();
+      for (const s of sources) {
+        const obj = resolve(s);
+        if (obj) {
+          for (const k of Reflect.ownKeys(obj) as (string | symbol)[]) {
+            all.add(k);
+          }
+        }
+      }
+      return [...all];
+    },
+
+    getOwnPropertyDescriptor(_, key) {
+      const src = findSourceWithKey(key);
+      if (!src) return undefined;
+      return {
+        configurable: true,
+        enumerable: true,
+        // eslint-disable-next-line @typescript-eslint/no-explicit-any
+        value: (src as any)[key],
+        writable: true,
+      };
+    },
+  }) as Intersection<{ [K in keyof Sources]: Sources[K] }>;
+}
@@ -0,0 +1,54 @@
+<script lang="ts" module>
+// eslint-disable-next-line @typescript-eslint/no-explicit-any
+export type FlexRenderProps<TProps = any> = {
+  content: unknown;
+  context: TProps;
+};
+</script>
+
+<script lang="ts">
+  import { isRenderComponentConfig, isRenderSnippetConfig, mountComponent } from "./render-helpers.js";
+
+  let { content, context }: FlexRenderProps = $props();
+
+  function renderAction(node: HTMLElement, contentVal: typeof content) {
+    let cleanup: (() => void) | undefined;
+
+    function render(c: typeof content) {
+      cleanup?.();
+      node.textContent = "";
+
+      if (isRenderComponentConfig(c)) {
+        cleanup = mountComponent(c.component, node, c.props as Record<string, unknown>);
+      }
+    }
+
+    render(contentVal);
+
+    return {
+      update(newContent: typeof content) {
+        render(newContent);
+      },
+      destroy() {
+        cleanup?.();
+      },
+    };
+  }
+</script>
+
+{#if isRenderSnippetConfig(content)}
+  {@render content.snippet(content.props)}
+{:else if isRenderComponentConfig(content)}
+  <div use:renderAction={content}></div>
+{:else if typeof content === "function"}
+  {@const result = content(context)}
+  {#if isRenderComponentConfig(result)}
+    <div use:renderAction={result}></div>
+  {:else if isRenderSnippetConfig(result)}
+    {@render result.snippet(result.props)}
+  {:else if typeof result === "string" || typeof result === "number"}
+    {result}
+  {/if}
+{:else if typeof content === "string" || typeof content === "number"}
+  {content}
+{/if}
--- a/Show More
+++ b/Show More
Author	SHA1	Message	Date
Xevion	669dec0235	feat: add timeline API with schedule-aware enrollment aggregation Implements POST /api/timeline endpoint that aggregates enrollment by subject over 15-minute slots, filtering courses by their actual meeting times. Includes ISR-style schedule cache with hourly background refresh using stale-while-revalidate pattern, database indexes for efficient queries, and frontend refactor to dynamically discover subjects from API.	2026-01-30 10:56:11 -06:00
Xevion	67ba63339a	fix: instructor/course mismatching, build order-independent map for association	2026-01-30 09:53:03 -06:00
Xevion	7b8c11ac13	feat: add calendar export endpoints for ICS and Google Calendar	2026-01-30 04:08:16 -06:00
Xevion	a767a3f8be	feat: add root error page handling	2026-01-30 04:07:53 -06:00
Xevion	8ce398c0e0	feat: add scraper analytics dashboard with timeseries and subject monitoring	2026-01-30 03:46:48 -06:00
Xevion	9fed651641	feat: add adaptive scheduling and scraper admin endpoints Subjects now have individually calculated scrape intervals based on their historical change ratio, consecutive zero-change runs, failure counts, and the current time of day. This reduces unnecessary scrapes during inactive periods while maintaining responsiveness during peak hours. Includes four new admin endpoints for monitoring scraper health and scheduling decisions.	2026-01-30 02:14:37 -06:00
Xevion	75a99c10ea	feat: add scrape job result persistence for effectiveness tracking	2026-01-30 01:37:41 -06:00
Xevion	857ceabcca	fix: prevent ts-rs serde warnings	2026-01-30 01:36:57 -06:00
Xevion	203c337cf0	feat: add confidence-based RMP matching with manual review workflow Replace simple auto-matching with scored candidate generation that considers department overlap, name uniqueness, and rating volume. Candidates above 0.85 auto-accept; others require admin approval.	2026-01-30 01:31:11 -06:00
Xevion	39ba131322	feat: add mobile touch controls with gesture detection	2026-01-29 23:56:45 -06:00
Xevion	2fad9c969d	fix: avoid title on icon, use simpler href-based login redirect	2026-01-29 23:44:05 -06:00
Xevion	47b4f3315f	feat: enhance login page with FAQ section and improved styling	2026-01-29 23:40:48 -06:00
Xevion	fa28f13a45	feat: add interactive timeline visualization for class times Implements a canvas-based timeline view with D3 scales showing class counts across subjects. Features drag-to-pan, mouse wheel zoom, subject filtering, hover tooltips, and smooth animations. Timeline auto-follows current time and supports keyboard navigation.	2026-01-29 23:19:39 -06:00
Xevion	5a6ea1e53a	fix: handle backend startup delays with retry logic in auth	2026-01-29 20:04:50 -06:00
Xevion	ba2b2fc50a	fix: increase Banner API timeouts to handle slow responses	2026-01-29 19:49:57 -06:00
Xevion	cfe098d193	feat: add websocket support for real-time scrape job monitoring	2026-01-29 19:31:04 -06:00
Xevion	d861888e5e	fix: proper centering for login page content, avoid unnecssary scrollbar	2026-01-29 18:05:50 -06:00
Xevion	f0645d82d9	refactor: persist audit log state in module scope for cross-navigation caching	2026-01-29 17:54:27 -06:00
Xevion	7a1cd2a39b	refactor: centralize number formatting with locale-aware utility	2026-01-29 17:53:38 -06:00
Xevion	d2985f98ce	feat: enhance audit log with smart diffing, conditional request caching, auto refreshing	2026-01-29 17:35:11 -06:00
Xevion	b58eb840f3	refactor: consolidate navigation with top nav bar and route groups	2026-01-29 17:01:47 -06:00
Xevion	2bc6fbdf30	feat: implement relative time feedback and improve tooltip customization	2026-01-29 16:44:06 -06:00
Xevion	e41b970d6e	fix: implement i64 serialization for JavaScript compatibility, fixing avatar URL display	2026-01-29 15:51:19 -06:00
Xevion	e880126281	feat: implement worker timeout protection and crash recovery for job queue Add JOB_TIMEOUT constant to fail stuck jobs after 5 minutes, and LOCK_EXPIRY to reclaim abandoned locks after 10 minutes. Introduce force_unlock_all to recover orphaned jobs at startup. Fix retry limit off-by-one error and update deduplication to include locked jobs.	2026-01-29 15:50:09 -06:00
Xevion	db0ec1e69d	feat: add rmp profile links and confidence-aware rating display	2026-01-29 15:43:39 -06:00
Xevion	2947face06	fix: run frontend build first with -e embed flag in Justfile	2026-01-29 15:00:13 -06:00
Xevion	36bcc27d7f	feat: setup smart page transitions, fix laggy theme-aware element transitions	2026-01-29 14:59:47 -06:00
Xevion	9e403e5043	refactor: modernize Justfile commands and simplify service management	2026-01-29 14:33:16 -06:00
Xevion	98a6d978c6	feat: implement course change auditing with time-series metrics endpoint	2026-01-29 14:19:36 -06:00
Xevion	4deeef2f00	feat: optimize asset delivery with build-time compression and encoding negotiation	2026-01-29 13:56:10 -06:00
Xevion	e008ee5a12	feat: show search duration and result count feedback	2026-01-29 13:15:25 -06:00
Xevion	a007ccb6a2	fix: remove out:fade transition from CourseTable	2026-01-29 13:08:45 -06:00
Xevion	527cbebc6a	feat: implement user authentication system with admin dashboard	2026-01-29 12:56:51 -06:00
Xevion	4207783cdd	docs: add changelog entries and roadmap completion tracking	2026-01-29 12:27:46 -06:00
Xevion	c90bd740de	refactor: consolidate query logic and eliminate N+1 instructor loads	2026-01-29 12:03:06 -06:00
Xevion	61f8bd9de7	refactor: consolidate menu snippets and strengthen type safety Replaces duplicated dropdown/context menu code with parameterized snippet, eliminates unsafe type casts, adds error handling for clipboard and API calls, and improves accessibility annotations.	2026-01-29 11:40:55 -06:00
Xevion	b5eaedc9bc	feat: add delivery mode indicators and tooltips to location column	2026-01-29 11:32:35 -06:00
Xevion	58475c8673	feat: add page selector dropdown with animated pagination controls Replace Previous/Next buttons with 5-slot page navigation centered in pagination bar. Current page becomes a dropdown trigger allowing direct page jumps. Side slots animate on page transitions.	2026-01-29 11:31:55 -06:00
Xevion	78159707e2	feat: table FLIP animations, improved time tooltip details & day abbreviations	2026-01-29 03:40:40 -06:00
Xevion	779144a4d5	feat: implement smart name abbreviation for instructor display	2026-01-29 03:14:55 -06:00
Xevion	0da2e810fe	feat: add multi-select subject filtering with searchable comboboxes	2026-01-29 03:03:21 -06:00
Xevion	ed72ac6bff	refactor: extract reusable SimpleTooltip component and enhance UI hints	2026-01-29 01:37:04 -06:00
Xevion	57b5cafb27	feat: enhance table scrolling and eliminate initial theme flash	2026-01-29 01:18:02 -06:00
Xevion	841191c44d	feat: integrate OverlayScrollbars with theme-aware styling	2026-01-29 01:05:19 -06:00
Xevion	67d7c81ef4	feat: implement interactive data table with sorting and column control Replaces static course table with TanStack Table featuring sortable columns, column visibility management, and server-side sort handling. Adds reusable data-table primitives adapted for Svelte 5 runes.	2026-01-29 01:04:18 -06:00
Xevion	d108a41f91	feat: sync RMP professor ratings and display in course search interface	2026-01-29 00:26:40 -06:00
Xevion	5fab8c216a	feat: add course search UI with ts-rs type bindings Integrate ts-rs for Rust-to-TypeScript type generation, build course search page with filters, pagination, and expandable detail rows, and refactor theme toggle into a reactive store with view transition animation.	2026-01-28 22:11:17 -06:00
Xevion	15256ff91c	docs: replace feature wishlist with organized roadmap and changelog	2026-01-28 21:07:10 -06:00
Xevion	6df4303bd6	feat: implement comprehensive course data model with reference cache and search	2026-01-28 21:06:53 -06:00
Xevion	e3b855b956	refactor: migrate frontend from React to SvelteKit	2026-01-28 21:06:53 -06:00