diff --git a/CLAUDE.md b/CLAUDE.md index 661b817..f1f1bb2 100644 --- a/CLAUDE.md +++ b/CLAUDE.md @@ -4,13 +4,14 @@ PDGA rating scraper and display app. Scrapes player ratings and course data from ## Tech Stack -- **Runtime:** Node.js 18 (Alpine in Docker) +- **Runtime:** Node.js 22 (slim/Debian-based in Docker) +- **Hosting:** Gitea (`gitea.shcizo.se/shcizo/pdga-rating`) — use `tea` CLI for issues/PRs, not `gh` - **Server:** Express with EJS templates - **Database:** SQLite3 (file: `ratings.db`, Docker: `/app/data/ratings.db`) - **Frontend:** HTMX + vanilla JS (in `public/js/`) - **Scraping:** Puppeteer (with stealth plugin) + direct HTTP - **Logging:** Pino (JSON in production, pino-pretty in dev) -- **CI/CD:** Release Please + Docker build/push to GHCR +- **CI/CD:** Gitea Actions (tag-triggered docker build/push to `gitea.shcizo.se/shcizo/pdga-rating`) ## Project Structure @@ -38,15 +39,25 @@ public/ - `LOG_LEVEL=debug npm start` — Enable debug logging - `docker compose up` — Run via Docker +**No test framework or lint setup** — `package.json` has only `start` and `dev` scripts. If adding either, document it here. + ## Conventions - **Logging:** Use `require('./logger')` (or relative path). Never use `console.log/error` in backend code. Use appropriate Pino levels: `debug` for verbose/diagnostic data, `info` for operational status, `warn` for retries/degraded state, `error` for failures, `fatal` for startup crashes. - **Frontend JS:** `console.error` is fine in `public/js/` — runs in browser, no Pino. -- **Commits:** Conventional commits (`feat:`, `fix:`, `refactor:`, `chore:`) — drives release-please. +- **Commits:** Conventional commits (`feat:`, `fix:`, `refactor:`, `chore:`, `ci:`). +- **Releases:** Manual version bump — edit `version` in `package.json` + `package-lock.json`, commit as ``, tag `v`, push commit + tag (`git push origin main v`). Triggers `.gitea/workflows/docker-build.yml` which builds and pushes the image. Auth uses repo secret `PACKAGES_TOKEN` (PAT with `write:package`) — the auto-injected `GITEA_TOKEN` does not have effective registry access. - **Scraping:** Two strategies per entity: direct HTTP (fast, preferred) with Puppeteer fallback (stealth plugin for anti-bot). Rate limiting must be respected. - **Database:** Migrations run automatically on startup in `db.js`. Schema changes go there. - **Templates:** EJS with shared layout in `views/partials/`. Pages use HTMX for dynamic content loading. +## PDGA Domain Notes + +- **Rating publication cycle:** PDGA officially recalculates ratings on the **second Tuesday of each month**. `getNextPDGAUpdateDate()` in `src/services/rating-calculator.js` computes this — round filtering uses it as cutoff. +- **Predicted rating algorithm:** `calculatePredictedRating(roundRatings)` replicates PDGA's formula — 12-mo window (expands to 24 if <8 rounds), outlier removal at ≥7 rounds (2.5σ + 100pt threshold), double-weighting of recent 25% at ≥9 rounds. Returns `{rating, stdDev, debugLog}`. +- **Rate limits:** `POST /api/refresh-round-history/:pdgaNumber` enforces a 24h cooldown per player (`src/routes/players.js`). Don't bypass — PDGA's site rate-limits aggressively. +- **Round history refresh** uses Puppeteer (stealth plugin), other scraping prefers direct HTTP. Predicted rating is recomputed and stored on each refresh. + ## Environment Variables | Variable | Default | Description |