RomeoPro Tracker

Tracker for Romeopro's GipsyTeam thread: bankroll progress, session history, and the full post feed around the marathon.

Live site: romeo.fish

Stack

• React 18 + Vite
• Static deployment on Vercel
• GitHub Actions scraper for forum updates
• Claude API for bankroll screenshot extraction and Romeo post translations
• Compact JSON payloads with client-side caching

Local Commands

npm ci
npm run dev
npm run build
npm run test
npm run check

Scraper commands:

npm run scrape
npm run scrape:dry-run
npm run scrape:no-push
npm run scrape:full
npm run scrape:reextract
npm run scrape:translate

Leaderboard probes:

npm run leaderboards:ggwf
npm run leaderboards:coin

leaderboards:ggwf fetches the public GGWF leaderboard group and searches the default Romeo candidate nick R Romanovskyi. leaderboards:coin checks the public CoinPoker CoinMasters leaderboard page for the known Romeo aliases. Both commands are diagnostic only; they do not write app data unless --write is passed.

For the app widget:

npm run leaderboards:ggwf -- --write data/leaderboards.json

npm run dev and npm run build automatically copy data/*.json into public/data so local builds work without hitting GitHub raw URLs.

Frontend Architecture

Main application files:

• src/App.jsx — primary UI composition, charts, feed, and topic views
• src/hooks/usePostsData.js — loading, polling, new-post detection, refresh flow
• src/hooks/usePersistentState.js — typed localStorage persistence wrapper
• src/storage.js — public data loader, cache, compact payload expansion
• src/i18n.js — UI dictionary and date helpers
• src/utils.js — formatting, chart helpers, and bankroll history deduplication

The app is still a single-screen SPA, but state handling and data loading are no longer buried directly inside App.jsx.

Data Loading Strategy

Runtime data is resolved from multiple sources:

1. VITE_DATA_BASE_URL if explicitly configured
2. Same-origin /data/* files generated during local builds
3. raw.githubusercontent.com/<repo>/main/data/*

The client compares meta.lastUpdated across successful sources and keeps the freshest payload instead of trusting the first response. That avoids stale Vercel assets when code deploys and scraper commits move at different cadences.

Vercel uses scripts/vercel-ignore-build.mjs as its ignored build step. Data-only scraper and leaderboard commits are skipped on Vercel because the deployed client can read fresher JSON directly from GitHub raw; source, dependency, config, and public asset changes still build normally.

Scraper

The scraper lives in scripts/scrape.mjs.

Modes:

• normal — scan recent pages, add new posts, refresh recent likes
• full — full thread scan, refresh likes across the whole dataset
• reextract — rebuild bankroll history from Romeo screenshot posts
• translate — backfill missing English and Spanish translations

Safety switches:

• --dry-run / SCRAPE_DRY_RUN=1 — hit the forum and compute changes without writing files or running git
• --no-push / SCRAPE_NO_PUSH=1 — update local data/*.json files but skip git pull / commit / push

GitHub Actions:

• every 5 minutes: normal
• every 12 hours: full
• manual dispatch: normal, full, reextract, or translate

Translation helpers are isolated in scripts/lib/translation.mjs so scraper-specific API logic is no longer mixed into the forum parsing flow.

Data Files

• data/posts.json — full post payload
• data/posts.min.json — compact payload used by the client
• data/meta.json — bankroll summary, scraper metadata, bankroll history
• data/leaderboards.json — GGWF leaderboard snapshot used by the leaderboard widget

Fork / Deploy Notes

1. Fork the repo.
2. Connect it to Vercel.
3. Add ANTHROPIC_API_KEY if you want screenshot extraction and translations.
4. Keep scraper automation on main; do feature work on separate branches.

There is no application backend in this repo. The deployed app is static and reads JSON data produced by the scraper.