WrTrueMeta

The repository behind wrtruemeta.com: a Wild Rift meta tracker built on first-party data -- the win rates of the top players on every champion's leaderboard, captured from the game itself, not aggregated from other sites.

How the parts fit

phone (ADB) -> src/ scraper -> data/captures/ -> extraction -> data/*.json|csv
                                                       |
                        scripts/ exports  ->  web-next/src/data/  ->  the site
                                                       |
                              web/ + api/  ->  the build advisor (own deploy)

Piece	What it is
src/	The ADB scraper: captures leaderboard/profile/build frames from a real device, plus the offline frame extractor (extract_frames).
data/	Everything captured and derived: win rates, items, runes, champions, ability formulas, the official patch archive.
scripts/	The pipeline: exports (export_captures, export_json, export_engine_data, export_champion_details, build_ladder_pulse), LLM extraction (extract_formulas), patch appliers (apply_patch_*, which assert before writing), the advisor deploy, and the admin ops runner.
web-next/	The site: Next.js, deployed on Vercel from origin/main. Reads only committed JSON under web-next/src/data/.
web/ + api/	The build advisor: a Python serverless function (own Vercel project, deployed by scripts/deploy_advisor.py --deploy) that generates builds with an LLM grounded in the scraped data.
tests/	Data-integrity tests, run by CI on every push. No API keys in CI, so nothing there can scrape or call an LLM.

Day-to-day operation

The admin page on the site has an Operations panel: scrape, extract, refresh exports, fetch patch notes, publish, and redeploy the advisor are all one click. Those buttons are executed on the collection machine by:

python -m scripts.ops_runner

which also starts hidden at logon via a Startup-folder launcher. Jobs queue while it is offline.

Local setup

python -m venv .venv
.\.venv\Scripts\Activate.ps1
pip install -r requirements-scrape.txt   # scraper + pipeline
pip install -r requirements-dev.txt      # test suite

requirements.txt alone is the advisor function's slim bundle -- do not grow it casually; the serverless bundle has a 250MB ceiling.

OCR needs the Tesseract binary (winget install UB-Mannheim.TesseractOCR); Gemini-engine extraction reads GEMINI_API_KEY from web-next/.env.local. Scraping needs ADB and a connected device with Wild Rift installed. Secrets live in web-next/.env.local (gitignored) and the Vercel projects, never in the repo.

When a new patch lands

Applying a balance patch touches five surfaces, and history shows missing one goes unnoticed for weeks: advisor sources, data/runes.json -> engine export, champion details export, change history + summary, and formula re-extraction for changed champions -- then a cache-version bump and an advisor redeploy. The scripts/apply_patch_*.py scripts document every applied delta and assert current values before writing, so a double-run or a shifted base fails loudly.