Azurdle

A daily Wordle-style guessing game for Azure services. See CLAUDE.md for the full design spec, security rules, database schema, and content pipeline. See ROADMAP.md for known gaps not yet scheduled for work.

Getting started

npm install
cp .env.example .env.local   # fill in Supabase + GitHub Models values
npm run dev

Commands

npm run dev               # next dev
npm run build             # production build
npm run test              # vitest — guess normalization, clue gating
npm run lint               # eslint
npm run generate:puzzle   # generate puzzle for today+14, writes to Supabase as queued (no human gate)
npm run validate:content  # schema + duplicate-window checks against Supabase puzzle rows
npm run calibrate         # standalone re-check: difficulty + fact/structure calibration
npm run check:queue       # warn if queued buffer < 7 days or reserves < 5

Apply supabase/migrations/ (in order) to a fresh Supabase project before running generate:puzzle.

How puzzles get approved

There is no human review step. npm run generate:puzzle writes a puzzle straight into the puzzles table as status = 'queued' — it's immediately live once its date arrives. The entire review gate is passesCalibration() (scripts/lib/calibration.ts): three model-graded checks (clue-1 difficulty, clue-1-3 shortlistability, and a fact/structure check covering what a human reviewer used to verify). A puzzle that fails any of them is discarded and regenerated, up to 3 attempts, before the run gives up for the day (the 14-day buffer absorbs the miss).