Calibrated

Company marketing site

The separate company website lives in marketing/. It has two reviewable homepage directions, enterprise and approach pages, a working practice-case excerpt, and a saved JEV persona comparison. It runs independently of the accounting app, without API keys or npm dependencies:

cd marketing
npm run dev

Open the site or compare both directions. The intended company domain is calibrated.co; this local preview does not publish or change DNS.

Accounting workspace

The current product is a prompt-first accounting workspace: choose a work type, compare two anonymous drafts, make your judgment, and keep the result in Notebook.

See the product decision, the cohort and credential runbook, and verification.

Start with backend/.env.pilot.example. Explicit authored samples work without credentials. Live prompting supports server-side OpenRouter configuration, or a temporary loopback-only Codex and Claude Code bridge using existing CLI sign-ins. Browser-bound identity is implemented; Clerk is an explicitly deferred cross-device upgrade.

Local dev

make dev-backend    # uvicorn :8000, SQLite in backend/
make dev-frontend   # vite :5173, /api proxied to :8000
make test           # backend pytest
make build          # frontend production build

The public benchmark boards (/benchmarks) are built from the JSON under benchmarks/ by backend/pipeline/benchmarks.py and backend/pipeline/boards.py (make benchmarks-validate, make benchmarks-build).

Docs

• docs/pilot/ — current product: decisions, runbook, QA, local and hosted setup
• docs/DEPLOY.md — Railway (API) and Vercel (frontend)
• docs/PRD.md, docs/TICKETS.md — product requirements and work log (parts predate the accounting pivot)

Retired prototype

The original "Benchmark Stadium" / Design Arena prototype (four-model tournament, synthetic seeded votes, email-code login, ARENA_LEGACY_API) was removed from the tree. It is in git history before commit cec8f89; docs/ARCHITECTURE.md carries the pointer. Only the Bradley–Terry ratings math survived, as backend/app/pilot_ratings.py.