- web/: no-build static SPA served by FastAPI locally and on HF Spaces;
IBM Plex Sans + JetBrains Mono (mono = verbatim material only),
restrained green-ink palette in OKLCH, verdict rows with evidence
quotes, repair cards with copy, honest empty state, skeleton loading,
reduced-motion support
- battle-tested via Playwright at desktop + mobile: fixed copy-button
overlap, /health no longer force-loads models (startup warms them in
a background thread), killed a bogus 0-part coverage repair card
(regression test added, 45 tests green)
- Dockerfile (HF Spaces, CPU torch, models baked at build), README
Space frontmatter, vercel.json for the static frontend
- PRODUCT.md + DESIGN.md capture the design system
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Step 1: scoring engine, improvement-prompt generator, API, 44 green tests
- Scorer adapter interface + LocalScorer (judge mode plugs in later)
- taxonomy: task type -> research-grounded parameters with default weights
- parameter evaluators with quoted evidence and honesty flags
- flagship generator: failure mode x task x model family, complete
paste-ready replacement prompts, no-generic-advice contract under test
- FastAPI app (/evaluate, /health, /warmup)
- curated planted-failure test pairs; classifier/constraint/regex fixes
found by the suite; question coverage reframed as honest topical proxy
after empirical probe showed bi-encoders cannot detect answerhood
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
