- shape: dark green console prints a white paper report; brief in DESIGN.md
- colorize/typeset/layout: console token set (OKLCH), mono etched labels,
LED accents, ring gauge for overall score, sticky topbar + console,
section rhythm, scroll-linked reading-progress bar
- animate: self-hosted Motion 12.42.2; 17 tagged features (M1-M17), all
gated on prefers-reduced-motion with instant final states
- critique 32/40 persisted; fixes: mobile scroll-to-report on evaluate,
cancellable evaluation, Ctrl+Enter accelerator, SR-safe counters, and
the generator no longer leaks the internal default-audience label into
repair prompts
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
One unified repair prompt + tone selector replaces Advanced section
- generator rewrite: failure modes contribute rules/problems/role/pre to
a single composed prompt built around the original instruction; per-
family mechanisms preserved (XML rules for Claude, ### Requirements +
tail emphasis for GPT, few-shot skeleton for Gemini); test contract
rewritten for the single-prompt guarantee (46 green)
- tone chips (Neutral/Simple/Professional/Technical) in the main flow;
each maps to an audience band so the choice drives the reading-level
check; Advanced section removed
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Model picker: searchable provider + model comboboxes, July-2026 catalog
Deploys blocked on user auth only: vercel login (own account) and an
HF write token. Both are one command after that.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
- 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>
Feasibility: full local-scoring version achievable with two labeled
carve-outs (open-world factual accuracy, deep subjective quality),
both unlockable later via the Scorer adapter interface.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
