Acuity — an independent, clean-room "cognitive value" TV/film index

Acuity scores 15,883 films & series on cognitive value: an overall Acuity Quotient (AQ) on a 0–200 scale, three lenses — Depth, Insight, Craft — and a five-tier scale (Idle · Ambient · Engaging · Absorbing · Profound), with multi-facet filtering including "filter to the streaming services you actually have."

Clean-room methodology (how this was built, and why it's defensible)

Acuity is an independent clean-room (Chinese-wall) re-implementation — a functionally comparable product built without copying any protected expression of a reference product. The legal core is a strict two-agent separation:

• The Analyst (one agent) studied a mirror of the reference product and wrote SPEC.md — a functional requirements specification only: information architecture, data-schema shape, UX flows, and idea-level design parameters. It copied no code, prose, taglines, assets, or score values; the original's score numbers, tier names, and thresholds were deliberately omitted.
• The Implementer (a separate agent, run as its own process) had the reference product explicitly off-limits — it never read the mirror, fetched the site, or searched for it. It authored all of Acuity's code, copy, brand, tier names, colors, and design solely from SPEC.md and our fact-derived dataset. The spec was the only channel between the two rooms.

So the agent that wrote the shippable expression never saw the original. The full protocol, what's independent, the evidence trail, and dated per-agent attestations (Analyst, Implementer, and every later pass) are in CLEANROOM.md.

What makes it independent / defensible

• Facts only as input. The catalog comes from IMDb public datasets (titles, types, years, genres, average ratings, vote counts) — facts, not anyone's creative data.
• Our own scores. build/score.py derives the three lenses from transparent genre + quality + reach signals, blends them, then maps the result onto our own deliberately flattened distribution (median 100, range ~0–200, flatter-than-normal). No external score values are used.
• Posters hotlinked, never downloaded — served live from TMDB with attribution; IMDb data credited for non-commercial use (see the site footer and CLEANROOM.md).
• Original everything else — brand (Acuity), tier names, copy, design, and code, all authored independently from the functional spec.

Layout

SPEC.md         functional requirements (clean-room analyst output)
CLEANROOM.md    protocol + dated attestations (the paper trail)
DATA.md         catalog.json schema
build/score.py  the scoring engine (facts -> our scores -> bell curve)
build/*.json    generated catalog + distribution stats
dist/           the static site (deployed to GitHub Pages)
  index.html explore.html title.html methodology.html kids.html
  styles.css app.js _headers   data/catalog.json data/stats.json

Build / run

# regenerate scores from IMDb datasets (downloads ~230 MB of facts; gitignored)
python3 build/score.py
# serve the site
python3 -m http.server 8911 --directory dist   # http://localhost:8911

Engineering notes (deliberately better than the reference)

• Single source of truth: every page renders from dist/data/catalog.json — no hand-edited numbers.
• No inline scripts/handlers → a strict CSP is enforceable (dist/_headers).
• All-relative paths → hosts cleanly from any subdirectory on GitHub Pages.
• No secrets in the client; nothing dynamic to attack (fully static).

Hosting note

dist/_headers (CSP, X-Frame-Options, etc.) is a Netlify/Cloudflare Pages feature and is not applied by GitHub Pages. The public surface is fully static with no secrets, so practical risk is low; for enforced security headers, deploy dist/ to Netlify or Cloudflare Pages instead.