This page reports two different things and keeps them apart. Extraction fidelity: the extractor reads a brand's own homepage and is compared, per axis, with the hand-graded catalog entry for that same brand — a test of the parser, published below. Judgment validity: whether a score on a new page tracks what independent designers judge — a test of the judgment, pre-registered and not yet measured. Every fidelity number is reproducible from open code. No validity number is claimed.
These five brands define what graded means on tasteHQ. The rates below are extraction fidelity: how faithfully the extractor recovers each hand-graded axis from the brand's own homepage. They are not a measure of agreement with designers about other pages.
Word-level underlines replace color as the sole emphasis mechanism in display-scale headlines.
| Axis | Extraction vs hand grade |
|---|---|
| Highest fidelity | |
| palette.warmth | 1.00 · exact |
| surface.ground | 1.00 · cream ✓ |
| emphasis.mechanism | 1.00 · underline ✓ |
| Lowest fidelity | |
| surface.radius | 0.50 · pred 12–16px / actual 8–10px |
| type.pairing | 0.00 · pred single-family / actual serif/sans |
| voice.archetype | pending · awaits ANTHROPIC_API_KEY |
Full-viewport hero: single product photo centered on white, one-line SF Pro Display name, nothing else above the fold — the page is a cinematic catalog, not a website.
| Axis | Extraction vs hand grade |
|---|---|
| Highest fidelity | |
| surface.ground | 1.00 · white ✓ |
| palette.contrast | 1.00 · high ✓ |
| surface.borders | ~0.92 · hairline (25-brand) |
| Lowest fidelity | |
| imagery.strategy | pending · n=1 only (Acne) |
| motion.budget | pending · Playwright required |
| voice.archetype | pending · per-brand tau to come |
Animated gradient mesh aurora runs behind the hero — brand colors signal financial infrastructure as a living, breathing system.
| Axis | Extraction vs hand grade |
|---|---|
| Highest fidelity | |
| type.pairing | 1.00 · sans/sans ✓ |
| palette.contrast | 1.00 · high ✓ |
| surface.ground | 1.00 · white ✓ |
| Lowest fidelity | |
| palette.warmth | 0.00 · accent recovered, hue-bucket wrong |
| palette.saturation | 0.50 · ordinal map needs recalibration |
| motion.budget | pending · pervasive vs static-CSS abstention |
Every action has a keyboard shortcut shown inline as a kbd-styled badge — the keyboard is treated as the primary input device, not an accelerator for power users.
| Axis | Extraction vs hand grade |
|---|---|
| Highest fidelity | |
| surface.dark_mode | 1.00 · dark-first ✓ |
| surface.borders | ~0.92 · hairline (25-brand) |
| palette.contrast | 1.00 · high ✓ |
| Lowest fidelity | |
| voice.sentence_length | ~0.56 · sparse <main> on SPA shell |
| motion.character | pending · scroll-recording analysis |
| voice.archetype | pending · per-brand tau to come |
Helvetica Monospaced Pro at 9px, 0.033em tracking, caps throughout — one face, one size, one weight across the entire UI.
| Axis | Extraction vs hand grade |
|---|---|
| Highest fidelity | |
| type.tracking | 1.00 · loose, em-median exact |
| imagery.strategy | 1.00 · photographic ✓ (only n=1 hit) |
| surface.ground | 1.00 · white ✓ |
| Lowest fidelity | |
| emphasis.mechanism | 0.00 · pred underline / actual scale (widget CSS swamp) |
| motion.budget | abstain · 56 Shopify @keyframes vs catalog "none" |
| palette.warmth / saturation | 0.00 · no :root vars, fallback poisoned |
Every brand entry on tasteHQ ships with canonical: false in /.well-known/taste/<slug> — we mark ourselves a mirror. The brand's own homepage is the source of truth.
When a brand publishes the same shape at <brand>.com/.well-known/taste/<slug> with canonical: true, tasteHQ's mirror auto-marks itself derivative and links upstream. No vendor lock-in. Read the spec at /cosign.
Status: not yet measured. Nothing above tests whether a score on a new page tracks what competent designers judge. That study is pre-registered here before any data is collected, so the criteria cannot be tuned to the result. No agreement number will be published without the human–human inter-rater band beside it.
learn_weights.py are reserved as a permanent validation set.docs/VALIDITY-REPORT-v1.md and linked here whatever it shows; if the judge fails the criterion, this page will say so.
The fidelity numbers on this page were measured on 17 May 2026 (weights w-2026-05-001; calibration note dated May) and have not been re-measured since. Six scored axes sat below 0.50. Listing them beats hiding them.
var() resolution, body-font capture, system stacks as kind evidence). The May number is now stale in the other direction and awaits re-measurement — it is deliberately not revised by hand.w-2026-05-001 is the only weights version ever published; the self-learning loop has not yet run on real evaluation records (kendall_w is null on every axis).Per-axis fidelity is the round-trip score: fetch a brand's homepage, project raw CSS/HTML signals into grammar axes, then compare against the hand-graded catalog value. Ordinal axes use normalized distance; nominal axes use binary match.
Reference material — the 30 axes across 8 categories used by the judge. Fidelity is the round-trip score from tools/spike_roundtrip.py on the 25-brand test set; bold scores are at-target.
Axes marked pending are honest non-predictions — the extractor abstains rather than guesses. For per-brand extraction-vs-hand-grade comparisons, see the Golden Set above.
| Axis | Scale | Extraction | Fidelity | Status |
|---|
Every figure on this page is generated by code you can run. No closed evaluation set, no secret prompts.
python tools/spike_roundtrip.py --weights w-2026-05-001