Redesign an existing website to make it better. Built on top of impeccable.
88%
Does it follow best practices?
Run evals on this skill
Adds up to 20 points to the overall score
Low
Low-risk findings worth noting
The scorecard is the audit's headline claim, so it must be reproducible:
two auditors handed the same audit.json measurements should land within
±5 points per dimension. This file makes that possible — every dimension
has observable anchors at 40 / 70 / 90, and the procedure below forbids
scoring without evidence.
| id | name | weight | primary evidence |
|---|---|---|---|
brand-expression | Brand expression | 10 | brandColorShare, radiusSprawl, T-tokens-unused, T-color-imbalance, Phase 5 benchmarks |
visual-hierarchy-craft | Visual hierarchy & craft | 15 | typeScale, impeccable critique, VISION-pass screenshot observations |
conversion-focus | Conversion focus | 20 | ctaFragmentation, T-cta-vocab, T-nav-conflict, above-the-fold analysis |
accessibility | Accessibility | 10 | contrast, altCoverage, landmarks, T-link-content-free, T-img-alt-* |
technical-seo | Technical SEO | 15 | Phase 3 check results (robots, sitemap, canonical, titles, OG, JSON-LD) |
content-llm-visibility | Content & LLM visibility | 15 | Phase 4 results (llms.txt, answerability, schema coverage, key-facts crawlability) |
performance | Performance (Core Web Vitals) | 15 | cwv measurement (LCP / CLS / TBT, mobile + desktop) |
Weights sum to 100. Conversion focus carries the most weight because the audit's promise is a better business outcome, and the conversion path is the shortest lever; brand expression and accessibility carry less not because they matter less but because their business effect routes through the other dimensions.
Score each dimension by finding the band the evidence fits. Anchors are observable states, not adjectives — if you cannot point at the measurement or screenshot that satisfies the anchor, you are in the band below.
brand-expression — does the site express its own brand, or a template's?T-tokens-unused); typography is a bare system stack with no
display register; the page could belong to any company in the vertical.visual-hierarchy-craft — does the eye know where to go, and does the craft hold up close?typeScale.kind === "ad-hoc", no
consistent ratio); three or more competing focal points in the first
viewport; the VISION pass names a dated pattern dominating the hero
(stock-photo hero + double CTA, carousel-as-homepage, wall-of-cards);
spacing is inconsistent between sibling sections.conversion-focus — does the page funnel attention to one ask?ctaFragmentation on one equivalence bucket ≥4); no primary CTA in the
first viewport, or the primary CTA competes with ≥2 equal-weight
alternatives; pricing or what-it-is is not reachable within one click of
home.T-nav-conflict).accessibility — can everyone use it?T-img-alt-empty); no
<main> landmark; content-free link labels present ("click here",
"read more" as bare links).<h1> count ≠ 1 on some page.<h1> per page with
ordered heading levels; no content-free link labels; animation respects
prefers-reduced-motion where motion exists.technical-seo — can search engines crawl, index, and represent it?<h1> count ≠ 1 on audited
pages; http does not redirect to https cleanly.content-llm-visibility — will an AI answer engine find, understand, and cite it?performance — does it load fast enough not to lose the visitor?Classify each metric against Google's thresholds first, then band:
| metric | good | poor | note |
|---|---|---|---|
| LCP | ≤ 2.5 s | > 4.0 s | |
| CLS | ≤ 0.10 | > 0.25 | |
| TBT (lab) | ≤ 200 ms | > 600 ms | lab proxy for INP — always label it as such |
| INP (field, PSI only) | ≤ 200 ms | > 500 ms | only when PageSpeed Insights field data was reachable |
audit.json#scorecard.dimensions.<id>.evidence —
measurement keys, finding ids, or tension ids. No evidence → the
dimension is not scored (see rule 3), never guessed."score": null and record the reason in notMeasured. The overall
renormalizes over the measured weights, and the report states the
omission in the scorecard hero — a missing dimension is visible, not
silently averaged.overall = round( Σ(scoreᵢ × weightᵢ) / Σ(weightᵢ over scored dimensions) ).
Integers only, no decimals anywhere in the scorecard.Findings carry P1/P2/P3 severity (P1 = actively losing business or excluding users; P2 = material drag; P3 = polish). Severity prioritizes the fix list; anchors set the score. A dimension can score 70 and still carry a P1 (one severe defect in an otherwise competent surface), and the scorecard must not be back-derived from severity counts.
.tessl-plugin
skills
audit
reference
deploy
reference
scripts
test
diff
direct
dynamics
extract
migrate
prepare-migration
prototype
assets
motion
fixtures
reference
qa
replica
reference
scripts
test
reskin
rollout
spec
stardust
uplift
reference