service-page-audit

name	service-page-audit
description	Audit, rate and fix a marketing service page under /services/*. Use when asked to review, rate, compare against a competitor, or fix leaked/generic copy on any of the 62 service pages. Encodes the conventions established across PRs

Service Page Audit & Fix

62 service pages render from one template. This skill is the repeatable version of a review that was rebuilt by hand four times and got two measurements wrong in the process. Run the tool first, then work the checklist.

---

0 · Hard rules — read before touching anything

Do not undo what already shipped. Every item below was a deliberate decision. If a fix seems to require reversing one, stop and ask.

Decision	PR	Do not
8 capability areas, 51 sub-capabilities, no numbering on cards (ai-cognitive-computing)	#350	Re-add numbering or collapse the taxonomy
hidePartnershipModel on pages where it was removed	#353	Re-enable the Partnership Model there
Titles trimmed to 45-60 chars, department word dropped where it cost length	#354	Re-add "Cognition" style prefixes
Bespoke hero SVG: PERCEIVE > REPRESENT > REASON > EVIDENCE	#354	Replace with the shared agentic diagram
All businessMetrics carry illustrative: true + visible disclaimer	#355	Present a number as a client result without a source
Opt-in hooks: architectureEyebrow / architectureTitle / architectureTitleHighlight / midCtaLabel / closingCta.primaryLabel / closingCta.secondaryLabel / practiceLede	#356	Hardcode these back into the template
Two-line hero statement	#357	Expand it back to a paragraph
eQORE conciergeChips / conciergeHeading / conciergeIntro; ConciergeSection takes heading + intro props; generator emits chips	#365	Remove the eQORE section (see constraints)
DEPT_SECTION_COPY in getParityService; agentic services pinned in data	#368	Move these back to per-service data, or change a department default without checking all pages in it
Real <table> semantics + explicit ARIA on the comparison section	#343	Return it to a <div> grid
Contextual internal links in toolchain and comparison	#346	Strip row.link / item.link
whatIsPara3/4 stay in the DOM when collapsed (grid-template-rows + inert)	#348	Re-gate them behind a mount condition

Standing constraints from the user:

• Do not touch images. Not the mapping, not dimensions, not the generator manifest. Report image findings, change nothing.
• The oversized logo (4500px rendered at ~104px) and the 152KB PNG are knowingly ignored.
• "Shield" appearing in breadcrumb schema is knowingly ignored.
• eQORE cannot be cut. It is integral. Make it page-specific instead.
• /services/ai-governance is parked.
• /services/intelligent-automation is blocked on a taxonomy decision (it is ~95 per cent duplicate of business-process-management).

---

1 · Set up

Work in the dedicated worktree, never the main tree — another session edits the main tree and uncommitted work there has been silently overwritten.

cd /Users/maheshkumar/kq-acc          # detached worktree, node_modules symlinked
git fetch origin && git checkout -B <branch> origin/main
# dev server on :3010 (main tree runs :3000 and is often a commit behind)

Confirm the server reflects the branch before trusting any measurement:

curl -s localhost:3010/src/data/servicesData.js | grep -c "<a string you just added>"

2 · Measure

node scripts/audit-page-quality.mjs <slug>
node scripts/audit-page-quality.mjs <slug> --compare=<competitor-url>
node scripts/audit-page-quality.mjs --all            # fleet ranking, worst first
node scripts/audit-page-quality.mjs --all --json     # machine-readable

Everything it reports comes from the rendered page, never from the data file. It exits non-zero on render errors or inherited copy, so it can gate a build.

Scoring is rubric-anchored (thresholds live in RUBRIC in the script). Two runs on the same page always agree. Do not hand out a score that is not from the tool — an unanchored number cannot be defended when challenged, and cannot be compared across pages or over time.

Calibration caveat: boilerplate percentage is measured against rendered innerText, so it includes UI chrome and is not comparable to per-data-block figures quoted in older sessions. It is comparable across pages, because every page is measured the same way.

3 · Work the checklist

The tool flags what is wrong. This is the order to fix it, cheapest lever first.

1. Inherited copy — any hit is a hard fail. Check whether the page needs its own value or whether the department default in DEPT_SECTION_COPY should change.
2. Render errors — hard fail. Usually a wrong data key (see §5).
3. Snapshot coverage — anything under 85 per cent means a crawler is missing content. Fix in scripts/generate-prerender.mjs, not the React page.
4. Content fit — does the copy belong to this service? Check off-topic word count, and read the capability names against the service name.
5. Boilerplate and second person — rewrite the offending block in the page's voice. Concrete nouns and "you", not "enterprise-grade frameworks".
6. Thin sections — under 10 words per 100px. Either give the section content or cut it.
7. FAQ depth — 8+ questions, 3 paragraphs each, with real sources.
8. E-E-A-T — blocked on a real name, title and LinkedIn from the user. Ask; do not invent one.

4 · Gate, in this order

node scripts/generate-prerender.mjs
npm run audit:all          # six gates — check the exit code, do not grep for "pass"
npm run check:prerender
npm run check:sitemap
cd frontend && npm run build

Then open the page in a browser. audit:all, check:prerender and npm run build have all passed on a page that fell to the error boundary and rendered nothing.

Finally re-run audit-page-quality.mjs <slug> and confirm the score moved for the reason you expected.

Revert shared/serviceIndex.json if only its generatedAt changed — it churns on every regeneration and conflicts across PRs.

5 · Traps that have actually bitten

The template resolves content through getParityService(). Most services carry a few hundred bytes of data and the parity layer synthesizes the rest, keyed by deptSlug. Never conclude anything about what renders by reading servicesData.js. Counting raw keys produced two wrong answers in one session: "industryHeading leaks on zero pages" (it leaked on 56) and "EXECUTION LOOP is on 3 pages" (it was on 56).

Dual emission. Content added to the React page must also be added to scripts/generate-prerender.mjs or crawlers never see it. Grep the regenerated snapshot for the new string.

Verify what renders, not what you wrote. Check the collapsed default state, not the expanded one. whatIsPara4 shipped in a PR and was never in the DOM.

Vite misses whole-file writes. After a scripted rewrite of servicesData.js, touch the file and re-curl it before measuring.

The copy gate scans code comments. US English only; every percentage needs illustrative: true or an entry in SOURCED_CLAIMS. Simplest fix is to spell the number out in prose. unauthorised / authorisation are absent from its UK list and slip through — do not rely on it for those.

Data key contracts:

Key	Contract
architectureNodes[].features	Not keyCapabilities. Wrong name throws and the whole page falls to the error boundary.
capabilityAreas[].items	Must be 'Name: prose' — the card front splits on the colon.
industryUseCases[]	Uses items or agents; the industry section only renders if this key exists.
closingCta	Whole-object override. Set title explicitly or the template default fills it in.
businessMetrics[]	Needs illustrative: true per metric to render the disclaimer.

Contrast measurement. Walk up the DOM for the first non-transparent background and composite the alpha. Measuring against document.body reports white-on-white and invents failures — that produced a bogus "9 failing footer nodes" (the real count was 1).

6 · Ship

One branch, one commit, pushed from the worktree. gh is not installed — hand the user the compare URL. State plainly in the commit message which changes are template-wide and which are page-scoped, so a reviewer is not misled into thinking a shared-component change is local to one service.