Repository navigation
Conversation
The four client quotes existed only in the CV PDF, so the page made the claims and the PDF carried the proof — reachable only by downloading it. This puts the same four on the page, directly above the CTA, so a visitor hits the evidence right before the ask. Two columns rather than reusing the three-up .fix grid: these quotes run longer than the "called in for" ones, and at three-up the attribution lines wrap to four. Styles follow the existing .fix grammar (Cormorant italic quote, muted sans attribution) and collapse to one column at the same 860px breakpoint as the rest of the page. Also adds the NLP Certification chip, which went onto the CV but never onto the page — same drift, one line to close. Quotes and attributions are copied verbatim from the merged CV, so the anonymization decisions carry over unchanged. Verified in Chromium at 1280/860/390: no horizontal overflow, no card overlap, all cards inside the wrap padding. Element positions measured directly rather than trusting scrollWidth alone.
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
|
Gate blocked — not merging. The reviewer-unavailable substitution in Confirmed on two separate invocations, so this is credit exhaustion rather than a transient stream error. No cross-vendor review has landed on this PR either ( The vault's 2026-08-13 entry is explicit that an outage relaxes which reviewer, never how many passes — so with no reviewer of any kind reachable, there is no version of this that satisfies the gate. Zero clean passes stand against the required two. Re-running in hope of a different result would be gaming the gate rather than satisfying it, so this PR stays open and unmerged until either the OpenAI billing is topped up or a second vendor is bound. Nothing here is urgent: the change is additive, My own verification stands on its own record (it does not substitute for the gate — it is the rendering evidence Codex could not have produced anyway): rendered in real Chromium at 1280 / 860 / 390, measuring element positions directly rather than trusting Generated by Claude Code |
On a page that is otherwise entirely about AI, "NLP Certification" reads as Natural Language Processing. The credential is neuro-linguistic programming — a behavioral qualification that backs the "Behavioral Strategist" half of the role line, and which the ambiguous form actively undersells by looking like a thin AI cert. Raised by the second Claude review pass. The same pass argued the chip should also name an issuing body like its siblings; that does not hold — "Licensed Chemist (RCh)" and "ITIL v3 Foundations" name none either, so the chip row has no such convention to break. Note: angeline.pdf still carries the short form, so the CV and the page now differ on this one chip until the PDF is regenerated.
Marivic's quote published her husband's private reaction — that he declined the offer, then wanted in. Her permission covers her own words, not his, and combined with "Founder, antique silver and collectibles" (a trade narrow enough to identify her in this market) he became identifiable by association. Now reads "someone close to me", which keeps the point the quote is making and drops the exposure. Raised by the second Claude review pass and decided by Chii. I flagged the trade descriptor's narrowness when we added it, but not how it compounds with the relationship detail — that interaction is what the review caught. Applied to the CV as well as the page. The same sentence was already live in the merged angeline.pdf, which the page links to, so changing only the page would have left the exposure in place while appearing to fix it. The CV regeneration also picks up the Neuro-Linguistic Programming chip expansion, closing the drift that commit opened. CV verified after regeneration: 2 pages, only the seven intended font faces (no fallback, no Type3), no header/footer white gaps, footer at 792.9pt. The new wording introduces no character outside the existing font subset, and the natural footer position is unchanged at 736.65pt, so the measured 75px spacer still holds.
These are real attributed quotes, but they shipped as bare sibling <p> elements, so a screen reader got no quote boundary and nothing tying a quote to its attribution. Raised by two separate Claude review passes. The #fix quotes stay plain <p> deliberately: those are composite lines written to voice a problem, not anyone's actual words, so blockquote would misrepresent them. The distinction is intentional, not drift. Default figure and blockquote margins are zeroed so rendering is unchanged — reverified in Chromium at 1280/860/390 and the geometry is identical to the previous commit (same maxRight/minLeft at every width, no overflow, no overlap). Note on the verification itself: the first re-run reported "no overlap" against zero elements, because the checker still selected `.says > div` after the cards became <figure>. Fixed the selector and re-ran; the numbers above are from the corrected check, not the vacuous one. angeline.pdf is unchanged by this commit — the print pipeline does not emit a tagged PDF, so the semantic markup would not carry through.
The four client quotes existed only in the CV PDF, so the page made the claims and the PDF carried the proof — reachable only by downloading it. This puts the same four on the page, directly above the CTA, so a visitor hits the evidence right before the ask.
What changed
/angeline, placed between the credentials block and the CTA..fixgrid: these quotes run longer than the "called in for" ones, and at three-up the attribution lines wrap to four. Collapses to one column at the same 860px breakpoint as the rest of the page..fixgrammar — Cormorant italic for the quote, muted sans for the attribution — so the section reads as part of the page rather than bolted on.Quotes and attributions are copied verbatim from the merged CV, so the anonymization decisions carry over unchanged.
Verification
Rendered in real Chromium at 1280 / 860 / 390 and measured element positions directly rather than trusting
scrollWidthalone (per the 2026-08-14 vault precedent — a page withoverflow-x: hiddenwill silently clip a real overflow instead of surfacing it):Screenshots of the section at desktop and mobile were reviewed; both read cleanly with no orphaned attribution lines.
Generated by Claude Code