A third detector joined the Monday quality job this week, measuring design the same way the lane already measures performance and accessibility: deterministically, against fixed thresholds, with no human sampling a page by eye. Its first run found 327 instances of visual debt across nine public surfaces — the bulk of it text rendered smaller than an 11-pixel legibility floor — and one approval turned that finding into nine pull requests the same day.
The detector engine is pbakaus/impeccable, running Apache-2.0 under the hood as a Rust binary pinned by the lockfile, driven through the quality lane's own Playwright Chromium. No LLM, no API key — it loads a rendered page and checks real computed values: font size against a floor, contrast ratio between text and background, line length, padding, whether anything is visually occluded. Only its QUALITY category counts toward the estate's baseline. The engine also ships a second category, opinionated "slop" rules that contradict our dark-editorial system by design — those are counted in a single line and never acted on.
It sits next to the Core Web Vitals and WCAG 2.1 AA checks the same lane already runs weekly across dekimu.com, InvoiceUp, miniterms, MiniReplies, HelpMeNegotiate, politeno, Commander, the Dekimu ID sign-in screen, and the public verification page — the same nine surfaces, one more lens on each.
The first census, captured locally, counted 327 instances across nine rules on nine surfaces. One rule dominated: undersized-ui-text, functional copy — kickers, labels, mockup chrome — rendered below 11 pixels. Dekimu.com alone carried 42 of those instances; every other surface had its own count, from 2 to 31. The pattern repeated estate-wide because 10-pixel type had quietly become house style across nine separately-built repos, each one small enough on its own that nobody had measured it against a floor.
A nudge that can never be satisfied is an alarm everyone learns to ignore.
The finding went out as a single approval card — raise the floor to 11 pixels, estate-wide — and came back approved. Execution became nine separate pull requests, one per surface, each built in its own worktree: dekimu, InvoiceUp, miniterms, MiniReplies, HelpMeNegotiate, politeno, Commander, Dekimu ID, and the verification site. Four of the nine briefs would have stopped at the scanner's sampled strings; the rule applied instead was the whole public surface, not just what the detector happened to flag on its single crawl. Authenticated dashboards were left alone on purpose — the approval scoped to public surfaces, and the shared dashboard shell's own 10-pixel tokens are a separate decision for later.
Every pull request shipped clean and merged the same day. A live rescan afterward measured undersized-ui-text at zero on all nine surfaces — down from counts as high as 42 on one surface alone — and the baseline file the lane ratchets against was shrunk in the same session to match. That baseline can only move down by a deliberate, measured commit; nothing in the automated lane is permitted to rewrite it, so the fixed floor cannot silently erode back to 10 pixels on the next redesign without the lane noticing. What is left on the design side is smaller and more debatable: 11-pixel body copy sitting right at the new floor, all-caps kickers that are a house-style choice rather than a defect, and low-contrast accent text already tracked by the accessibility baseline.
This post was drafted by an AI system from Dekimu's public engineering record and published with automated checks, without per-post human editing.
← Back to blog