audit: the preview's pre-flight, before/after shots, and two filed rows

PREFLIGHT.md - taste-skill §14 plus the Felhom additions, run in writing against
the DEPLOYED preview, Pass/Fail per line. Measured, not assumed:

  page height 1440: 10 922 -> 6 780 px (38% shorter)
  page height  390: 19 913 -> 10 865 px (45%, 23.6 -> 12.9 phone screens)
  initial load:     13 req / 308 KB -> 11 req / 260 KB   (same conditions)
  fully loaded:     16 req / ~540 KB -> 15 req / ~545 KB (ESSENTIALLY UNCHANGED -
                    the 93 KB portrait is paid for by dropping the 49 KB hero logo
                    and the 31 KB mono font; stated rather than spun)
  third-party hosts: 0 -> 0
  em dashes in our copy: 18 -> 0
  layout families: 8 for 11 sections (3 repeats) -> 10 for 10 (none)
  no horizontal scroll: clientWidth == scrollWidth at 390 and at 1440

Contrast: three NEW styles measured 3.88 / 3.14 / 3.57 on the first pass, all under
WCAG AA. Fixed to --text-2 and re-measured on the deployed file: 7.56 / 6.12 / 6.96.

A measurement that lied, recorded because it nearly shipped a false verdict: the
first helper reported .limits p at 1.17:1 by treating the translucent --warn-dim
as opaque. Compositing the whole ancestor stack gives the true 5.67:1, a pass. The
three real failures in the same run were real. An instrument that can report a
wrong number silently is not a measurement.

Two rows filed, both needing something this session may not do alone:
  R-930 (Business & legal, P3) the primary CTA is 4.01:1, under AA. MEASURED
        IDENTICAL ON THE LIVE PAGE, so pre-existing and site-wide, not a regression
        of this redesign. Three fixes offered, each touching brand, so it goes to
        the operator rather than being changed unilaterally.
  R-931 (Process & tooling, P4) felhom.eu serves NO compression at all - site.css
        goes over the wire as 94 310 raw bytes with no Content-Encoding even when
        gzip is offered. An nginx setting in manifests/, not a website change.
        Worth more than anything on this page.

Register: 127 -> 129 rows, 2 opened, 0 closed. register_shape_gate.py OK.

documentation/audits/redesign-2026-10-10/PREFLIGHT.md

@@ -0,0 +1,119 @@
+# Pre-flight — taste-skill §14 plus the Felhom additions
+
+Run against the **deployed** preview at `https://felhom.eu/preview/`, 2026-10-10, not against the
+file on disk. Every measurement came from the browser.
+
+## Headline numbers
+
+| | Old home page | New preview | |
+|---|---|---|---|
+| Page height, 1440 px | 10 922 px | **6 780 px** | 38% shorter |
+| Page height, 390 px | 19 913 px | **10 865 px** | 45% shorter, 23.6 → 12.9 phone screens |
+| Initial load (same conditions) | 13 requests, 308 KB | **11 requests, 260 KB** | |
+| Fully loaded (all assets) | 16 requests, ~540 KB | 15 requests, ~545 KB | **essentially unchanged** |
+| Third-party hosts | 0 | **0** | |
+| Em dashes in the page's own copy | 18 | **0** | |
+| Layout families for 11 / 10 sections | 8 (3 repeats) | **10 (no repeat)** | |
+| JetBrains Mono fetched | yes, 31 KB, for 13 characters | **no** | |
+
+**Honest reading of the weight.** The *initial* load is 48 KB lighter. The *fully scrolled* page is
+about the same: the new 93 KB portrait is paid for by dropping the 49 KB decorative `logo.png` and
+the 31 KB mono font. The redesign is not a weight win once you scroll; it is a weight win for the
+first screen, which is the one most visitors see. R-931 (no gzip anywhere on the site) is worth far
+more than anything on this page.
+
+## taste-skill §14
+
+| Check | Verdict |
+|---|---|
+| Brief inference declared (§0.B) | **Pass** — AUDIT.md §A |
+| Dial values explicit and reasoned | **Pass** — 6 / 3 / 4, AUDIT.md §I |
+| Design system chosen or aesthetic labelled honestly | **Pass** — Felhom design system v2, no new system |
+| Redesign mode detected, audit performed (§11) | **Pass** — preserve; AUDIT.md |
+| **ZERO em dashes** | **Pass with two stated exceptions** — 0 in the page's own copy. 6 remain in `<title>`/description/OG/Twitter/JSON-LD **by operator decision at STOP 1** (indexed strings), 1 in the shared footer, which gate 3 forces to be identical on 14 pages. Neither swept silently |
+| Page Theme Lock — one theme | **Pass** — dark throughout, no section inverts |
+| Color Consistency Lock — one accent | **Pass** — `--blue` / `--blue-bright` only |
+| Shape Consistency Lock — one radius | **Pass** — `--radius: 2px` everywhere |
+| Button Contrast Check | **FAIL, pre-existing, filed** — `.cta-button` is 4.01:1 (white on `#0083D8`, 16px/600) against a 4.5 floor. **Identical on the live page**, so not a regression. R-930, needs a brand decision |
+| CTA Button Wrap — no 2-line label at desktop | **Pass** |
+| Form Contrast Check | **n/a** — no form on this page |
+| Serif discipline | **n/a** — no serif |
+| Premium-consumer palette ban | **n/a** — not that brief |
+| Italic descender clearance | **n/a** — no italic display type |
+| Hero fits viewport, ≤2-line headline, ≤20-word subtext | **Pass** — headline 2 lines, subtext **20 words**, CTA and a real screenshot both above the fold at 1440 and at 390 |
+| Hero top padding ≤ `pt-24` | **Pass** — 128 px, and the hero is sized by its content; the old one was a forced `100vh` band with the content floating in the middle |
+| Hero stack ≤ 4 text elements | **Pass** — headline, subtext, CTA, figure caption. No eyebrow, no tagline under the CTA, no trust strip |
+| **Eyebrow count ≤ ceil(sections/3)** | **Pass — 0 of an allowed 4.** The cluster headings are not eyebrows (no uppercase, no tracking, they head real groups), and the comparison's column headers were deliberately de-capitalised for this rule |
+| Split-Header Ban | **Pass** — every section header stacks headline over body |
+| Zigzag Alternation Cap | **Pass** — the two image+text splits (hero, portrait) are separated by two other families |
+| No Duplicate CTA Intent | **Pass** — one label, „Jelentkezem tesztelőnek", used for the one intent in both places |
+| Logo wall = logos only | **n/a** — no logo wall |
+| Bento Background Diversity | **n/a** — no bento |
+| „Used by" wall under the hero | **n/a** |
+| **Copy Self-Audit** | **Pass** — every visible string re-read. The ten capability texts are **verbatim from the live page** on purpose: the audit's finding was that the writing is right and the layout is wrong, so no hedge („ha van", „szinte mind") was rewritten away |
+| Motion motivated | **Pass** — there is no motion. The old hero's infinite float is gone |
+| Marquee max one | **n/a** — none |
+| Navigation one line, ≤80 px | **Pass** — unchanged from the live nav, spliced verbatim |
+| **Section-Layout-Repetition** | **Pass** — 10 sections, 10 families. Was 11 sections over 8 families, `why-grid` ×3 and `apps-showcase` ×2 |
+| Bento rhythm and exact cell count | **n/a** |
+| Long lists use the right component | **Pass** — the ten-card grid became three labelled clusters (§4.9 „grouped chunks") |
+| **Real images, no fake screenshots** | **Pass** — one real dashboard screenshot in the hero, four in the gallery, one real photograph of the operator. No AI image, no stock, no picsum, no div-built preview. The decorative logo is gone from the hero |
+| No pills over images | **Pass** |
+| No photo-credit decoration | **Pass** |
+| No version footers | **Pass** |
+| No micro-meta sentences under eyebrows | **Pass** |
+| No hero-bottom decoration strip | **Pass** |
+| No floating top-right sub-text | **Pass** |
+| No filled progress-bar comparisons | **Pass** — the comparison is text in two columns |
+| No locale / weather strips | **Pass** |
+| No scroll cues | **Pass** |
+| No version labels in the hero | **Pass** |
+| **No section-numbering eyebrows / generic step labels** | **Pass** — `1..5` and `01..05` are gone; the steps are named by their verb-noun and ordered by layout |
+| No decorative dots | **Pass** |
+| No `border-t`+`border-b` on every row | **Pass** — the comparison uses one divider between rows, not a box per row |
+| Content density sane | **Pass with a stated deviation** — the ten capability paragraphs run past §4.9's 25-word guide. Deliberate: see Copy Self-Audit |
+| Quotes ≤3 lines | **n/a** — no testimonials |
+| Motion claimed = motion shown | **Pass** — `MOTION_INTENSITY: 3` claims none and ships none |
+| GSAP skeletons | **n/a** |
+| No `window.addEventListener('scroll')` | **Pass** — the only JS is the existing nav toggle and `gallery.js` |
+| Reduced motion honoured | **Pass** — the global `prefers-reduced-motion` block is present, and there is no animation to suppress |
+| Dark mode tested | **Pass** — single locked dark theme, by brand decision |
+| Mobile collapse explicit | **Pass** — every new multi-column rule has its `<900px` and `<600px` fallback written in the same block |
+| Viewport stability, no `h-screen` | **Pass** — the new hero uses no viewport-height sizing at all; the old `.hero` used `100vh` |
+| `useEffect` cleanup | **n/a** — no framework |
+| Empty / loading / error states | **n/a** — static marketing page |
+| Cards omitted in favour of spacing | **Pass** — the ten cards became three clustered lists; the comparison is a divided table, not eight boxes |
+| Icons from an allowed library | **Pass, by the Felhom override** — the self-hosted `icons.svg` sprite is the library; no npm icon package |
+| Motion isolated in client leaves | **n/a** |
+| No AI tells from §9 | **Pass** |
+| Core Web Vitals plausible | **Partly** — the hero image is `fetchpriority="high"` with width/height set, so LCP and CLS are in hand; **but nothing is compressed on the wire** (R-931), which is the real limiter |
+| One design system | **Pass** |
+
+## The Felhom additions (`felhom-web-design` §9)
+
+| Check | Verdict |
+|---|---|
+| UTF-8 BOM on every page file | **Pass** — asserted by the builder and by `site_gates.py` gate 1 |
+| Zero emoji | **Pass** — gate 2 |
+| Zero `—` including meta | **Pass with the two stated exceptions above** |
+| No `<style>` block; `site.css` `?v=` bumped | **Pass** — `?v=11` |
+| **Every request to `felhom.eu` or `stats.felhom.eu`, listed from the browser** | **Pass** — 11 requests: `felhom.eu` 8, `stats.felhom.eu` 2, third-party **0**. Full list in the session report |
+| Page weight compared and said out loud | **Pass** — table above, including the part that did not improve |
+| Body contrast at or above baseline | **Pass after a fix.** Three new styles measured **3.88 / 3.14 / 3.57** on first pass, all under AA. Moved to `--text-2` and re-measured on the deployed file: **7.56 / 6.12 / 6.96**. One pre-existing failure remains and is filed (R-930) |
+| Focus visible | **Pass** — `:focus-visible { outline: 2px solid var(--blue-bright) }` applies |
+| Reduced motion honoured | **Pass** |
+| **No horizontal scroll at 390 px** | **Pass** — `clientWidth == scrollWidth == 375` at 390, and `== 1425` at 1440 |
+| English twin in the same commit | **Deferred to Part D by design** — a preview URL cannot be a `TWINS` pair (AUDIT.md §K.6). The twin is written in the switch commit, as a real twin |
+| `repo_gates.py` green, new page registered | **Pass** — `site_gates.py` green; `preview/index.html` registered in `NO_TWIN` with its reason |
+
+## A measurement that was wrong, and how it was caught
+
+The first contrast pass reported `.limits p` at **1.17:1**, which would have been a severe failure.
+It was an error in the measuring script, not in the page: the helper walked up for the first
+non-transparent background and found `--warn-dim`, which is `rgba(224,169,62,.12)`, then treated that
+translucent colour as if it were opaque. Compositing the whole ancestor stack over the page
+background gives the true value, **5.67:1**, which passes. The three real failures in the same run
+were real and were fixed.
+
+The lesson is the project's own: an instrument that can report a wrong number silently is not a
+measurement. The corrected helper is in the session transcript and composites every layer.