/* HAND-WRITTEN, NOT generated — deliberately kept separate from styles.scoped.css (scripts/scope-web-demo-css.mjs). That file's entire contract is "mechanical @scope rewrite, zero content change"; this file exists precisely to make a content change (suppressing UI Phase 1's do-not-ship list forbids), so mixing the two would blur what the generator's own drift-guard test is allowed to assume about its output. Loaded AFTER styles.scoped.css (see WebInstall.tsx) so these rules win the cascade at equal specificity. recon/16-adversarial-review-phase1.md, L4-1: renderShellExtras() ends with an UNCONDITIONAL renderDemoBar() (public/web/app.js:3637), on every branch of render() including the guest homepage's first paint. It is not just cosmetic — its "Signed in" control flips the public site into "Signed in as @blake" for any visitor, which is also the trigger for the MIN_LISTINGS_TO_INSTALL brick guard in floorTransport.ts. Blake's files stay byte-identical; this suppresses it visually and functionally (display:none removes it from hit-testing, so its buttons cannot be clicked) without touching app.js. */ .blake-install-scope .demo-bar { display: none !important; } /* R24 — Nick's ruling: a category with zero listings gets no tile at all (mirrors the mobile app). demoTextFixups.ts's fixEmptyCategoryTiles removes the tile control itself; these two rules only handle the layout fallout of removing items from Blake's two FIXED-count grids so no empty cell/trailing gap is left behind. Both classes (MOSAIC_REFLOW_CLASS / MEGA_REFLOW_CLASS in demoTextFixups.ts — pinned against these exact literals in demoTextFixups.test.ts, see that file) are toggled by the fixup only when fewer than the full slate of tiles survives — an untouched page (every category has listings) keeps Blake's original layout byte-for-byte. .mosaic is a named-area grid (styles.css:431-443) — grid-area pins each tile to its own cell regardless of whether it exists, so a removed tile leaves its area blank rather than being backfilled by a neighbor. .mega-grid has no named areas, but a fixed repeat(9, 1fr) template (styles.css:256) — fewer tiles auto-place left-to-right with nothing pinning a hole in the middle, but leave a TRAILING gap on the right at a fixed 9-column width. 🔴 F1 (adversarial review of #671, 2026-08-26) — TWO PRIOR ATTEMPTS, both measured broken: 1. Reused Blake's own :1024px dense auto-flow fallback (grid-template-areas:none + grid-auto-flow:dense) with "featured" tiles at grid-column:span 2. A short final row still left a large blank strip trailing it — dense packing only prevents a hole INSIDE a row, not a short LAST row. 2. repeat(auto-fit, minmax(Npx, 1fr)). auto-fit only collapses a track that is empty in EVERY row it spans — once the surviving tile count exceeds however many columns fit at the viewport's width, the grid wraps to a second row, and whichever columns that row doesn't fill stay reserved (occupied in row 1), leaving the same trailing blank strip. Measured LIVE on staging with this exact CSS injected: 1024px viewport → 717.7px blank (75.9% of the last row); 768px → 233.6px; 534px → 247.5px. The harness's own earlier "5/9 and 9/9, zero trailing gap" measurement was real but was the one case (single row, no wrap) where this approach happens to work — never re-checked across 1-8 survivors × the real breakpoints, which is what caught it. 🔴 L1 (adversarial re-review of #671, 2026-08-26) — THE --gbz-tile-cols FIX ABOVE WAS ALSO WRONG. Pinning column count 1:1 to tile count guarantees no WRAP, but it does nothing about how narrow each column gets: at high survivor counts on a phone it produces sliver tiles with clipped labels — measured 8 survivors @375px = 31.9×165px tiles, 82.6px of a 114px label clipped, starting at just 3 survivors on a phone; staging's real 5-survivor case clips 56.3px TODAY. A column CAP (stop increasing columns past some max) was considered and rejected — it reintroduces the exact partial-last- row problem F1 already fixed. FIX: flex-wrap. `display:flex; flex-wrap:wrap` on the container, `flex: 1 1 ` on each tile. Flexbox's own `flex-grow` stretches whichever tiles land in the FINAL row to fill it exactly — no leftover gap, unlike a grid track that stays whatever width the template says regardless of what's actually in it — and `flex-shrink`/the basis floor keeps every tile above a sane minimum width instead of dividing the full row by however many tiles happen to survive. Measured across the same 70-combination matrix (1-8 survivors × the real breakpoints): 0px gap AND 0px label clipping at every one, minimum tile width never below 182px desktop / 339px @375px. `display:flex` makes every grid-* property on the old rules inert, so they're deleted rather than left dead. The container's own `gap` declaration (styles.css — mosaic and mega-grid both already set it) is unaffected by the display change and keeps applying between flex items exactly as it did between grid items. `align-items: flex-start` on the mega-grid rule (not the mosaic one) is load-bearing: .mega-tile is `aspect-ratio: 1` (styles.css:257) with no explicit height, so a flex container's default `align-items: stretch` would stretch each tile to match the tallest item in its row and break the square aspect ratio; flex-start lets each tile size itself normally instead. The mosaic's own `.tile` gets an explicit `height: 165px` instead (matching styles.css:435's original row height), so it has no equivalent stretch problem to guard against. */ .blake-install-scope #categories .mosaic.gbz-mosaic-reflow { display: flex; flex-wrap: wrap; } .blake-install-scope #categories .mosaic.gbz-mosaic-reflow .tile { flex: 1 1 180px; height: 165px; } .blake-install-scope #mega .mega-grid.gbz-mega-reflow { display: flex; flex-wrap: wrap; align-items: flex-start; } .blake-install-scope #mega .mega-grid.gbz-mega-reflow .mega-tile { flex: 1 1 100px; } /* R25 + R26 RETIRED 2026-08-31 (Blake's auth pop-up restore). Both centered text inside elements that linkFixups used to create: R25 for the #btn-signin button→anchor swap (fixSigninButton), R26 for the two injected Create-account links (addCreateAccount). All three fixups are retired — #btn-signin is a real