Focus slot — fidelity & spacing harness · candidate A, “the dissolve”

Candidate A chosen 21 Sep 2026. This page is not about whether the idea is right — it is about whether it survives the real geometry. Every value is extracted from source: the card DOM from QuoteCard.tsx, its CSS verbatim from six theme files, the focus transform from focus-mode.css, the φ arithmetic from WelcomeHomeView.swift + GoldenSplit, the window sizes from BristlenoseApp.swift, and the timings from WelcomeBaton.swift. The cell renders at true pixel size (1px = 1pt @1x), so what you see is what fits. Drag Window and watch the budget readout go red.

Window 1000 Sidebar Quotes Length Copy Tags Slot Palette Appearance Motion

1 · The live cell — true pixel size, real card CSS, real transform

The φ arithmetic, in full: the spiral is framed width: w, height: w / 1.618 with .padding(20), so w = window − sidebar − 40. GoldenSplit(.horizontal) hands the major cell width: (w − 8) × 0.618 and the spiral's full height w × 0.618 — square to within the gutter's share. At the 700pt minimum window with the sidebar out, that is a 267pt cell, and the furniture (tag, title, the two-line summary, CTA, the 26pt nav row, the drop card, six 8pt gaps) does not shrink with it. The budget readout above is measured from the DOM after layout, not calculated — if it says the slot is negative, the illustration genuinely has nowhere to go at that size.

2 · The three windows — minimum · shipping default · large, side by side

Scaled to fit this page; the scale factor is stated under each so you can read the real size off it. These are the sizes that matter: 700 is the enforced minimum (BristlenoseApp.swift:149), 1000 is what every new window opens at (:267), and 1600 stands for a maximised display. Sidebar out at 220 in all three.

3 · What the harness is for

The card is not what the ideas page drew. The real blockquote.quote-card has a left border only and is rounded on the right only (border-radius: 0 6px 6px 0); the layout is a hanging indent — .quote-row is a baseline-aligned flex with the timecode in a left gutter and .quote-body carrying text, speaker and badges. Timecodes render with square brackets, [11:30]. The earlier draft drew a full box with a meta row underneath, which was an impression of a quote card rather than the product's one.
The dissolve touches two properties, not four. background: transparent and border-left-color mixed to 40% — there is no full border to dissolve. The recede set is exactly .badges, .timecode, .context at 0.14, plus headings at 0.4. .speaker is not in it (axis 1 keeps "whose voice" lit) and neither is .star-btn — there is no Focus rule for the star anywhere in the system, which the illustration inherits for free by having none either.
Timing is two clocks, and mixing them is the easy mistake. Beats stretch by WelcomeTempo — speed 0.6 → stretch 1.667, and a tool webview carries pace 1.3, so ×2.167 — plus a 3s rest on the opening frame and 3s on the finished frame. The 320ms dissolve does not stretch: micro-transitions stay at authored speed by stated policy, because slowing a fade reads as lag rather than calm. So the loop is slow and the transition is not.
At the minimum window the slot is GONE, and that is the headline finding. Derived from source, sidebar out at 220: a 700pt window gives a 272pt cell, and the furniture — tag 14 + title 20 + a two-line summary 32 + CTA 15 + nav row 26 + drop card 78 + six 8pt gaps 48 + 32 cell padding — is ~265pt. That leaves about 7pt. Not "cramped": arithmetically nothing. At the 1000pt shipping default the cell is 457pt and the slot gets ~192pt, which is why the pool’s caps cluster at 160–196 — they are tuned to the default window and are already at its limit. At 1600pt it is ~563pt and everything is easy. So the design target is the default window, and the minimum window is a degradation case every illustration already handles by self-scaling — which is exactly why BookShelfView’s fan reaches 0.29 there. Collapse the sidebar and 700pt becomes a 408pt cell, which is survivable; it is the sidebar-out minimum that is brutal.
The first-draft copy was the longest line in the pool, and copy length is a spacing decision here. Measured against the eight shipped slots: Miro 28 chars, clips 38, codebooks 55, starHide 65, autocode 66, agent 69, tag 84, ingest 96 — and the 21 Sep draft of this line, 110. At the 700pt window’s ~36 characters per line that is four lines against the pool’s typical two, spending ~26pt of a budget that is already at zero. The default here is now the cut at 83 chars, written inverted-pyramid with an em-dash hinge so WelcomeClauseFit can drop the trailing clause and leave a grammatical 33-char core — the rewrite discipline the 20 Aug pass adopted across all 45 slots. Switch Copy to compare the three.
The spacing question has a measurable answer, and it is not flattering at the minimum window. The furniture is a near-constant ~210–230pt while the cell is square and scales linearly, so the slot's share collapses as the window narrows. Switch Slot between cap 176 and fill to see the two mechanisms: every shipped illustration takes a cap and self-scales under it, while fill is the full-bleed proposal that measures its box and stops when the next card will not fit. Only the second answers "how many quotes" honestly at every size.

Ideas and the rejected candidates: welcome-focus-cell.html. The pool's reference spec: welcome-studytools-animations.html. Nothing here is built — WelcomeIllustration has no .focus case.