Design: Compass Bridge re-clothe (SEA-2111)

Status: Draft Owner lane: compass-ux (design) → compass-ui (execution) Refs: SEA-2111 (live board doesn’t match the rigel.build reference); sequenced after the SEA-2034 DS-token cutover (merged, main 18e988b5). Two adjacent concerns are split into their own lanes/PRs, not folded here: the state-dot pixel-art glyph adoption → SEA-2118 (mechanical frozen-spec adoption, global across surfaces); review/CI badge semantic clarity → SEA-2117 (its own design pass — see T4). This record is the board-structure re-clothe and ships on the current state-dots. Governing spec: apps/ui/src/design/surfaces.md §“Bridge — the Issues and PRs board” (L196–268, frozen T6/SEA-1816) Reference render: apps/rigel.build/src/components/BridgeBoard.astro (brand repo, read-only)

Problem / Intent

The live Bridge board (apps/ui/src/components/Bridge.tsx) renders on the merged --cx-* tier but still wears its pre-DS clothing: GitHub-board-style bordered cells, rounded priority-striped cards, a plain 13px toolbar heading, and a PRs tab that is a grouped list rather than the board the frozen spec describes. The rigel.build reference render (BridgeBoard.astro) is the excellence bar the surfaces spec names as “the starting point … not a redraw” (surfaces.md:202-203). This record lifts the reference’s visual structure onto the live component — real data, real interaction, only the clothing changes.

Global Constraints

Every task below inherits these; none restates them.

• Stack: SolidJS ^1.9.13 + Vite + TypeScript; pure CSS custom properties. No Tailwind, no component library, no CSS-in-JS.
• Tokens: consume the --cx-* semantic tier ONLY. No raw hex, no --rigel-* refs, no literal durations/easings outside design/tokens.css — the D7 stylelint guard bans all four at error (apps/ui/.stylelintrc.cjs:26-61: color-no-hex, hex-anywhere regex, /var\(\s*--rigel-/, legacy-var vocabulary, raw motion literals on transition|animation). Reference --rigel-* values translate through their existing --cx-* aliases (tokens.css:111 --cx-accent: var(--rigel-blue), tokens.css:139-143 --cx-issue-*, tokens.css:146-151 --cx-ci-*/--cx-review-*).
• tokens.css is read-only for this lane. No new token is coined here; a need the tier can’t answer is an Open Question, never a unilateral mint.
• Motion axis is D9/foundation-T8’s (apps/ui/src/design/motion.md). This record consumes existing motion tokens (--cx-pulse-period, --cx-ease-out, --cx-motion-fast/base) and never coins a duration or easing. The advancing-card affordance the reference shows is not in motion.md’s primitive list — it ships as a dormant hook (T5, Decision D2), not a coined token.
• State-dots and badge-clarity are OUT of this lane. The state-dot pixel-art glyph adoption is SEA-2118 (its own global PR — Matt ruled the glyphs ship at 9px; grounded blast radius is 5 files + the .r-tab overlay, too wide to fold into a board-scoped PR). The review/CI badge semantic-clarity redesign is SEA-2117 (its own design pass); T4 here recolors the existing badges onto the semantic tier and consumes whatever badge FORM that design freezes. The board keeps today’s StateDot unchanged; no .state-dot rule is touched here.
• Not a redraw. surfaces.md:202-203: “The starting point is BridgeBoard.astro (authored against the real Bridge.tsx / IssueCard.tsx / board.ts), not a redraw.” The live component’s data wiring (board.ts / board-render.ts accessors), selection model (store.selectIssue / store.openAgent), click/dblclick/chip a11y compromises (IssueCard.tsx:56-59, Bridge.tsx:69-71), and the Swimlanes/Status grouping toggle (Bridge.tsx:98, 160-175) all survive unchanged.
• No fabricated data. Where the Astro mock shows a fact the live store cannot source (the advancing flag, BridgeBoard.astro:39,70), the gap is handled as a dormant hook — never a stubbed fake field presented as real.
• Sticky behavior is load-bearing. The live board scrolls (app.css:510-512 .swimlane { flex: 1; overflow: auto; … }) with sticky column heads (app.css:536-539) and sticky lane gutters (app.css:566-569). The reference does not scroll (it caps at max-width: 1040px, BridgeBoard.astro:434-437, with a mobile overflow-x: auto fallback only, :612-618). Every lifted visual must keep the sticky/scroll model working.
• Ledger coupling: this is a product design record; its freeze requires a same-PR docs/designs/product/DECISIONS.md delta or a Ledger-impact: PR line. The driver handles this at PR time. Candidate ledger rows are flagged in §Ledger below.
• Verification vehicle: the Playwright visual-smoke harness from the cutover (apps/ui/e2e/visual-smoke.spec.ts, apps/ui/playwright.config.ts) is extended, not reinvented. Every visual slice re-runs it and the driver attaches before/after PNGs for Matt.
• Format + lint + tests (biome, moon run compass-ui:stylelint, vitest, test:visual) run per slice before it is called done.

Decisions (Matt-ruled 2026-08-16, via ask)

These closed the load-bearing forks the draft surfaced; the plan below is written against them.

• D1 — PRs view is board-ified (was OQ-1). The PRs tab becomes the same swimlane grid on PR-lifecycle columns, per surfaces.md:206-210 and the reference. Two sub-rulings: (a) unresolved review threads gate a PR to “In review” — an approved + CI-green PR with open threads is not “Ready to merge”; (b) the “Merged” column shows PRs merged within a 1-day window, then they drop — and that windowing is handled server-side / in the store lane, not in this UI. The UI renders whatever merged rows the store hands it; T6 does not implement retention.
• D2 — advancing card ships as a dormant CSS hook (was OQ-2). Existing motion tokens only; wired to no data until a real source lands. The motion-spec addition routes to D9/foundation-T8, the data field to the store lane.
• D3 — CI/review badges: recolor onto the semantic tier; FORM per SEA-2117. T4 re-points the badges from the raw --cx-ok/error/warn they use today onto the dedicated --cx-ci-*/--cx-review-* tier, and adopts whatever badge form the SEA-2117 badge-clarity design freezes (pixel-art 1-bit glyphs leading). The .cx-badge naming in surfaces.md is the token/vocabulary contract, not a labelled-DOM-box mandate. The live commented verdict maps to --cx-review-pending. If SEA-2117 has not frozen by the time T4 runs, T4 ships the interim compact recolored pip and the glyph form lands as a fast follow — T4 does not block on it.
• D4 — roving-tabindex 2-D keyboard grid is OUT of scope (was OQ-5): filed as a follow-up interaction issue at dispatch; this record changes only the clothing, not the keyboard model.
• D5 — state-dot glyph adoption is a separate lane (SEA-2118, see Global Constraints); the board ships on the current StateDot.

Approach

One PR train of six right-sized slices over Bridge.tsx, IssueCard.tsx, and the .bridge*/.swim*/.card*/.pr-* blocks of app.css, in the order: harness extension → grid shell → cards → badges → advancing hook → PRs board. Each slice deletes the legacy selectors it replaces in the same diff (surfaces.md:267 flip item 8).

The reference is lifted at three levels:

1. Structure — the hairline grid. The reference draws its grid as a 1px gap over a --rigel-night-2 container background with cells on --rigel-night (BridgeBoard.astro:434-450), which reads as hairlines. That exact technique is not sticky-safe: the live board scrolls under sticky heads, and grid gaps are windows to the scrolled content behind them. The re-clothe therefore keeps the live border technique (app.css:609-611) but tunes it to the hairline look — uniform 1px --cx-border cell borders, cells on --cx-bg, lane gutters on --cx-bg-panel (the reference’s .bridge-lane { background: var(--rigel-panel) }, BridgeBoard.astro:459-465), and the board frame on a single outer --cx-border border. Same optics, sticky-safe mechanics.
2. Composition — the frozen surfaces.md §Composition contract (surfaces.md:224-234): cells are .cx-card on the panel tier; selection is --cx-bg-selected plus the accent left rule, never a raised background; the column-head tint consumes --cx-issue-* at low alpha in the lane head only; badges live on --cx-ci-*/--cx-review-*; the issue cross-link is a .cx-chip. The DS component CSS exists but is imported nowhere (App.tsx:4-6 imports only tokens.css, base.css, app.css) — the card/badge component files get wired in as part of their adoption slices.
3. Type hierarchy — the reference quiets the issue key (.card-issue { color: var(--rigel-mute); letter-spacing: 0.5px }, BridgeBoard.astro:515-520) and brightens the title (.card-title { color: var(--rigel-fog) }, :556-561); the live card inverts that (key on --cx-accent 600-weight, app.css:678-683). The toolbar heading goes display-face (--cx-font-display at --cx-display-sm, tokens.css:196,202 — the reference’s font-family: var(--rigel-display); font-size: 22px, BridgeBoard.astro:379-384) from the live 13px/600 (app.css:477-480). Column heads go uppercase letterspaced mono per the reference (BridgeBoard.astro:451-458) — the live heads are already close (app.css:536-552) and keep their live-data count badge.

Where the DS and the reference disagree on primitive shape (the DS .cx-card carries --cx-radius-md and motion-token transitions, design/components/card.css:11,17-19; the reference card is square, BridgeBoard.astro:501-508), the DS wins on shape — the reference guides structure, spacing, and hierarchy; frozen DS components govern primitive shape. Where no DS component governs the element (the CI/review pips, the seg control), the reference’s shape wins (OQ-1, generalizing the draft’s pip-only carve-out).

The PRs view is the one structural change (D1): surfaces.md:206-210 freezes “both are the same swimlane grid, only the columns differ — … PRs use the PR-lifecycle columns (In progress, In review, Ready to merge, Merged)”, and the reference renders exactly that (BridgeBoard.astro:120-129,309-359). The live PRs tab is a grouped flat list (Bridge.tsx:274-320, .pr-tab/.pr-group/ .pr-row, app.css:783-861). The board-ification is the last slice, driven by a new pure derivation over the real PullRequest shape — no fabricated column field.

Alternatives considered

• Port the reference’s gap-grid hairlines verbatim. Rejected: not sticky-safe (above). Border-tuning reproduces the optics without breaking the scroll model.
• Restyle the PRs list in place instead of board-ifying. Rejected by D1: it satisfies “PRs view rows carry the PR’s own facts” (surfaces.md:219-220) but contradicts the same section’s “both are the same swimlane grid” (surfaces.md:206-207) and the reference render. Matt ruled board-ify.

Gap analysis

Reference (BridgeBoard.astro, cited astro:) vs live, each row mapped to the live file+line it changes and the token it lands on. Live line numbers at main 18e988b5. (The former G9 state-dot row is removed — it is its own lane, SEA-2118.)

#	Reference has	Live has	Live change site	Token(s)
G1	Hairline 1px grid: container bg --rigel-night-2, gap: 1px, cells --rigel-night (astro:434-450)	Per-cell 1px borders bottom/right on --cx-border (app.css:609-611), corner/head borders (app.css:524-525,541-542)	.swim-cell, .swim-corner, .swim-colhead, .swim-gutter border/bg rules; add outer board frame	--cx-border, --cx-bg, --cx-bg-panel
G2	Lane gutter on the panel tier, padding: 14px 12px; gap: 9px (astro:459-465)	Gutter on --cx-bg-raised, padding: 10px 12px; gap: 8px (app.css:566-578)	.swim-gutter (rename .bridge-lane per surfaces.md:211)	--cx-bg-panel
G3	Col heads: 11px mono, letter-spacing: 1px, uppercase, padding: 12px 14px (astro:451-458)	10px/600, letter-spacing: 0.05em, padding: 8px 12px, colored lane-dot + count (app.css:536-563; dot fed lane.color from constants.ts:17-31)	.swim-colhead (rename .bridge-col-head); drop the lane-dot, keep the live count	column tint: color-mix(in srgb, var(--lane-tint) 8%, var(--cx-bg-raised)) where --lane-tint = the lane’s --cx-issue-* (OQ-2 alpha)
G4	Column-head tint per surfaces.md:213-215 (“tinted by the issue-state color at low alpha in the lane head only — contrast rationing”) — NOT actually present in the reference CSS (astro:451-458 is untinted --rigel-haze)	No tint	.swim-colhead background; Bridge.tsx:190-194 passes --lane-tint inline style	--cx-issue-queued/blocked/in_progress/in_review/done (tokens.css:139-143)
G5	Flat square card: bg --rigel-raised, 1px border, padding: 9px 10px, no radius, no left stripe (astro:501-508)	.card: --cx-bg-panel, --cx-radius-sm, 3px priority left border (app.css:626-643,659-670), selected = accent ring (app.css:654-657)	IssueCard.tsx:46 container class → .cx-card + data-selected; delete app.css .card base/selected/priority rules; add the board-scoped internal-layout block (F3)	.cx-card contract: --cx-bg-panel, --cx-bg-selected + accent left rule (design/components/card.css:5-37)
G6	Quiet key / bright title: .card-issue on mute + 0.5px tracking (astro:515-520), .card-title on fog (astro:556-561)	Key on --cx-accent 11px/600 (app.css:678-683); title default text (app.css:695-698)	.card-issue, .card-title rules in app.css (propagates to DoneView’s shared sub-parts, F6)	--cx-text-faint (key), --cx-text (title)
G7	Display-face board heading, 22px (astro:379-384)	13px/600 UI face (app.css:477-480)	.bridge-toolbar .heading	--cx-font-display, --cx-display-sm (tokens.css:196,202)
G8	Square seg control, mono labels, bordered (astro:385-403)	.seg on --cx-bg-panel + active --cx-bg-active, --cx-radius-sm (app.css:487-507)	.seg, .seg button — align metrics + square off (no DS component governs the seg, OQ-1); keep Solid onClick mechanism (radios are the marketing no-JS hack, astro:225-229, not ported)	--cx-border, --cx-bg-raised
G9	7px square CI/review pips on state colors (astro:526-550)	8px border-radius: 2px CI square + circle review dot on --cx-ok/error/warn (app.css:746-773)	.ci-badge/.review-badge rules; consumers IssueCard.tsx:79,83, Bridge.tsx:61,64, DoneView.tsx:37,41	--cx-ci-pass/fail/pending, --cx-review-approved/changes/pending (tokens.css:146-151); compact-pip shape per D3
G10	One advancing card: blue border + chase-light sweep, 1.8s, reduced-motion display: none (astro:586-610,620-631)	Nothing	New .cx-card[data-advancing="1"] rule + keyframe (consumes existing tokens only); no live data source for the flag	--cx-accent (blue flow per surfaces.md:232-234), --cx-pulse-period, --cx-ease-out — D2
G11	PRs view is a board: same lanes, PR-lifecycle columns, PR cards with coord-as-key + ci/review + resolved/total threads foot (astro:120-129,309-359)	Grouped flat list: .pr-tab > .pr-group > .pr-row (Bridge.tsx:274-320, app.css:783-861)	Bridge.tsx PRs branch; new prLifecycle/prBoardRows in board-render.ts/board.ts; PR_LANES in constants.ts; delete .pr-row* selectors	--cx-issue-* reused for column tint; card tokens as G5 — D1
G12	Empty cell renders empty (astro:490-496 min-height: 56px, no placeholder)	Status-mode empty cell renders a "—" placeholder (Bridge.tsx:207 .term-empty fallback); swimlane-mode empties are dimmed (.swim-cell.dim, app.css:620-623, applied Bridge.tsx:252-255)	Remove the Bridge.tsx:207 fallback; min-height on .bridge-cell; keep .dim (a live-data affordance the reference lacks, consistent with the KEPT list — F11)	—
G13	Cell padding: 10px; gap: 8px; min-height: 56px (astro:490-496); gutter col 168px, cols minmax(0,1fr) (astro:260)	Cell padding: 7px; gap: 7px, no min-height (app.css:609-618); gutter 180px, cols minmax(210px,1fr) (Bridge.tsx:131-134)	.bridge-cell metrics; keep live minmax(210px,1fr) (the board scrolls; minmax(0,1fr) is the reference’s fit-to-1040px constraint, not ours)	--cx-space-2/3

Live-only features the reference lacks — all KEPT (real data / real affordances): the N agents · M in-flight toolbar sub (Bridge.tsx:140-142), the Swimlanes/Status grouping toggle (Bridge.tsx:160-175), the gutter’s N items meta + → open affordance (Bridge.tsx:236-242), the lane-head counts (Bridge.tsx:193), PR-chip and issue-chip cross-links, selection sync, the current StateDot (untouched here — its glyph adoption is SEA-2118).

Plan

Slices are dependency-ordered; T2–T6 are each independently reviewable and re-run the visual harness. Class renames land with the slice that owns the element (.swim-* → the surfaces.md:211-216 .bridge-lane / .bridge-col-head / .bridge-cell vocabulary), updating Bridge.test.tsx and visual-smoke.spec.ts selectors in the same diff.

T1 — Harness extension (first; the acceptance vehicle)

Extend the visual-smoke spec with the shots this record’s review needs: the PRs tab (click the PRs · seg button, wait for the PRs surface), a cropped column-head strip (tint review), and a cropped single-card close-up. Baseline run lands the “before” set.

Interfaces: consumes apps/ui/e2e/visual-smoke.spec.ts:11-20 conventions (SCREENS = "e2e/__screens__", selector-waits, no sleeps); produces three new test() blocks writing bridge-prs.png, bridge-colheads.png, bridge-card.png. No production code.

T2 — Grid shell re-clothe (G1–G4, G7, G8, G12, G13)

The board frame, hairline borders, panel-tier gutters, tinted column heads, display heading, seg metrics (squared off), empty-cell fix, cell metrics. Renames .swim-* selectors/classes to .bridge-lane/.bridge-col-head/ .bridge-cell/.bridge-corner in Bridge.tsx + app.css. The .pr-group-head head reuses .swim-gutter (Bridge.tsx:291, styled app.css:797-805) — its class reference updates in THIS slice’s rename, so the grouped PRs list stays coherent until T6 replaces it (F10). Column tint: Bridge.tsx sets style={{ "--lane-tint": lane.color }} on each head (lane colors are already --cx-issue-* var strings, constants.ts:18-30); CSS applies the 8%-mix background (OQ-2). Drops the lane-dot, keeps lane-count.

Interfaces: consumes BOARD_LANES: Lane[] (constants.ts:17-31, Lane = { state: IssueState; label: string; color: string }); produces the renamed markup in Bridge.tsx:179-271 and replacement app.css rules; updates visual-smoke.spec.ts waits if selectors changed (.bridge root kept). Test cycle: vitest Bridge.test.tsx green after rename; harness re-run.

T3 — Card re-clothe: .cx-card adoption (G5, G6)

Wire design/components/card.css into the cascade (import in App.tsx after base.css, before app.css, preserving the load-bearing order the cutover froze), flip the IssueCard container from class="card" + classList={{ selected }} (IssueCard.tsx:46-48) to class="cx-card" + data-selected, drop the priority left-stripe (selection owns the left rule per card.css:33-37), and restyle the card-issue/card-title sub-part rules to the quiet-key/bright-title hierarchy. Delete the superseded .card base/hover/selected/priority rules from app.css (the grandfathered raw transitions at app.css:637-642 die with them — the cx-card transition is already tokenized, card.css:17-19).

Card internal layout (F3): the bare .cx-card is display: block with --cx-space-3 (12px) padding (card.css:5-8), while the live .card is a display: flex; flex-direction: column; gap: 6px; text-align: left button (app.css:626-630) and the reference card is 9px 10px padding (astro:503-508). So T3 adds a board-scoped layout block — .bridge-cell > .cx-card { display: flex; flex-direction: column; gap: 6px; text-align: left; padding: 9px 10px } — overriding the DS card’s block/12px default for the dense board only. This is a spacing/structure override (the reference governs spacing), scoped so it never leaks to other .cx-card surfaces.

Cross-surface note (F6): DoneView.tsx:31 renders the shared .card-issue/.card-pr/.card-diff sub-parts, so the quiet-key restyle intentionally propagates to the Done view (one convention, not two). The existing done.png harness shot covers it — the T3 diff notes the propagation so it isn’t read as an accidental out-of-scope change in Matt’s screenshots.

Interfaces: consumes IssueCard props ({ issue: Issue; onOpenPr?: () => void }, IssueCard.tsx:18-21) unchanged; produces the class flip + app.css sub-part restyle + board-scoped layout block + deletions. Test cycle: vitest; harness card close-up.

T4 — Badge recolor + form onto --cx-ci-* / --cx-review-* (G9; D3)

Re-point every .ci-badge[data-status] / .review-badge[data-verdict] rule (app.css:746-773) from the generic --cx-ok/error/warn onto the dedicated --cx-ci-pass/fail/pending and --cx-review-approved/changes/pending namespaces (tokens.css:146-151), and map the live commented verdict (board-render.ts:102-112 returns "changes" | "approved" | "commented") onto the --cx-review-pending color (D3). All three consumer sites (Bridge.tsx:61,64, IssueCard.tsx:79,83, DoneView.tsx:37,41) share the rules, so one edit covers them.

Badge FORM per SEA-2117 (D3). The badge-clarity design (SEA-2117, pixel-art 1-bit glyphs leading) owns what the badge looks like; T4 consumes that frozen form. If SEA-2117 is frozen when T4 runs, T4 renders the chosen glyph/pip form (a BadgeGlyph-style inline SVG mirrors the state-dot emission contract and the consumer sites flip from bare <span> to the component). If SEA-2117 has NOT frozen, T4 ships the interim compact square pip recolored onto the tier (7px, no radius — astro:526-532) and the glyph form lands as a fast follow — T4 does not block on SEA-2117.

Interfaces: consumes ciBadge(pr): "pending" | "success" | "failure" | undefined (board-render.ts:119-121) and reviewBadge(pr): "changes" | "approved" | "commented" | undefined (board-render.ts:102-112) unchanged; produces the recolor + the SEA-2117 form (or interim pip). Test cycle: Bridge.test.tsx:145-148 badge-presence assertions stay green; stylelint; harness.

T5 — Advancing-card hook (G10; D2, dormant until data lands)

Add the .cx-card[data-advancing="1"] rule: border-color: var(--cx-accent) plus a sweeping ::after chase-light gradient (color-mix(in srgb, var(--cx-accent) 32%, transparent) — the reference’s 32% mix, astro:598-603), animated at var(--cx-pulse-period) var(--cx-ease-out) infinite (existing tokens only — no coined literal). Fidelity/guard notes (F12): --cx-pulse-period (tokens.css:224) resolves to 1.6s via --rigel-pulse-period (tokens.css:39) where the reference hard-codes a 1.8s sweep (astro:606) — we accept the token value rather than coin the literal. The period token is zeroed under reduced motion (tokens.css:241-248), which would leave a 0s-infinite animation running; the explicit display: none reduced-motion guard (mirroring astro:620-631) is therefore load-bearing, not redundant — a reviewer must not strip it. The attribute has NO live data source (Issue carries no transition timestamp — stub-data.ts has no updatedAt/ advancing), so it ships dormant: IssueCard renders data-advancing only when a future store accessor provides it. Verified in the harness by toggling the attribute via page.evaluate before the shot.

Interfaces: produces the CSS rule + keyframe and an optional advancing?: boolean-style prop seam on IssueCard left unwired; consumes --cx-accent, --cx-pulse-period, --cx-ease-out. Test cycle: stylelint (proves no raw motion literal); harness advancing shot via attribute toggle.

T6 — PRs view board-ification (G11; D1, last)

Replace the grouped list with the same swimlane grid over PR-lifecycle columns, per surfaces.md:206-210 and the reference (astro:309-359). New pure derivations (red-green: unit tests first per house rule):

constants.ts

export type PrLifecycle = "in_progress" | "in_review" | "ready" | "merged";
export const PR_LANES: { state: PrLifecycle; label: string; color: string }[];
// labels: "In progress" | "In review" | "Ready to merge" | "Merged"
// colors: --cx-issue-in_progress / --cx-issue-in_review / --cx-accent (OQ-3) / --cx-issue-done

// board-render.ts — defined in terms of the EXISTING roll-up seams (F2),
// NOT raw pr.reviews, so it inherits the latest-per-author +
// changes_requested>approved>commented precedence:
export function prLifecycle(pr: PullRequest): PrLifecycle;
//   pr.forgeState === "merged"                                  → "merged"
//   reviewBadge(pr) === "approved" && ciBadge(pr) === "success"
//     && every thread resolved (D1a)                            → "ready"
//   reviewBadge(pr) !== undefined || pr has open threads        → "in_review"
//   else (incl. draft-open)                                     → "in_progress"
// Totality (F2): input is narrowed to board rows (forgeState !== "closed"),
// so "closed" is unreachable; a defensive `default → "in_progress"` keeps it
// total if the type ever widens.

// board.ts
export function prBoardRows(all: readonly Issue[]): PrRow[];
// like prRows (board.ts:132-134) but forgeState !== "closed" — merged PRs
// (within the store's 1-day window, D1b) included so the Merged column is
// sourceable. The 1-day retention is SERVER-SIDE / store-lane (D1b); this
// function renders whatever merged rows the store hands it.
export function prBoardGroups(
  agents: readonly Agent[], all: readonly Issue[],
): { agent: Agent | null; rows: PrRow[] }[];
// grouping/order contract identical to prRowGroups (board.ts:139-152),
// INCLUDING the agent:null "Unassigned" group LAST iff it has rows.

Unassigned lane gutter (F8): prBoardGroups keeps prRowGroups’s trailing agent: null group (board.ts:149-150), so the swimlane grid gets an Unassigned lane — but a null agent has no StateDot/handle/count/→ open affordance (the gutter anatomy at Bridge.tsx:225-242). T6 designs that gutter explicitly: no StateDot, a faint “Unassigned” label, no open affordance — mirroring the current list’s distinct .pr-group-head.unassigned head (Bridge.tsx:284-287) that this slice deletes.

The PR card reuses the T3 card anatomy with the PR’s own facts (coordinate-as-key via the existing pr-row-coord derivation Bridge.tsx:55-58, ci/review pips, title, foot = @assignee + resolved/total threads from pr().threads as Bridge.tsx:44 counts them) and keeps the issue-key .cx-chip cross-link flipping to Issues (Bridge.tsx:72-88 semantics). prCount tab badge stays open-only (board.ts:158-167) — consistent with the reference, which excludes the Merged column from its count (astro:218-221). F7: prCount’s doc comment (board.ts:154-155 “equals the visible row count”) becomes false once the tab shows merged rows the count excludes — the comment updates to “equals the non-Merged visible rows” in this diff, and Bridge.test.tsx:57-64’s framing follows. Delete PrRowItem and the .pr-tab/.pr-group/.pr-row* selector family (app.css:783-861) in the same diff.

Interfaces: consumes PrRow = { issue: Issue; pr: PullRequest } (board.ts:128), prBadge/ciBadge/reviewBadge/issueKey (board-render.ts); produces the three functions above + PR_LANES + the board markup + deletions + unit tests for prLifecycle/prBoardRows (precedence: merged > ready > in_review > in_progress; thread-gating per D1a). Test cycle: new vitest units red→green; Bridge.test.tsx PR-tab assertions updated; harness bridge-prs.png — and the stub board set (F2) carries at least one merged PR whose parent issue is in store.issues(), so the Merged column is non-empty in that shot and the D1 board-ification is visually verified rather than vacuously green (G11).

Out of scope

• Roving-tabindex 2-D grid (surfaces.md:236-239,264-265 flip item 6) — a keyboard/interaction change, not clothing; filed as a follow-up issue at dispatch time (D4).
• State-dot 1-bit glyph adoption — its own lane, SEA-2118 (D5); the board ships on the current StateDot.
• Review/CI badge form redesign — its own design lane, SEA-2117 (D3); T4 recolors onto the tier and consumes the badge form SEA-2117 freezes.
• Motion primitive spec authorship — D9/foundation-T8 owns adding a card-advance topology to motion.md; T5 here only consumes tokens (D2).
• Merged-column retention windowing — server-side / store lane (D1b); the UI renders the merged rows it is handed.
• The board-empty centered message (surfaces.md:242-243) — the stub store always has issues; deferred with the roving-tabindex follow-up.

Ledger

Candidate DECISIONS.md rows for the driver at PR time (this record does not edit the ledger): (1) PRs view board-ification supersedes the grouped-list PRs tab, with thread-gating and a server-side 1-day Merged window; (2) board-card priority left-stripe retired in favor of the cx-card selection left rule; (3) advancing-card affordance ships as a dormant hook pending D9 motion-spec coverage and a real data source.

Tasks

• T1 — extend visual-smoke.spec.ts: bridge-prs.png, bridge-colheads.png, bridge-card.png; capture baseline set
• T2 — grid shell: hairline borders, panel gutters, --lane-tint column tint, display heading, seg metrics (squared), .swim-*→.bridge-* rename (incl. the .pr-group-head head reference, F10), empty-cell fix; tests + harness green
• T3 — wire card.css; IssueCard → .cx-card[data-selected]; drop priority stripe; quiet-key/bright-title; board-scoped card layout block (F3); note DoneView propagation (F6); delete legacy .card rules
• T4 — pips → 7px squares on --cx-ci-*/--cx-review-*; commented → review-pending color; stylelint green
• T5 — data-advancing chase-light rule on existing motion tokens + load-bearing reduced-motion guard (F12); harness shot via attribute toggle (D2)
• T6 — PRs board: prLifecycle/prBoardRows/prBoardGroups/PR_LANES (unit tests first; thread-gating D1a; totality F2), unassigned-lane gutter (F8), PR-card markup, prCount comment update (F7), delete PrRowItem + .pr-* CSS (D1)
• Driver: ledger delta / Ledger-impact: line in the PR body; attach before/after PNGs; file the roving-tabindex + empty-board follow-up issue at dispatch

Open Questions

Remaining forks are non-load-bearing detail (the load-bearing ones closed as D1–D5 above); each carries a stated assumption the design proceeds on.

OQ-1 (non-load-bearing) — primitive-shape fidelity where DS ≠ reference. Reference cards/pips/seg are hard-square; the DS .cx-card carries --cx-radius-md (card.css:11) and the live pips/seg carry small radii. Assumption: DS components govern shape; the reference governs structure, spacing, and hierarchy — except where no DS component governs the element (the CI/review pips and the seg control), where the reference’s square shape wins. Reviewable in the screenshot pass.

OQ-2 (non-load-bearing) — column-head tint alpha. surfaces.md:213-215 specs “low alpha” without a number, and the reference CSS carries no tint at all (astro:451-458). Assumption: color-mix(in srgb, var(--lane-tint) 8%, var(--cx-bg-raised)). Reviewed in bridge-colheads.png; a one-line knob.

OQ-3 (non-load-bearing) — “Ready to merge” column color. --cx-issue-* has five issue-axis values and no “ready” (tokens.css:139-143). Assumption: --cx-accent (the blue flow color — ready-to-merge is work in motion, not a lifecycle state). No token minted.