feat(ship): observe-a-run V-2 — OutcomeHeader replaces meta-grid run header

[sarath-soman]

Summary

• Productionises the OutcomeHeader prototype from docs/capabilities/observe-a-run/design/prototype/OutcomeHeader.tsx into ship/src/web/lib/OutcomeHeader.tsx. Three render branches (complete / failed / running) with explicit failing-phase prominence and aggregate cost/tokens/duration row.
• Swaps it in for the legacy state-strip + MetaGrid block in RunDetail.tsx (lines 131-194).
• Extends RunDetail.test.tsx with failed-run integration coverage (claims 7+8); guards GlobalRegistrator.register() so happy-dom tests compose in one bun test run.

Closes lambda-brahman/ship#21.

Why manual

Produced by the SWE-pipeline implementer agent (run 2026-05-25-232338-issue-429), which halted at impl on bun scripts/am-i-done.ts Claim 10 due to pre-existing baseline lint/throw violations in unrelated files — not caused by this change. All 9 in-scope claims passed. Committing manually so V-2 ships while the baseline-cleanup tickets proceed in parallel.

Companion tickets that came out of the halt diagnosis:

• chore(ship): disable biome useLiteralKeys + clear lint errors #451 — chore: biome useLiteralKeys conflict cleared (biome rule vs tsconfig strict)
• lambda-brahman/ship#3 — refactor: relocate design prototypes into `src/gallery/` (kills the tsc-prototype-reach noise; pairs with a Designer-character update for future runs)
• lambda-brahman/ship#4 — feat: impl emits Claude session-id so `ship run --resume` continues in-context (turns cheap baseline-induced halts into actual resumes instead of $8 re-runs)

Test plan

• `bun test src/web/lib/OutcomeHeader.test.tsx` — claims 1-5, 9 (three branches mount, complete shape, failed shape with errorSummary, running shape with disabled Rerun, onRerun fires, static no-hex source scan)
• `bun test src/web/pages/RunDetail.test.tsx` — claims 7-8 (failed-run integration; failing-phase prominent)
• `bun tsc --noEmit` — no new errors (pre-existing prototype-reach errors unchanged; lambda-brahman/ship#3 will address them)
• Local dashboard at http://127.0.0.1:8080 — render verification once chore(ship): disable biome useLiteralKeys + clear lint errors #451 merges (so the dev server's biome layer is clean)

🤖 Generated with Claude Code

[sarath-soman]

Status (2026-05-26): merge-conflict with main; deferred pending re-run.

The V-2 implementation (commit 540b696 on this branch) was authored against a main that didn't yet have:

• PR feat(ship): NowBar surfaces in-flight runs with no-progress >5m indicator #448 (V-1 NowBar autonomous impl)
• PR feat(ship): land DS extensions 1-4 — StatePill primitive + 3 tokens #423 (DS extensions — StatePill replaces StateBadge)
• PR fix(ship): guard Happy DOM register in RunDetail.test.tsx #458 (Happy DOM register guard in RunDetail.test.tsx)
• PR docs(capabilities/observe-a-run): lock in derive-at-read for V-1 stuck-suspect #455 (V-1 derive-at-read spec sharpening)

Two files conflict on rebase: ship/src/web/pages/RunDetail.tsx and ship/src/web/pages/RunDetail.test.tsx.

The impl agent halted on over-scoped claim gates (Claims 6 + 10 cover project-wide typecheck + lint, which fail on pre-existing unrelated red — tracked at #451 / lambda-brahman/ship#3). All 8 in-scope claim gates (1–5, 7–9) pass.

Two paths to land V-2

A. Resolve conflicts + merge as-is. Manual rebase of this branch on current main, resolve the two conflicting files (RunDetail's header section + its test fixture), force-push, merge. Faster but commits to V-2's specific impl.

B. Re-run V-2 from scratch via the orchestrator. Close this PR; re-fire ship run 426 (orchestrator skips V-1 automatically per #463); V-2 spawns fresh against current main, no rebase needed. Burns ~10 min + tokens but guarantees compatibility.

Path B is more honest given how much main moved since 540b696. Path A is cheaper if the conflict is small (it is — just RunDetail's render branch).

Operator's call.