[Frontend][Inbox] Session-epoch follow-ups: list snapshot path, dropped-read reporting, peekDetail wording, spec mock hygiene

[Chris0Jeky]

Follow-up to PR #2631 (#2571), from the read-only verification pass on its round-2 fixes. None blocks that merge; all are post-logout-only edges of the new session epoch, tracked here for one small follow-up PR.

1. MEDIUM: the map clear inverts the summary-side guard the same way it inverted the detail guard, and the epoch does not cover the list path. applyBackgroundListSnapshot (captureStore.ts ~184-193) keeps a locally newer row only when latestSummaryGenerationById.get(id) ?? 0 exceeds the snapshot's observed generation; after resetForLogout() clears that map the read is 0, so an older in-flight background list snapshot wins. Scenario: a batch triage poll is active; a detail read moves an item to a terminal status via upsertSummary (generation N above the in-flight snapshot's observed generation); the user logs out; between the shell watcher's resetForLogout() and the Inbox unmount's cancelBatchTriagePolling(), the pre-logout list response lands with isCurrent() still true and items regresses the row. The window is sub-millisecond, the route is already leaving, and the next Inbox mount replaces items through the scope-replacement seam, so it is not user-visible today; but the pre-existing spec at captureStore.spec.ts ~2763-2796 asserts the regressed behaviour as intended ("nothing stale-high is left to pin the row"), so the two halves of one guard now carry opposite post-logout contracts. Fix: the list snapshot path captures the session epoch like the detail path and drops a snapshot issued before the reset; retarget that spec.

2. LOW: onRefreshed fires for reads the epoch dropped (captureStore.ts ~651), so refreshedDetailIds in the batch poll can claim an id was reconciled while detailById still holds pre-batch state; post-logout only.

3. LOW: fetchDetail's drop is unreported to callers (three drop paths, no boolean, unlike fetchItems); selectItemById returns true with an empty detail and the Legacy panel shows "Unable to load capture detail." for a read that succeeded; the new path is post-logout only, the other two are pre-existing.

4. LOW: the docstring says every detail read captures the epoch, but peekDetail writes detailById through cacheDetail with no epoch and no generation compare (its only guard is the hash re-check); cacheSummary: false keeps it out of items. Qualify the wording or add the capture.

5. LOW: captureStore.spec.ts ~2560 installs a persistent mockImplementation returning never-resolving promises; the suite's only reset is vi.clearAllMocks(), so a test added below it that triggers an unconfigured detail read hangs instead of receiving undefined. Replace with per-test mockReturnValueOnce or restore the implementation in that test's teardown.

Refs #2571, PR #2631, PR #2567.

[Chris0Jeky]

[Claude lane claim v2]
lane: alpha-product-trust (session taskdeck-alpha-85, Taskdeck-Alpha checkout)
issue: #2640
base: origin/main at claim time (after PR #2631, 6f665bc or later)
residual acceptance: all five items of the issue, one PR: (1) the list snapshot path captures the session epoch like the detail path and drops a background list snapshot issued before resetForLogout(), so the summary-side stale-high guard cannot be inverted by the map clear; the spec at captureStore.spec.ts ~2763-2796 that asserts the regressed post-logout behaviour as intended is retargeted; (2) onRefreshed does not fire for a read the epoch dropped; (3) fetchDetail reports whether it cached (a boolean like fetchItems, or an outcome), and selectItemById surfaces a dropped read honestly instead of returning true with an empty detail (the Legacy panel's "Unable to load capture detail." must not render for a read that succeeded but was dropped; choose the smallest honest surface and state it); (4) the docstring no longer says every detail read captures the epoch, or peekDetail captures it (state which); (5) the persistent never-resolving mockImplementation in the spec is replaced with per-test setup or restored in that test's teardown.
owned paths/modules: frontend/taskdeck-web/src/store/captureStore.ts; src/tests/store/captureStore.spec.ts; src/composables/useInboxOrchestrator.ts and its spec only for item 3's caller-side surface
shared-path leases: PR #2631 (captureStore.ts, AppShell.vue) merged 6f665bc; PR #2635 (AppShell.vue, in the merge gate) does not touch captureStore; no open PR touches captureStore.ts or useInboxOrchestrator.ts (checked gh pr list files at claim time)
parallel-safe work: worktree codex-2214-pin-leg-truth (Review composable and API) and codex-2621 (shell seam, PR in the merge gate) are disjoint
open PRs/branches/review threads checked: both claim forms on #2640 (none); the D-12 Codex-session PRs touch none of the owned paths
human or decision gates: none
status: active (worktree codex-2640-session-epoch-followups, being created)

[Chris0Jeky]

[Claude lane release v2]
lane: alpha-product-trust
issue: #2640
PR: #2659, exact head d230b99, merge commit a1f1d0f, merge commit not squash
disposition: closed by the merge. Shipped: the batch poll's tick captures the session epoch when it issues its request and its currency check carries it, so an older in-flight background list snapshot issued before a logout can no longer win the summary-side stale-high guard after the map clear (the post-logout contract is now the same for both guard halves: a response issued before the reset is dropped, one issued after it caches normally); refreshTerminalDetails fires onRefreshed only when the read cached; fetchDetail reports a DetailCacheOutcome (cached, superseded, generation, epoch) through an additive onCacheOutcome callback (not a return-type change; a Paper row editor awaits it as a CaptureItem), selectItemById re-reads once on a generation drop with no cached detail, keeps the selection on a superseded read, and restores the previous selection and active row only on the epoch case, so every live-session drop either opens or recovers and the restore path is reachable only after a logout; peekDetail documents that it accepts and ignores the callback; the spec's persistent never-resolving mock became per-test setups plus a settles-versus-hangs guard, and the orchestrator spec resets fetchDetail per test.
review: one fresh-context independent reviewer subagent (SHIP; two MEDIUMs taken in round 2: the dropped-read restore branch was reachable in a live batch race and stripped a deep-link hash or turned a click into a silent no-op; the restore set the selected row without the active row) and one read-only verification pass scoped to the round-2 diff (SHIP; a contract-doc MEDIUM and two LOWs recorded below). Evidence: four red-first outputs in round 1 and three in round 2; 5 files / 288 across the store, orchestrator, Inbox view and shell specs; 61 files / 1181 across the Inbox, resilience and property sets; typecheck, ESLint, build, diff check; hosted CI green at the exact head.
NOT verified: Playwright (no E2E crosses a logout or a batch write with an in-flight Inbox read); the Legacy detail panel and list in a real browser (proven at the orchestrator level only); the full frontend suite.
recorded for the next touch of captureStore (no third round): the outcome named superseded is also what a poll-owned read crossed by a logout reports, because the poll's currency check is tested first and it folds six reasons including a moved epoch (no live path reads the distinction today); a second generation collision inside the re-read window returns true with an empty panel and a retained deep link (explicitly chosen); the orchestrator spec block's docstring about clearAllMocks now contradicts the mockReset the same diff added.
CLAUDE_SYNC_PACKET: merged=#2659a1f1d0f closes=#2640 refs=#2571 docs=STATUS thirteenth block (coordinator, in assembly) worktree=codex-2640-session-epoch-followups removed after this comment