Closes #170

[realproject7]

Closes #170

Pair each dashboard History row with its own session's index entry by identity, instead of by a positional reversal that swapped the duration/languages/cost subtitle between two sessions saved in the same clock-minute.

EPIC Alignment

Part of EPIC #134 (v1.3 audit follow-up), Batch 40 (re-pointed from #173 to #170 per operator #939 to restore the standing-order sequence #172 → #171 → #170 → #173). Dashboard/archive cluster: src/dashboard.ts (consumer) + packages/archive/src/dashboard.ts (the aggregate/index contract). Honors the Batch 36 amendment (SessionIndexEntry has no name; aggregateSessions takes only ParsedSession[], so identity is threaded by constructing name-paired entries before the positional step). Does not touch resample.rs/render.ts/parse.ts/pipeline.rs (reserved for #178/#179) or settings.

Self-Verification

The bug. renderHistory() paired each newest-first session (m.sessions) with the oldest-first chronological index (m.stats.index) via index[index.length - 1 - i], assuming the second array is the exact reverse of the first. Array.prototype.sort is stable, so when two sessions tie on (meta.headerDate, meta.startClock) — two back-to-back captures in the same clock-minute, or the <name> (2).md filename-collision the archive tests anticipate — the tie sub-group keeps the same relative order in both arrays rather than being mirrored. The blind full-array reversal then swapped the tied sessions' entries: a correctly-titled, correctly-linked row displayed the OTHER session's duration/date-time/languages/cost subtitle (via rowMeta()).

The fix (pair by identity, per the amendment).

• Extracted a pure toSessionIndexEntry(session) in @livecap/archive (packages/archive/src/dashboard.ts:83-96), derived solely from the session — aggregateSessions now calls it (identical output, so stats.index and its tests are unchanged).
• IndexedSession gains an entry field (src/dashboard.ts:60-64); buildDashboardModel builds it from each session alongside its file name (:96-101), so every history row carries the entry derived from its OWN session.
• renderHistory now uses indexed.entry (src/dashboard.ts:452) — pairing by identity, never by position. The positional reversal and its assumption comment are removed.

This is the amendment's accepted approach — "construct name-paired entries before aggregation, then key the history mapping by name" — realized as each IndexedSession owning its entry (identity ≡ file name).

Note on stats.index. aggregateSessions still produces the chronological index (a public @livecap/archive type, covered by its own tests); the app's history no longer consumes it, but it remains a valid part of the archive contract, so it stays. No duplication — both paths build entries through the one shared toSessionIndexEntry.

Security invariants. No new dependencies. No caption/audio content logged — toSessionIndexEntry carries only meta numbers/langs/title (never transcript text; same fields the index already exposed). Simplest fix that solves the ticket; no drive-by refactors.

Tests / evidence.

• test/dashboard.test.ts — new regression pairs each session with its OWN index entry when two tie on date+clock (#170): two sessions tied on date+clock with distinct duration (45 vs 10)/languages (EN→KO vs JA→FR)/cost, asserts each row's entry matches its own session, plus an entry↔session invariant over all rows.
• Negative control (scratch, not committed) confirmed the old positional pairing swapped them: [{title:"Alpha",oldDur:10,ownDur:45},{title:"Beta",oldDur:45,ownDur:10}] — proving the regression is not vacuous.
• Existing dashboard tests green: 18 app (test/dashboard.test.ts) + 8 archive (packages/archive/test/dashboard.test.ts, incl. the aggregateSessions index-row assertions — unchanged).
• Full app suite 154 tests pass; archive package suite green; pnpm typecheck (app + packages), pnpm lint, pnpm build all clean.

Design Fidelity

No visual change — the fix corrects WHICH session's data a row shows, not how it renders. The historyRow markup, classes, and rowMeta() subtitle format are untouched; only the (now correct) entry feeds them.

Element	Design (unchanged)	Source pointer	Markup/token/layout change?
History row	title span + meta subtitle + optional snippet; click → detail	src/dashboard.ts historyRow (:548-576); .dash-row/.dash-row-title/.dash-row-meta in styles.css	None — same nodes/classes
Row subtitle (meta)	rowMeta(entry) — date · time · duration · langs · cost	src/dashboard.ts rowMeta (:228)	None — same formatter; now fed the row's OWN entry
Row title / click target	bound to session/name (was already correct)	historyRow openDetail(name, session)	None

The only user-visible effect: two sessions saved in the same clock-minute now each show their own duration/languages/cost, instead of each other's.

Deviations

• Threaded identity by giving each IndexedSession its own entry (built in buildDashboardModel) rather than adding a name field to SessionIndexEntry + changing the aggregateSessions signature — this keeps the archive contract and its ~9 test call sites untouched, and moves the pairing into the pure, node-testable buildDashboardModel (renderHistory is DOM-coupled). It matches the ticket's fix option (a) ("build the SessionIndexEntry alongside each IndexedSession").
• Left DashboardStats.index in place (still produced + tested as archive API) though the app history no longer reads it — removing it would be an out-of-scope contract change.

🤖 Generated with Claude Code

[project7-interns]

Verdict: APPROVE

Epic Alignment: PASS

The #170 fix preserves session/index identity before aggregation and eliminates the stable-sort tie mispair without changing the archive index contract.

Checked (evidence)

• Identity pairing: src/dashboard.ts:57-64,91-107,445-454 derives and carries each session’s own entry; the positional reversal is removed.
• Shared contract: packages/archive/src/dashboard.ts:83-104,142 centralizes entry derivation in toSessionIndexEntry, used by both aggregate output and dashboard model.
• Regression: test/dashboard.test.ts:188-228 covers equal date/clock sessions with distinct duration, languages, and cost.
• Design Fidelity: live PR table spot-checked; historyRow markup/classes and rowMeta() formatting are unchanged, with no CSS/token/layout/animation diff.
• Riskiest part of this diff: tie ordering in stable sorts; row metadata is now object-bound rather than position-bound.
• Kill-list: scanned new ranges — clean.
• CI: gh pr checks 183 → app-macos, color-guard, no-stub-gate, packages-linux all pass.

Findings

• None.

Decision

The amended identity approach and its tie regression fully satisfy #170 without scope creep.

[realproject7]

RE2 — APPROVE at PR #183 head (pending app-macos/packages-linux green)

The #170 positional-pairing bug is correctly fixed via identity pairing, the amendment is honored, and the toSessionIndexEntry extraction is byte-identical to the old inline push. Reviewed the whole diff (4 files) and independently ran the app + archive suites + both typechecks + lint on Linux with a real pnpm install (correct workspace linkage).

Checked (evidence)

• Bug correctly diagnosed & fixed. Old buildDashboard's history pairing read index[index.length-1-i] against newest-first m.sessions, assuming the chronological index is the exact mirror. On a (headerDate, startClock) tie, Array.sort is stable so BOTH the ascending index and the descending sessions preserve the tie-group's input order (not mirrored) — the reversal lands index[len-1-i] on the other tied session, swapping duration/langs/cost subtitles (title/click target came from m.sessions so stayed correct). Fix: each IndexedSession now carries its OWN entry = toSessionIndexEntry(session) built in buildDashboardModel (src/dashboard.ts:96-101), and renderHistory uses m.sessions directly (:449, positional reversal + comment deleted) — pairing by identity, never position.
• Amendment honored. Per the Batch-36 amendment (SessionIndexEntry has no name; aggregateSessions takes only ParsedSession[]), the fix builds the entry alongside each session before the positional step (fix option a) and keys the row by the session it was derived from — it does NOT try to thread a name into SessionIndexEntry. Verified SessionIndexEntry's shape is unchanged.
• Extraction is byte-identical (so stats.index is unchanged). Compared toSessionIndexEntry (packages/archive/src/dashboard.ts:83-96) field-by-field against the old inline index.push at 5b02a74: every field matches, and the old talkRatioMic/smoothScore locals were metrics ? finite(metrics.talkRatioMic|smoothScore) : null — exactly what the helper computes. aggregateSessions now just calls the helper in the same loop/order, so stats.index output is identical. Confirmed empirically: the archive package's 8 dashboard tests pass unchanged.
• No other positional consumer. Grep of PR-head src/dashboard.ts for .index/stats.index → the history row was the only positional pairing; it's gone. Trends etc. use stats.index content, not position-against-sessions.
• Test is non-vacuous. New tie regression: two sessions on the same 2026-06-10 09:00 (the "(2)" filename-collision case) with distinct duration(45/10)/langs(EN→KO/JA→FR)/cost → asserts each row's entry carries its OWN numbers (would swap under the old reversal — dev's negative control confirmed) + an invariant loop that every entry is derived from its own session's meta.
• Independently reproduced (node 24 / vitest 3.2.6, real pnpm install): app test/dashboard.test.ts 18/18 (incl. the tie regression); full app suite 22 files / 154 tests; @livecap/archive package 12 files / 95 tests (incl. the 8 unchanged dashboard tests); app-tsconfig AND host-tsconfig tsc --noEmit both clean; eslint on the 4 changed files clean.
• Boundary/invariants/design: 4 files only (src/dashboard.ts, packages/archive/src/dashboard.ts + barrel index.ts, test/dashboard.test.ts); package.json/lock untouched (no new deps); no reserved files; no caption content logged (entry carries only meta numbers/langs/title). No visual change — historyRow markup/classes and rowMeta() untouched; only the (now correct) entry feeds them.

CI condition

color-guard/no-stub-gate pass; app-macos/packages-linux pending — packages-linux runs the archive-package tests, which I reproduced green locally, but the merge gate should still wait for both green. (Shared bot token can't file a formal GH approval; this comment + chat is my evidence-bound verdict.)