feat(mobile): bring inbox analytics to parity with desktop

[Gilbert09]

Summary

Brings mobile-app inbox analytics up to parity with the desktop app so events fired from both clients land in the same PostHog buckets and can be funnelled together.

Mirrors the events introduced in desktop PRs #2228, #2287, #2295, and #2369.

Events wired on mobile

• Inbox viewed — fired once per inbox-tab focus when reports settle. Carries the priority (priority_p{0..4}_count, priority_unknown_count) and actionability (actionability_immediately_actionable_count, etc.) breakdowns from feat(code): add priority + actionability to all inbox events #2295, plus report_count/total_count/ready_count/source_product_filter/status_filter_count/has_active_filters/is_empty/is_gated_due_to_scale: false.
• Inbox report opened — fired from the /inbox/[id] detail screen on mount, with rank + list_size snapshotted from the list view, open_method (currently "click" by default — mobile has no Cmd/Shift modifiers), previous_report_id chained across opens via the inbox store, and the priority + actionability snapshot from feat(code): add priority + actionability to all inbox events #2295.
• Inbox report closed — fired when the detail screen unmounts (close_method: "deselected"), with time_spent_ms, scrolled, and snapshotted priority/actionability.
• Inbox report scrolled — fired once per open on the first scroll inside the detail pane, with time_since_open_ms.
• Inbox report action — fired for:
  • Dismiss / snooze via DismissReportSheet (with dismissal_reason and dismissal_note truncated to 1000 chars per feat(inbox): capture dismissal_note in dismiss analytics event #2287). Snooze is detected the same way as desktop, via the snoozesInsteadOfDismiss flag on the reason option.
  • Start task on the detail screen → create_pr, surface detail_pane.
  • Tinder swipe-right (accept) → create_pr, surface list_row.
  • Tinder swipe-left (dismiss) → dismiss, surface list_row.
  • Expanding the Signals list on the detail screen → expand_signal.

For #2369 (tag discuss-launched task events with signal_report_id): the task detail screen registers a signal_report_id PostHog super-property whenever the loaded task has one, and clears it on unmount. Every subsequent event fired from that task screen carries the signal report id, matching the desktop setActiveTaskAnalyticsContext behaviour.

Events intentionally skipped

• feat(inbox): Capture "Signal source connected" analytics event #2380 Signal source connected — mobile has no signal-source toggle or DataSourceSetup wizard surface. There is no place in apps/mobile to fire this from without inventing a flow.
• feat(analytics): tag discuss-launched task events with signal_report_id #2369 Task created tagging — mobile does not currently fire a Task created analytics event at all, so there is no event to add the signal_report_id property to. Only the super-property half of feat(analytics): tag discuss-launched task events with signal_report_id #2369 ports cleanly to mobile, and that's what's implemented here.

Implementation notes

• New module apps/mobile/src/lib/analytics.ts houses ANALYTICS_EVENTS, the typed property interfaces, useAnalytics(), useActiveTaskAnalyticsContext(), and computeReportAgeHours(). Event names + shapes are literal mirrors of apps/code/src/shared/types/analytics.ts — kept in lockstep manually since the desktop types live inside apps/code (no shared workspace package to import from).
• New hook useInboxEngagementTracker (mobile-adapted port of the desktop hook of the same name) owns the OPENED/CLOSED/SCROLLED lifecycle keyed on the open report.
• inboxStore gained lastVisibleReportIds + previousOpenedReportId so the detail screen can record rank/list_size/previous_report_id on OPENED without prop-drilling.
• DismissReportSheet now passes the dismissal { reason, note } back via onDismissed so the parent can route it through analytics — keeping the sheet stateless about which surface launched it (toolbar vs. detail).
• No desktop code touched.

Test plan

• pnpm --filter @posthog/mobile test — 119 passed (14 new across lib/analytics.test.ts, features/inbox/utils.test.ts, features/inbox/hooks/useInboxEngagementTracker.test.ts).
• pnpm --filter @posthog/mobile lint — clean.
• No new TypeScript errors (npx tsc --noEmit count unchanged at 47 pre-existing errors).
• Smoke in dev: open inbox tab → confirm Inbox viewed in PostHog activity feed with the new breakdown counts.
• Open a report → confirm Inbox report opened with rank/list_size/previous_report_id/priority/actionability. Scroll inside → confirm one Inbox report scrolled. Tap back → confirm Inbox report closed with realistic time_spent_ms and scrolled: true.
• Dismiss a report via the sheet with a reason + note → confirm Inbox report action with action_type: "dismiss", dismissal_reason, dismissal_note (truncated at 1000 chars if longer). Empty/whitespace-only note → no dismissal_note. Reason already_fixed → action_type: "snooze" and no dismissal_reason/note attached.
• Tinder swipe-left → Inbox report action with action_type: "dismiss", surface: "list_row". Tinder swipe-right (accept) → Inbox report action with action_type: "create_pr", surface: "list_row".
• Tap "Start task" on detail → Inbox report action with action_type: "create_pr", surface: "detail_pane". Subsequent task screen events carry the signal_report_id super-property. Navigating back to the inbox clears the super-property.

[greptile-apps]

Comments Outside Diff (1)

1. apps/mobile/src/features/inbox/hooks/useInboxEngagementTracker.ts, line 798-839 (link)

   Spurious CLOSED+OPENED pairs when rank/listSize/report change while a detail screen is open

   rank, listSize, and report are all included in the useEffect dependency array. If the inbox list refreshes in the background (React Query background refetch, a report is added/removed, or the order changes), every one of these changes while the user is actively reading a report will fire a cleanup INBOX_REPORT_CLOSED (with near-zero time_spent_ms) immediately followed by a new INBOX_REPORT_OPENED. This inflates both event counts and makes funnel analysis unreliable.

   The fix is to make only reportId drive the open/close lifecycle. Snapshot rank, listSize, openMethod, and previousReportId once at the moment the report opens by feeding them through refs and reading the ref values at mount rather than keeping them as live deps. Similarly, report can be read from a ref inside the effect body (since only its initial value matters for the OPENED event) — only reportId should gate the re-run.

   Prompt To Fix With AI

   This is a comment left during a code review.
   Path: apps/mobile/src/features/inbox/hooks/useInboxEngagementTracker.ts
   Line: 798-839
   
   Comment:
   **Spurious CLOSED+OPENED pairs when rank/listSize/report change while a detail screen is open**
   
   `rank`, `listSize`, and `report` are all included in the `useEffect` dependency array. If the inbox list refreshes in the background (React Query background refetch, a report is added/removed, or the order changes), every one of these changes while the user is actively reading a report will fire a cleanup `INBOX_REPORT_CLOSED` (with near-zero `time_spent_ms`) immediately followed by a new `INBOX_REPORT_OPENED`. This inflates both event counts and makes funnel analysis unreliable.
   
   The fix is to make only `reportId` drive the open/close lifecycle. Snapshot `rank`, `listSize`, `openMethod`, and `previousReportId` once at the moment the report opens by feeding them through refs and reading the ref values at mount rather than keeping them as live deps. Similarly, `report` can be read from a ref inside the effect body (since only its initial value matters for the OPENED event) — only `reportId` should gate the re-run.
   
   How can I resolve this? If you propose a fix, please make it concise.

Prompt To Fix All With AI

Fix the following 3 code review issues. Work through them one at a time, proposing concise fixes.

---

### Issue 1 of 3
apps/mobile/src/features/inbox/hooks/useInboxEngagementTracker.ts:798-839
**Spurious CLOSED+OPENED pairs when rank/listSize/report change while a detail screen is open**

`rank`, `listSize`, and `report` are all included in the `useEffect` dependency array. If the inbox list refreshes in the background (React Query background refetch, a report is added/removed, or the order changes), every one of these changes while the user is actively reading a report will fire a cleanup `INBOX_REPORT_CLOSED` (with near-zero `time_spent_ms`) immediately followed by a new `INBOX_REPORT_OPENED`. This inflates both event counts and makes funnel analysis unreliable.

The fix is to make only `reportId` drive the open/close lifecycle. Snapshot `rank`, `listSize`, `openMethod`, and `previousReportId` once at the moment the report opens by feeding them through refs and reading the ref values at mount rather than keeping them as live deps. Similarly, `report` can be read from a ref inside the effect body (since only its initial value matters for the OPENED event) — only `reportId` should gate the re-run.

### Issue 2 of 3
apps/mobile/src/app/(tabs)/inbox.tsx:50-79
**`Inbox viewed` silently skips every re-focus when data is cached**

`viewedFiredRef.current` is reset to `false` by the `useFocusEffect` blur cleanup, which is the correct intent ("fire once per focus"). However, the reset only takes effect when the React `useEffect` re-runs, and that effect only re-runs when one of its listed dependencies changes. If the user navigates away and back while the data is still fresh (no loading, no filter change, no new reports), none of the deps have changed so the `useEffect` doesn't re-run — `viewedFiredRef.current` remains `false` but the event is never sent.

The fix is to expose the focus change as a React state value so `useEffect` actually re-runs on re-focus. For example, replace the `useFocusEffect` reset with a `useState` focus-version counter (`setFocusVersion(v => v + 1)` on focus) and add that counter to the `useEffect` deps, or move the event-firing logic entirely into a `useFocusEffect` callback that waits for `isLoading` to be `false`.

### Issue 3 of 3
apps/mobile/src/app/inbox/[id].tsx:169-185
Side effect inside a `setState` updater — React Strict Mode (development) calls state updaters twice to detect impure functions, which means `tracker.signalAction(...)` fires twice per toggle in development. This can cause confusion during the smoke test step that checks for one `expand_signal` event. Move the analytics call outside the updater.

```suggestion
  const handleToggleSignals = useCallback(() => {
    const next = !signalsExpanded;
    if (next && report) {
      tracker.signalAction({
        report_id: report.id,
        report_title: report.title ?? null,
        report_age_hours: computeReportAgeHours(report.created_at),
        action_type: "expand_signal",
        surface: "detail_pane",
        is_bulk: false,
        bulk_size: 1,
      });
    }
    setSignalsExpanded(next);
  }, [report, tracker, signalsExpanded]);
```

_{Reviews (1): Last reviewed commit: "feat(mobile): bring inbox analytics to p..." | Re-trigger Greptile}

[stamphog]

The PR exceeds the automated review size ceiling (1273 lines, 14 files). It introduces new analytics infrastructure (a new analytics module) and a new engagement tracker hook, which warrants human review to validate the event naming conventions, correctness of tracking logic, and absence of data leakage.

[k11kirky]

would prefer we run a /simplify skill on these as part of the prompt <3 very verbose comments