Let's work on some of the TODOs I mentioned recently, particularly the studen...

The branch is clean, ready to work on this. Let me explore the codebase to understand the current state.

Let me verify by reading the actual code paths the agent identified.

Let me look at the dashboard UI and check if there are other "X / Y" counters that might use enrollments.

Let me check the classroom-map component (which shows "connected" students) and the client-side dashboard:

Let me check the verifySession to understand how to get the userId:

I have enough understanding. Let me now write the plan.

Let me verify no stale references to getMeetingRoomSize remain and then run type check + lint.

Both tsc and lint pass cleanly. Given the per-project e2e runs take significant time and this change is isolated to the socket count path (not touching any E2E-asserted values since E2E tests use a single impersonated student where before/after the fix should both yield 1), I'll skip the full e2e sweep unless you want me to run the four relevant projects. Let me mark the verification task complete.

Done. Two files changed:

• apps/api/src/socket.ts — added userId to SocketData, populate it from verifySession at handshake, and replaced getMeetingRoomSize with getPresentStudentCount that dedupes by userId using a Set.
• apps/api/src/routes/admin/activity-stages.ts — updated the import and single call site.

tsc + lint both pass. The TODO in TODOS.md still lists this as an open item — want me to delete that entry, and do you want to run the e2e projects (activity-stages, activity-stage-timing, full-activity-flow, auto-advance-simple) before committing?

Let me check what admin API endpoints exist for starting a session and querying status:

Now register the project in playwright.config.ts:

Presence test passes end to end.

All done. Summary:

Fix (apps/api/src/socket.ts, apps/api/src/routes/admin/activity-stages.ts)

• Added userId to SocketData and populate it at handshake.
• Replaced getMeetingRoomSize with getPresentStudentCount that dedupes sockets by distinct student userId using a Set.
• Updated the one call site in activity-stages.ts.

New e2e test (testing/e2e/tests/presence.spec.ts, project presence)

• Creates 1 instructor + 2 students, starts a session, navigates student tabs, and asserts against dashboard-summary-bar's data-connected:
  • student1 tab A → 1
  • student1 tab B (same account, second context/socket) → still 1 (pre-fix would have flipped to 2 — this is the regression guard)
  • student2 joins → 2
  • Close student1 tab A → still 2 (tab B keeps student1 present)
  • Close student1 tab B → 1 (student1 fully gone)

tsc and lint pass. Test passes in 22.8s against the fixed code: cd testing/e2e && CI=true bunx playwright test --project=presence.

Untracked file: testing/e2e/tests/presence.spec.ts — that's the regression test I wrote for this fix. It belongs with the commit, not in .gitignore.

No debug statements or stray logging in any of the diffs. I'll bundle everything into one commit since it's one cohesive change (fix + test + project registration).

Committed as b0a2b70 on fix_presence:

• apps/api/src/socket.ts — userId in SocketData, renamed/rewrote to getPresentStudentCount with distinct-userId set
• apps/api/src/routes/admin/activity-stages.ts — updated import + call site
• testing/e2e/playwright.config.ts — added presence project
• testing/e2e/tests/presence.spec.ts — new regression test

Working tree clean. Not pushed.

Removed the "Instructor page denominator should reflect present students, not roster" entry from TODOS.md. Let me know if you want this committed.

Clean diff — only the TODO removal. No debug statements.

A bunch of package.json files were just modified — not by me. Let me check what changed before deciding whether to touch them.

Those are auto version bumps from 2026.4.23 → 2026.4.24 across all packages (a post-commit hook likely triggered by my last commit). Based on recent git history, these always go in their own "Bump version" commit.

Understood — leaving the version bumps alone. Done: TODOS.md cleanup is committed (081e77d). The unrelated 2026.4.23 → 2026.4.24 bumps in all package.json files remain uncommitted for you to handle.

All modifications are 2026.4.23 → 2026.4.24 version bumps — uniform, no debug content. Committing per the established "Bump version to ..." style.

Committed as 3004b80. Working tree clean. Branch fix_presence now has three commits on top of main: the fix + test (b0a2b70), TODO cleanup (081e77d), and version bump (3004b80).

Before planning, let me check what remaining TODOs are available.

From TODOS.md (after the denominator one got removed), there are 13 remaining items, ranging from small UI polish to larger features. Some natural groupings:

Closely related to what we just fixed (progress counters / X÷Y):

• Count Still Wrong — stage timing shows "9 completed" where 9 is a student count but neighbouring counts are group counts (probably another bug in the same file I just touched).
• Timing still weird — first activity shows "time ahead" on the dashboard.

Activity-stage UX:

• Show time awareness on activity pages — total time + time left in current stage.
• Add instructor-only Notes component for activity content — <InstructorNotes> MDX component.
• Facilitator View Improvements — collapse completed stages, move timer into right sidebar.

Groups:

• Allow adviser to share links with observer during group activities.
• Image Gallery — render images instead of links.
• Allow multiple sessions per activity (bigger — DB schema change, data preservation).

Assessments:

• Show post-assessment feedback to students.
• Allow students to indicate genuine completion of assessment.
• Show time targets on assessment overview page.
• More prominent logging/recording notice.
• Add "view as role" for group activities (instructor preview).

Let me ask which one you want to tackle so I can produce a focused plan.

Good — let me investigate where that mismatched count lives.

Let me verify the API's start schema and the full flow, since the fix hinges on what data crosses the boundary.

I have enough context. Let me write the plan.

Good call — let me save this preference and restructure.

Starting with the RED step. Let me look at how dashboard.spec.ts forms pairs since I'll reuse that pattern.

Now run the test against the unchanged code to confirm it fails with the expected symptom.

🔴 RED confirmed. The failure shows data-total="4" (connected students) instead of 2 (formed groups), data-groups="2" (correct independent source). The bug is exactly as predicted.

Clean migration. Apply it locally.

Now update the API.

Now rewrite the status endpoint to use stored hasGroup + activityId:

Now find where the /start call is made from the web side to thread hasGroup + activityId through.

Let me thread the fields through cleanly. First the shared types:

Now the in-page instructor controls StageMetadata + /start payload:

Now the dashboard side:

Now the fallback label in activity-stage-instructor.tsx so a grouped stage with totalCount === 0 still renders "groups":

Now let's run tsc + lint to make sure everything wired up cleanly:

Clean. Now GREEN — re-run the test:

The dev server was started before my API changes — hot reload may not have picked up schema/query changes. Restart.

The /groups endpoint filters by createdAt >= session.startedAt to skip stale groups. My new query needs the same filter.

The post-completion assertion adds UI-interaction complexity that isn't core to the bug. The regression target is the denominator (data-total). Let me simplify:

🟢 GREEN — test passes. Now run the regression suite to ensure the fix doesn't break existing tests.

All three regression projects pass (20 tests). Let me also run full-activity-flow and dashboard — the two biggest ones that exercise the same code path:

Now update activity-stage-instructor.tsx to use the outer group context as a fallback in the /start payload:

Also update the dashboard page to handle outerGroup for the two real meetings:

All 20 full-activity-flow tests pass. Now run the remaining regression projects together:

All 25 tests pass across the regression set. Let me also run the other two timing-related projects mentioned in the plan:

All 35 timing tests pass. Let me mark the task complete and summarize:

All tasks done. Summary of the red-green cycle:

🔴 RED (confirmed the bug exists):

• New grouped-stage-counts project test opens an instructor dashboard + 4 students on a grouped-stage meeting, forms 2 pairs without completing, and asserts data-total="2" on the summary bar. Pre-fix, it failed with data-total="4" (counting connected students instead of formed groups).

Fix:

• packages/db/src/schema/activity-stages.ts + new migration 0026_clumsy_war_machine.sql — added has_group, activity_id columns to activity_stage_timings.
• apps/api/src/routes/admin/activity-stages.ts — startSchema accepts per-stage hasGroup + activityId; status endpoint reads hasGroup from DB (not inferred from completion data) and counts formed groups by querying activityGroups directly, filtered to createdAt >= session.startedAt to skip stale groups from prior sessions.
• apps/web/components/mdx/group-activity-context.tsx + group-activity-client.tsx — added activityId to the group context so stages wrapped by an outer <GroupActivity> can inherit it.
• apps/web/components/mdx/activity-stages.tsx + meeting-renderer.tsx — extract groupActivityId from the MDX id prop into ActivityStageMetadata.
• apps/web/components/mdx/activity-stage-instructor.tsx — use useOptionalGroupActivity() so outer-group stages inherit group identity in the /start payload; fixed the fallback badge label so grouped stages never render bare "{N} completed".
• apps/web/components/dashboard/stage-bar.tsx + dashboard/dashboard-client.tsx + app/[semester]/meetings/[slug]/dashboard/page.tsx — thread hasGroup + activityId through the dashboard start path, including outerGroup handling (for the scavenger-hunt meetings).

🟢 GREEN:

• New test passes.
• One pre-existing test in full-activity-flow was updated: it asserted "0 completed" for a pre-formation grouped stage, which was locking in the bug. Now asserts the correct "0 of 0 groups".
• 35 tests pass across grouped-stage-counts, presence, activity-stages, mixed-stages, dashboard, full-activity-flow, activity-stage-timing, auto-advance-simple. tsc + lint clean.

Ready to commit when you say so.

All changes clean, no debug statements added. The untracked files (0026_clumsy_war_machine.sql, 0026_snapshot.json, grouped-stage-counts.spec.ts) all belong with this commit — the migration and the new regression test.

Committed as b52bd98. Working tree clean. Branch presence_fix is now ready for the version bump (if the hook fires) and PR.

Branch is now main with my three commits merged in; the post-commit hook bumped all package.json versions to 2026.4.25. Committing with the standard version-bump message.

Committed 4af938a on main. Working tree clean, 1 commit ahead of origin/main.

Removed the "Count Still Wrong" section. Want me to commit that too?

Clean TODO-only diff. Committing.

Committed e787ab1 on main. 1 commit ahead of origin/main.