feat(seed): the lead Activity tab shows logged interactions out of the box (#1258)

[objectstack-fleet]

Fixes #1258

Route A (maintainer ruling 5386129833): the lead Activity tab now shows logged interactions out of the box, through the platform's ActivityPointer (sys_activity). No app-side imitation object and no extra writer.

What changed

• src/sales/data/activity.seed.ts: a new leadInteractionPointers dataset on sys_activity. It is derived from the same EVENT_SPECS that seed the crm_event rows. Each pointer is the row log_call / log_meeting would have written for a held lead event: type: 'completed' (the task feed kind the tab filters to), the action's own summary format, object_name/record_id pointing at the lead by email, source_object/source_id pointing at the event by subject, and the action's metadata shape. Only call/meeting events (the two writers that exist) younger than the 14d telemetry retention are included.
• The same file adds two earlier held interactions, so each tab reads as a short history instead of a single row: CloudFirst — follow-up meeting after the Cloud Summit booth (Lisa Thompson, 8d) and EduTech Labs — referral intro call with David Kim (9d). Both leads already carry last_contacted_date: today() for their later call, so no recency clock moves.
• src/sales/data/index.ts and objectstack.composition.ts: the dataset is exported and registered in CrmSeedData right after the event attendees, because it resolves against leads and events seeded earlier.
• test/seed-consistency.test.ts: the sys_* guard block is deleted on purpose. Its own comment asked for this: "If a future platform release makes an ActivityPointer target seedable, delete this test deliberately — do not widen it". objectstack#11339 did that, and the 17.4.0 pin carries it. A positive pin replaces it. Every pointer must name a seeded lead and a held event on that same lead, with type: 'completed', inside 14 days. Every held lead call or meeting inside the window must have its pointer. At least 3 pointers across at least 2 leads.
• One changeset (hotcrm: patch).

Measurement first (17.4.0, booted app, fresh DB)

I re-measured the three 17.1.0 blockers in the old guard comment before writing the final seeds. First I booted with a scratch version of the dataset (2 rows), then again with the final 4 rows. Commands: rm -rf .objectstack/data && pnpm build && objectstack start -p 4158. I read the rows from sqlite and checked them in the browser.

leg	17.1.0 (old guard)	17.4.0 (this PR)
(i) readonly strip	rows land (isSystem write)	rows land: timestamp, type, summary, metadata, object_name, source_object all stored, although every field is readonly: true
(ii) pointer resolution	record_id stored the literal natural key	record_id = the lead's runtime id (pxDFU7RFoMejXWbp = lisa.t@…, Cmp7aiCyo4Dr81PH = dkim@…). source_id = the event's runtime id (all 4 match crm_event.id by subject)
(iii) retention / shards	14d telemetry, day-sharded	still 14d / 14 day-shards: rows live in sys_activity__r20260925 behind the sys_activity view. Seeds use daysAgo(1..9), and the builder filters out anything 14d or older
Activity tab	empty	Lisa Thompson: Activity (2) and David Kim: Activity (2), each entry with "View source". Controls: Alice Martinez (new, planned call only) and Noah Patel both show Activity (0) / No activity recorded. The console's query was GET /api/v1/data/sys_activity?filter=["and",["object_name","=","crm_lead"],["record_id","=",LEAD_ID]], and it answered 200 for all four
replay boot	not measured	restarted on the same DB: still 4 pointer rows, 4 distinct source_id, 0 null record_id (the composite ['source_object','source_id'] key dedupes on resolved ids)

Verification (at e0563de)

pnpm verify chain, one step at a time, exit codes captured before any pipe:
validate 0 · typecheck 0 · lint 0 · lint:i18n-gate 0 · hygiene 0 · hygiene:tokens 0 · build 0 · test 0 (Test Files 172 passed (172), Tests 3688 passed | 1 skipped (3689)).

Reverse check of the new pin, with the fix committed first. I deleted the leadInteractionPointers, registration line from objectstack.composition.ts (grep count 1, then 0). Expected direction: red. Observed: red, Tests 2 failed | 36 passed (38). The two that failed were "seeds a handful…" and "every held call or meeting … has its pointer". The per-row test stays green with zero rows, as expected, because the first test guards it. Restored with git checkout HEAD -- objectstack.composition.ts under a trap, and git diff HEAD came back empty.

Token ratchet before and after (base aa61e1d vs this branch): all 12 ceilings are unchanged and green. Seed data is outside the ratchet by ruling (src/*/data). Its out-of-scope line moved from ~95,777 to ~96,270 tokens. objectstack.composition.ts is outside src/, so the ratchet does not measure it.

Acceptance notes

• File surface. The claim named src/sales/data/index.ts "if registration is needed", but CrmSeedData is assembled in objectstack.composition.ts, because the replay order interleaves packages. So the registration is a one-line addition there, plus the import. No open PR touches that file.
• Brief window on first boot. On a fresh DB the pointer rows were first stored with record_id/source_id null. The loader's later seed pass filled them about 30s later (created_at 13:17:03, updated_at 13:17:33). This is the platform loader's deferral, not something this repo shapes. A tab opened in the first seconds after first boot can read empty. Noted only.
• The entries show the actor as System, because a seed cannot name a user (actor_id / actor_name are not authored). That is the same doctrine as owner_id.
• sys_activity retention is 14d, so on a long-running demo instance these rows age out. A replay boot re-evaluates daysAgo(), but it did not rewrite the existing rows in my replay measurement (updated_at unchanged).

---

Generated by Claude Code

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

1 Skipped Deployment

Project	Deployment	Actions	Updated
hotcrm	Ignored		Sep 25, 2026 1:25pm UTC