release: dev → prod (2026-09-06, second cut: notification copy + roster doc) - #1525

…live DB (#1521)

## Summary

A 2026-09-06 E2E run found that the "By org archetype" credential roster in `docs/enterprise/90-audits/03-verification-guide.md` had drifted from the live seed data. Specifically, the doc listed `rachel.anderson@hotmail.com` as an IIT Madras LEARNER, but that user has no live membership at IIT Madras at all — the actual IIT Madras LEARNER cohort is the Brown family (`sophia`, `thomas`, `victoria`, `william`).

While regenerating the roster from Supabase project `pzmbxqdgibfkhjwzeprf`, three more stale rows turned up in the same table:

- **Wipro LEARNERs**: doc said `olivia.anderson@gmail.com`, `patrick.anderson@outlook.com`, `priya.anderson@yahoo.com` (none have live Wipro memberships); live LEARNERs are `robert.brown@gmail.com`, `samantha.brown@outlook.com`, `sarah.brown@yahoo.com`.
- **Wipro OWNER**: doc said `samantha.anderson@yahoo.com`; the live persistent OWNER (besides `tour-owner@familiarise.dev`, which is correct) is `zara.brown@yahoo.com`.
- **Platform admin**: doc said `olivia.brown@protonmail.com`, but that user's live `role` is `CONSULTANT`, not `ADMIN` — the login would not reach the admin surface. A live `ADMIN`-role user is `robert.davis@yahoo.com`.
- **Arjun Anderson's Coaching slug**: doc said `arjun-anderson-coaching-mrpk`; the live slug is `arjun-anderson-coaching-2ncb`.

The Wipro OWNER (tour), IIT Madras OWNER/EXPERT rows, LearnPro OWNER/EXPERT rows, and the multi-org-consultant note all checked out against the live data and were left unchanged.

**Scope note:** per the task, only the credential/roster table and its accompanying note were rewritten. The step-by-step Flow B (`olivia.anderson@gmail.com` as the Wipro learner), Flow C (`rachel.anderson@hotmail.com` as the IIT learner to assign), and Flow G (`olivia.brown@protonmail.com` as the ADMIN login) narrative sections still reference the pre-drift logins and were intentionally left untouched to avoid widening this PR's scope — those should be corrected in a follow-up pass so the walkthroughs point at working logins. The "Seeded data summary" counts line (`4 orgs · 78 users · 22 enterprise memberships · ...`) also was not reverified beyond the active-membership count for the 4 archetype orgs (26, not 22) and total live `users` (250, not 78) — those higher-level counts likely reflect years of accumulated dev/prod activity on the single shared Supabase project rather than a clean reseed, and reconciling them was out of scope here.

## Regeneration SQL

```sql
-- Active enterprise memberships for the four seed orgs, joined to email/name/role.
SELECT o.name AS org_name, m.role, u.email, u.name, m.status
FROM "Membership" m
JOIN users u ON u.id = m."userId"
JOIN organizations o ON o.id = m."organizationId"
WHERE o.slug IN ('wipro', 'learnpro-academy', 'iit-madras', 'arjun-anderson-coaching-2ncb')
ORDER BY o.name, m.role, u.email;

-- Live ADMIN-role users (the "Platform admin" persona).
SELECT email, name, role FROM users WHERE role = 'ADMIN';
```

## Verification

- `npx prettier --check docs/enterprise/90-audits/03-verification-guide.md` fails, but this is pre-existing formatting debt unrelated to this change — the same command fails identically against the pre-edit `dev` version of this file (verified via `git stash`). Running `prettier --write` on the whole file would reformat ~150 unrelated lines (table alignment, `*emphasis*` → `_emphasis_`, etc.), which the task explicitly asked me not to do, so the diff here is scoped to only the touched roster lines.
- Read-only `SELECT`s only were run against Supabase; no writes were made.

## Undone

- Flow B / Flow C / Flow G narrative logins are not yet updated to match the corrected roster (see Scope note above).
- The "Seeded data summary" counts line was not reconciled against live totals.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

https://claude.ai/code/session_01EyngsXG829TRTBof4CSGT1