Launch Sprint

The MVP doc records what is shipped; the Roadmap records the quarter direction. This doc is narrower and time-boxed: the single sprint that ends with launch. Anchor date: 2026-06-20, HEAD febb7da3b.

The thesis is simple. After the June deploys — transportation Phase 2–5, the conference notification/attendance integration, the public features + community hubs, exam-and-grade automation, the lumos/LMS optimize pass, the saas-dashboard de-fabrication, timetable production-readiness, and demo-seed auto-provision — almost every epic is code-complete. What stands between here and launch is no longer feature-building. It is four things: making the last few correctness fixes, confirming the database, turning on the ops and credentials, and running a QA pass that actually proves it. This sprint does exactly that, for all three launch audiences at once: the King Fahad pilot (Sudan, Sept 2026), the public SaaS at ed.databayt.org, and the Aldar enterprise pilot (UAE).

For requirements see PRD. For the pilot context see Pilot.

Where we are

State verified against block records and git log at HEAD febb7da3b. The "old status" column is the last roadmap reconcile (2026-05/06-15); the gap is real because a lot shipped since.

The launch gap

Three buckets. Bucket A is small and surgical, bucket B is mostly confirmation, bucket C is where the real lead time sits.

A — Surgical code (critical path)

B — Database (audit + apply, Neon-branch-first)

Audited against prod 2026-06-21 (prisma migrate diff, read-only). The bucket reduced to a single genuinely-pending migration — everything else was already live:

prisma migrate diff also surfaces pre-existing drift (live_class_* FK renames, roomName NOT NULL, index renames, live_class_events.eventId unique, PricingRule.updatedAt default) — deliberately not applied: reconciliation noise from earlier conference/live-class refactors, not this sprint's work. The catalog TOCTOU fixes, billing atomic CAS, exams AI cost cap, and RBAC/calculator tests are code-only — no DB change.

C — Ops / env / infra (the real lead time)

The 4-week sprint

Five tracks run in parallel across all four weeks: A Code, B Data, C Ops/Infra, D QA/Proof, E Polish. Each week has a theme and an exit gate.

Week 1 — Reachable & Truthful

• A · Fix the Playwright CI command (tests → test) so e2e runs · make STRIPE_API_KEY optional in /api/health · add the manage/ + paper/ exam index pages and confirm sidebar surfacing.
• B · Audit Neon prod vs Prisma; apply any additive drift on a Neon branch, then promote.
• C · Set Sentry + Upstash env in Vercel (immediate observability + correct rate limiting).

Week 2 — Money & Messaging

• A · ✅ Wired postFeeAssignment (assignFee + bulkAssignFees) — the accrual fix that stops the receivable going negative. Investigation found the other three blocked beyond wiring (no invoice/expense payment-recording flow; salary rule unbalanced) and fixed a pre-existing toCents ×100 ledger-inflation P0 — rules now store whole units, posting-rules.test.ts green (15/15). The debit=credit invariant test already existed, and the finance/content.tsx + salary/content.tsx 1/100 display bug is fixed. ✅ payroll progressive tax (replaces flat 15% in payroll + salary, unit-tested). The ÷100 ledger migration is a verified no-op safety net — a read-only prod check found zero inflated rows (all 200 entries are seeded whole-units), so the toCents fix deploys safely alone; the script is now self-verifying/fees-scoped, run post-deploy only as a sweep. ✅ invoice PDF + branded RTL email (logo into the PDF; email rebuilt on react-email — RTL, branding, itemized, signature). ✅ invoice + expense payment flows (markInvoicePaid + markExpensePaid). ✅ salary poster — rebalanced the salary posting rule + wired postSalaryPayment in processPayments → 6 of 6 posters wired; the double-entry ledger is fully wired for every finance money event. Week-2 finance code is complete — and an adversarial multi-agent review of the whole finance diff found + fixed 12 money-path defects (multi-tenant schoolId on updates, concurrent double-post races via conditional updates + a JournalEntry @@unique idempotency constraint, salary tax base, invoice partial-payment, an expenses/approve permission gate, 4 dashboards' ÷100 display, RTL discount sign). ✅ the expense list now surfaces Approve/Reject + Mark-paid (ExpenseRowActions), closing the expense flow end-to-end. What remains is ops (Track C: Oracle VM, Tap/FCM env, ADEK) and two additive deploy-pending migrations (the toCents ÷100 sweep — a verified no-op — and the JournalEntry @@unique).
• C · Stand up Oracle VM + Evolution API (6 env vars; per-school QR for King Fahad + Albayan) · set Tap + FCM env · file the ADEK inquiry now (longest external lead time).

Week 3 — Conference & Proof

• C · LiveKit 6-gate provisioning; schedule gate-6 in-school network test to a King Fahad / Aldar site visit.
• D · Real e2e pass now that CI runs: triage the test.skip backlog, un-skip what runs headless, add a named section-grid attendance case, give the assertion-less specs real expect()s · i18n RTL Arabic pass on the swept surfaces + parent-portal dictionary-key migration.
• E · Exams schema cleanup (bilingual → lang, reconcile Result/ExamResult, harden the ExamCertificate cascade chain).

Week 4 — Harden & Launch

• E · Drain the highest-value SaaS-dashboard backlog: ✅ invoice state-machine guard + atomic CAS (3be3506ed + 2026-06-21 — no voiding a paid invoice, no concurrent double-transition; buttons gated) · ✅ exams RBAC + calculator test coverage + AI generation/grading cost cap (2026-06-21) · ✅ catalog TOCTOU closed (proposal reject + 4 exam-approval actions now atomic; chapter/lesson count denormalization wrapped in transactions, 2026-06-21). A two-pass adversarial review (find → verify, then re-verify the fixes) confirmed + closed 8 further correctness defects across finance/catalog/exams/billing incl. a P0 ledger-reversal no-op. Remaining: i18n sweep · the Kanban demo route (still linked from analytics nav + command menu — full removal is a product call, not dead code).
• D · Per-pilot smoke on King Fahad + Albayan · load + rate-limit verification (now Redis-backed) · final secure pass · wire the hardcoded-string ratchet into the PR gate.
• Run the launch gate.

Launch gate

The sprint ends when this is green. Items that are audience-specific are tagged.

• Bucket-A code DEPLOYED to prod 2026-06-21 (febb7da3b..626b5d690 → origin/main; Vercel dpl_FJzP8oBEvcYVnw3L8YiraqkfqUcr ● READY on ed.databayt.org; smoke /en /ar /en/login /en/features /en/community /en/docs/sprint /en/docs/finance = 200; /api/health = healthy, Stripe-optional confirmed live)
• Neon prod schema confirmed against Prisma + additive manifest applied (Neon-branch-first): JournalEntry @@unique live on prod 2026-06-21; saas cols + attendance/lumos indexes already present; only the verified-no-op toCents sweep remains (post-deploy) — see bucket B
• Sentry + Upstash + FCM env set; Sentry receiving events
• Playwright CI command fixed (tests → test, sharded) — now runs; first green CI run still pending
• Finance ledger correct: toCents ×100 resolved + all 6 posters wired & balanced + idempotent createJournalEntry + a two-pass adversarial review closed 8 further defects (incl. the P0 reverseJournalEntry no-op, negative-net Cash inversion, payroll post-before-flip recovery); posting-rules.test.ts green
• King Fahad + Albayan tenants smoke-clean end-to-end
• Oracle VM live; WhatsApp QR scanned per pilot school
• Tap keys set — UAE / Aldar
• LiveKit network-test passes inside school WiFi — Aldar
• ADEK eSIS status acceptable (DRYRUN is fine for King Fahad / SaaS; live submission only gates Aldar compliance go-live) — _Aldar

External blockers

These sit outside the team's keyboard and are called out so launch is not falsely gated on them.

Deferred (post-launch)

Explicit non-goals this sprint, listed so no speculative work starts.

See also

• MVP — shipped state across the epics with maturity badges
• Roadmap — quarter outlook and active initiatives
• PRD — requirements, role model, NFRs
• Pilot — the pilot programme and success criteria