# Acceptance checks An implementation is complete only with automated evidence where practical and a written manual test report. These are required scenarios, not claims that a build has passed. ## Storefront and branding 1. Seed 12 synthetic products, three collections and at least three variant products, including sold-out stock. Browse, search, filter, add variants and edit the cart at 360px and desktop widths. 2. Render the same catalog under minimal, editorial and bold presets. Changing a theme alters layout and typography without changing SKUs, prices, order data or canonical URLs. Preview, publish, rollback and concurrent edits work; stale edits conflict. 3. Core browse/checkout/admin flows work by keyboard and screen reader; audit contrast/focus/error announcements. Reduced motion is respected. Measure public product page performance using realistic image/network budgets and document results; optimize responsive images and avoid loading admin code publicly. 4. Product and collection HTML exposes accurate metadata before JavaScript runs. A variant share reopens the right option; QR, WhatsApp and native share fallback work. A shared cart leaks no PII and revalidates price/stock. Draft/archived products cannot be purchased through stale links. ## Payment and stock (use deterministic concurrency barriers) 5. Two buyers compete for the last unit: exactly one reservation succeeds. Concurrent multi-SKU carts cannot deadlock indefinitely or partly reserve stock. 6. Repeated checkout key returns one order; changed payload conflicts. Provider-create timeout does not produce uncontrolled duplicate payable attempts. 7. Duplicate and out-of-order captured/failed webhooks produce one inventory consumption and one confirmation intent; mismatched amount/currency/order, bad signature and test/live mismatch do not mark paid. 8. Crash before/after event persistence and before/after order transition; replay recovers without double consumption. Polling and webhook settlement racing give the same result. 9. Reservation expiry racing capture releases or consumes once. Late success with available stock reacquires atomically; unavailable stock creates a visible paid exception and refund path. No paid order silently disappears. 10. Two captured attempts on one order do not ship twice; extra payment is flagged. Two parallel refunds cannot exceed capture, including pending refunds. Timeout/retry does not issue a duplicate refund; rejected refund remains visible. Refund does not automatically restock; explicit restock is idempotent. 11. Mobile Android/iOS UPI round trip, unavailable UPI app, back/cancel, browser termination/reopen, in-app browser and desktop QR have documented real-device results. Mock/sandbox tests do not substitute for these. Do not execute live transactions without approval. ## Operations 12. Import malformed/duplicate SKU/negative stock/stale CSV: dry-run identifies rows and commits nothing. Valid import commits once; formula-like cells are safely exported. Stock corrections preserve reservations and audit history. 13. Delivery outside allowed PIN zones is rejected; overlapping-zone precedence, free-shipping threshold, pickup, coupon expiry/caps and tax rounding match documented examples. Concurrent final coupon use is not oversubscribed. Order snapshots survive later edits. 14. Catalog staff cannot refund or change credentials; fulfillment staff cannot alter prices; logged-out and cross-role API requests fail. Owner MFA/session revocation/CSRF/rate limits are tested. Private links and exports do not leak addresses or payment data. 15. Partial cancellation, shipment and refund quantities reconcile per line. A notification-provider outage does not roll back payment; retry and dead-letter state are visible. Disabled email/WhatsApp integrations never claim delivery. 16. Restore a backup into an isolated environment; verify image references, order history and recovery procedure. Reconcile payment state safely without sending duplicate messages or making financial writes automatically. 17. With coupons/reviews/pickup switched off, their entry points disappear without broken routes. If reviews are enabled, non-purchasers cannot impersonate verified buyers and moderation/rate limits work. 18. Validate dashboard against a fixed fixture with taxes, shipping, discounts, partial refunds and day-boundary timestamps. Shared campaign attribution contains no buyer PII. ## Handoff evidence Provide test commands/results, dependency versions, migration/rollback instructions, configuration inventory, screenshots for all three themes, failure-mode runbook and known limitations. Label each integration: unconfigured, sandbox-verified or live-verified. No real customer data, secrets, fake endorsements or fabricated validation claims in fixtures or demos.