frontend phase 10

dev/shared-working-context/reports/mocks-registry.md

@@ -73,3 +73,4 @@ the frontend can build before the backend phase merges, and swap to the real HTT
 | `ILocationProvider` | `client/src/services/bookings/evv/locationProvider.ts` | **EVV GPS capture** — the only client seam f8 introduces. `getCurrentPosition()` never rejects (denied/unavailable → `null`, so a GPS problem is **advisory, never a block**). The **real** provider wraps `navigator.geolocation.getCurrentPosition`; the **mock** returns canned coordinates per mode so the in-range / advisory-out-of-range / denied paths are all demoable without a device (the mock `BookingsApi` computes the match against the same seeded reference point) | `NEXT_PUBLIC_EVV_MOCK_GPS` = `in_range` \| `out_of_range` \| `denied` \| `off` (default `in_range` while `USE_BOOKINGS_MOCK`, else `off`) | Set `NEXT_PUBLIC_EVV_MOCK_GPS=off` (or flip `USE_BOOKINGS_MOCK`) → the real `navigator.geolocation` provider is selected. Real **address-match math** stays server-side (backend geocoding seam), not here — this seam only *captures* the position | 🟡 |
 | `PaymentApi` | `client/src/services/payment/apis/mockApi.ts` | **The f9 checkout money path** — plays the PSP + webhook roles the client can't reach: `getCheckoutSummary` serves the unserved C6 breakdown (REQ-016; commission-net/VAT/service split via **integer parts-per-10000 BigInt math**, 12% fee / 10% VAT, reconciles to the rial); `initiatePayment` enforces b10 idempotency (same `Idempotency-Key` → same attempt; repeat after capture / lapsed window → **`409`**) and returns a `redirectUrl` into the local mock-gateway harness; `confirmGatewayReturn` on success is the **webhook-confirm stand-in and the missing f7↔f8 bridge** — flips the request `converted` (+ client-augmented `bookingId`, via `mockMarkBookingRequestConverted` in the f7 mock), inserts a **confirmed** booking into the f8 store (`mockInsertConvertedBooking`), and auto-issues the b11-shaped invoice (`moadianStatus: pending`, `pdfUrl: null` so the print path exercises); replayed returns converge idempotently; `getInvoice` 404s until issued | `USE_PAYMENT_MOCK` (`services/payment/constants.ts`, default `true`) | b10 initiate + b11 invoice are live and `paymentClientApi` maps them 1:1 (`Idempotency-Key` header, `GET invoices/{bookingId}`); deliver **REQ-016** (checkout summary — the real client already targets the proposed `booking_requests/checkout_summary/{id}` slug) + **REQ-017** (transaction status / `bookingId`; until then the real outcome poll maps `booking_requests/get` statuses and can't distinguish declined from slow) + **REQ-018** (invoice reachable post-capture), make the upstream `bookingRequests` flow real, then set flag `false`. No hook/component change | 🟡 |
 | Mock-gateway page (test harness) | `client/src/app/[locale]/(private-routes)/(customer)/bookings/checkout/gateway/page.tsx` | **Not a product feature** — a dev stand-in for the PSP's hosted payment page so the initiate → redirect → return round-trip is exercisable without a gateway: the mock `redirectUrl` points here, and its success/failure buttons drive both branches of the return surface (`?outcome=success\|failure`). Clearly labelled «درگاه پرداخت آزمایشی», dashed border | _none — only reachable via the mock's `redirectUrl`_ | On the real path b10's `redirectUrl` is the PSP's **absolute** URL (the checkout does a full `window.location.assign` for `http(s)` URLs), so this page is simply never linked; delete it when `USE_PAYMENT_MOCK` retires. The PSP's return deep-link into `/bookings/checkout/return` is backend/PSP config | 🟡 |
+| `RefundsApi` | `client/src/services/refunds/apis/mockApi.ts` | **The f10 customer cancel + refund surface** b11 doesn't serve (refunds are admin-only; no customer cancel command, no policy preview, no refund-by-booking, no fee-leg decomposition on the customer status → REQ-019/020/021). Reads the shared **f8 bookings store** (`mockGetBookingForRefund`) to resolve the tier by lead time (`free_24h` >24h / `partial_under_24h` <24h / `customer_no_show` started — client-invented codes → i18n keys) and the per-session refundable(un-started)/locked(completed-and-verified) breakdown, decomposing the refund across the two fee legs via **integer parts-per-10000 BigInt math** (`refundAmount + fee = refundableGross` to the rial). `cancelBooking` flips the booking → `cancelled` (`mockMarkBookingCancelled` stamps the b9 snapshot + cancels only un-started sessions) and creates a refund: **card → `succeeded`** immediately (no ETA); **BNPL → `approved`→`processing`→`succeeded`** over status polls with a `expected_customer_refund_eta` ~10 business days out (Fridays skipped) so the ~7–10-day banner renders. Enforces the outside-policy **`409`** (already-cancelled / nothing-refundable / non-refundable session). Seeds a **`failed`** refund on the cancelled booking 5004 so the contact-support state demos; booking 5002 is pinned to the BNPL channel; booking 5003 (new, mid-engagement) demos the mixed refundable/locked breakdown. Also adds bookings-store seeds 5003/5004 + the two non-seam exports | `USE_REFUNDS_MOCK` (`services/refunds/constants.ts`, default `true`) | Deliver **REQ-019** (customer cancel command — the real `refundsClientApi.cancelBooking` already targets `POST bookings/{id}/cancel`) + **REQ-020** (cancellation-policy preview → `GET bookings/{id}/cancellation_policy`, incl. the canonical `cancellation_policy_code` set) + **REQ-021** (`GET refunds/by_booking/{id}` + the decomposition fields on the customer `refunds/{id}/status`), then set flag `false` — the real client maps the published `refunds/{id}/status` 1:1 and targets the proposed slugs for the rest. No hook/component change | 🟡 |