diff --git a/controllers/adminController.go b/controllers/adminController.go
index 609ba1f..b47fbb4 100644
--- a/controllers/adminController.go
+++ b/controllers/adminController.go
@@ -2116,8 +2116,26 @@ func GetAdminBookings(c *fiber.Ctx) error {
 		return utils.Internal(c, "failed to count bookings")
 	}
 
+	// Newest first, and deterministically so.
+	//
+	// Without an ORDER BY the row order is unspecified — Postgres returns heap
+	// order, which in practice is oldest first. Two things follow, and both bit:
+	//
+	//   1. The newest booking sits on the LAST page. The console drains a
+	//      bounded window, so once pickupbookings outgrows that window a
+	//      just-created customer-app booking can never reach the Orders page at
+	//      all. It is written correctly and is simply never fetched.
+	//   2. OFFSET pagination over an unordered result is not stable: the same
+	//      page can return different rows across two requests, so draining pages
+	//      can duplicate and skip rows well before that threshold.
+	//
+	// bookingid rather than createdat: it is the primary key and unique, so the
+	// sort needs no tiebreaker and the paging cannot wobble between equal
+	// timestamps. It also matches the order the console already sorts into
+	// client-side, so page 1 is the newest page by both definitions.
 	var bookings []models.PickupBooking
 	if err := query.Preload("Parcels").Preload("ServiceOptions").
+		Order("bookingid DESC").
 		Offset(offset).Limit(pagesize).Find(&bookings).Error; err != nil {
 		return utils.Internal(c, "failed to fetch bookings")
 	}
diff --git a/docs/express-console-api.md b/docs/express-console-api.md
index 10e4e69..152540c 100644
--- a/docs/express-console-api.md
+++ b/docs/express-console-api.md
@@ -251,10 +251,25 @@ delivered, riderkms, ridercharges, dutyminutes }`.
 
 | Method | Path | Notes |
 |---|---|---|
-| GET | `/admin/bookings` | tenant-scoped list |
+| GET | `/admin/bookings` | tenant-scoped list, **newest first** (`bookingid DESC`) |
 | POST | `/admin/expressbooking` | create one — passes CityGate |
 | POST | `/admin/expressbooking/bulk` | `{ "bookings": [ ... ] }`, max 200, per-row results |
 | GET | `/admin/bookings/:id` | 404 if outside your tenant |
+
+`GET /admin/bookings` is ordered `bookingid DESC` — page 1 is always the newest
+page. The order is guaranteed, not incidental: `bookingid` is the primary key
+and unique, so `OFFSET` paging over it is stable and two requests for the same
+page return the same rows. Before this was explicit the result order was
+unspecified (Postgres heap order, in practice oldest first), which put the
+newest booking on the LAST page — a client draining a bounded number of pages
+could never reach a just-created booking.
+
+It also accepts `?status=` as an EXACT, case-sensitive, single-value match
+against the stored enum. The stored values are capitalised
+(`Pending_Pickup`, `Converted_To_Consignment`); `?status=pending_pickup`
+matches nothing and returns an empty list rather than an error. There is no
+multi-status or date filter, so a client grouping several statuses into one tab
+still has to filter its own rows.
 | GET | `/admin/bookings/:id/track` | **the tracking screen** — see below |
 | POST | `/admin/bookings/:id/assign-miler` | |
 | POST | `/admin/bookings/:id/assign-vehicle` | |
@@ -393,7 +408,12 @@ rather than making the console reconcile two stores.
 - **Envelope**: `{ "success": true, "data": ... }` on success,
   `{ "success": false, "message": "..." }` on failure. Lists add `total`,
   paginated lists add `page`.
-- **Pagination**: `?pageno=1&pagesize=100`. Default 500, cap 1000.
+- **Pagination**: `?pageno=1&pagesize=100`. Default 20, cap 100. Both numbers
+  are enforced server-side (`min(100, max(1, pagesize))`); a request for a
+  larger page silently receives 100 rows and the response's own `pagesize`
+  field reports 100. This entry previously read "default 500, cap 1000", which
+  was wrong on both counts — clients sizing a page budget off it fetched a
+  twelfth of what they expected.
 - **Rate limits**: 300/min per IP globally, 10/min shared across all credential
   endpoints. Behind the ingress this keys on the proxy IP unless
   `TRUSTED_PROXIES` is set.