docs(quick-261016-nqk): plan and summary for UAT gaps G-18.1-7 and G-18.1-4; D-21 amended, COVERAGE cancel_at row

.planning/phases/18.1-payment-gateway-subscriptions/18.1-CONTEXT.md

@@ -66,6 +66,7 @@ Research (`18.1-RESEARCH.md` § Premise Shifts S-1..S-10) found premises that no
 - **D-20 (S-8, S-9):** The Stripe subscriptions subpackage is its own party plugin, `golem15.pgstripesubscriptions`, at module path `git.golem15.com/golem15/sm-pgstripe-plugin/subscriptions`. It gets its own importable or package-local Stripe fake; P18's root `stripefake_test.go` stays unchanged.
 - **D-21:** Behaviour defaults: keep PHP's `cancel_at_period_end` mapping (a manual test-mode check decides whether `cancel_at` from the billing portal also needs handling); out-of-order events behave as PHP does; the job reads both pre-basil and basil payload shapes; customer `GET /subscription` returns the latest subscription with `has_access` and `in_grace_period` flags.
 - **D-21 amendment (2026-10-14, gap closure 18.1-07, user decision on verification human item 1):** out-of-order events no longer behave as PHP; created and updated events apply the subscription Stripe holds, serialised per subscription, and a deletion before its subscription is stored is retried. D-21's other defaults are unchanged.
+- **D-21 amendment (2026-10-16, quick 261016-nqk, UAT test 7 / gap G-18.1-7):** the test-mode check settled the `cancel_at` question. On API `2026-09-30.endive` a billing-portal cancel reports `cancel_at` (the period end) with `cancel_at_period_end` false; an API cancel reports both. pgstripe now stores `cancel_at_period_end` true when Stripe reports that flag or a `cancel_at` date on an active or trialing subscription; clearing `cancel_at` maps back to false and fires `reactivated`. The row keeps PHP's single flag; `cancel_at` itself is not stored.
 - **D-22 (plan split, 6 plans):** (1) framework: cabana flat reorder + `taglist`; (2) pgsub core tracer in the new repo (created with `tea repos create --owner golem15`); (3) pgstripe secret columns + import mapping + `subscriptions` plugin (gateway, both webhook routes, River job, host test); (4) pgsub admin screens + customer API; (5) `vue-paymentgateway-layer` pieces (tag `v0.2.0` only on go-ahead); (6) unit tests across all repos + `scripts/check-phase18.1.sh`. Plan 2's repo is committed before plan 3's `go.mod` points at it. No push, tag or deploy without the user's go-ahead.
 
 ### Claude's Discretion

.planning/phases/18.1-payment-gateway-subscriptions/COVERAGE.md

@@ -17,7 +17,7 @@ Scope: the Stripe Billing surface the frozen PHP `SubscriptionGateway` contract
 | subscriptions.retrieve (syncSubscription / reconcile) | INTEGRATE | also the webhook job's fresh-state read: created/updated apply Stripe's current subscription, 404 falls back to the snapshot (CR-01, 18.1-07); used by Service.SyncSubscription (WR-06) |
 | subscriptions.update cancel_at_period_end true (cancel at period end) | INTEGRATE | |
 | subscriptions.cancel (cancel immediately) | INTEGRATE | |
-| subscriptions.update cancel_at (flexible-billing cancel enums) | OPT-OUT | D-21 keeps PHP's cancel_at_period_end mapping; a manual Stripe test-mode check at UAT decides a follow-up |
+| subscription cancel_at (read from events and retrieval) | INTEGRATE | a cancel_at date on an active or trialing subscription maps to cancel_at_period_end true; cleared maps back (portal cancels report cancel_at only; D-21 amended 2026-10-16) |
 | subscriptions.update (resume or un-cancel through the API) | OPT-OUT | not in the frozen PHP contract: reactivation happens in the billing portal and arrives as customer.subscription.updated |
 | subscriptions.update (plan change, proration) | OPT-OUT | not in the frozen PHP contract: plan changes go through the billing portal |
 | subscriptions.list / subscriptions.search | OPT-OUT | not in the frozen PHP contract: reconcile walks local rows and retrieves each |