Make notifications action sweep wait for cold-load shell

docs/implplan/SPRINT_20260311_012_FE_live_notifications_harness_truthful_waits.md

@@ -0,0 +1,45 @@
+# Sprint 20260311_012 - FE Live Notifications Harness Truthful Waits
+
+## Topic & Scope
+- Remove the remaining cold-load false negatives from the live ops/policy Playwright sweep.
+- Wait for the real notifications operator shell before asserting `New channel` and `New rule` availability on `/ops/operations/notifications`.
+- Keep the harness truthful so page/action failures represent product defects rather than premature DOM checks.
+- Working directory: `src/Web/StellaOps.Web`.
+- Expected evidence: updated live sweep harness, clean rerun on the live stack, and a local QA-only commit.
+
+## Dependencies & Concurrency
+- Depends on the authenticated live stack already being healthy at `https://stella-ops.local`.
+- Safe parallelism: none while the live sweep is running because it mutates auth state and captures shared Playwright output.
+
+## Documentation Prerequisites
+- `AGENTS.md`
+- `docs/qa/feature-checks/FLOW.md`
+- `docs/implplan/SPRINT_20260309_002_FE_live_frontdoor_canonical_route_sweep.md`
+
+## Delivery Tracker
+
+### FE-NOTIFY-HARNESS-012-001 - Wait for the real notifications shell before checking actions
+Status: DONE
+Dependency: none
+Owners: QA, Developer
+Task description:
+- Live manual probes showed `/ops/operations/notifications` was healthy while the sweep still reported `New channel` and `New rule` as missing. The remaining gap was harness timing: the script checked for buttons before the operator notifications panel finished hydrating on cold loads.
+- Add a notifications-shell wait gate to the live ops/policy action sweep so the harness only evaluates those actions once the real operator surface is present.
+
+Completion criteria:
+- [x] The harness waits for the notifications shell before probing `New channel` and `New rule`.
+- [x] Live rerun confirms the previous notifications false negatives are gone.
+- [x] The scoped QA harness repair is committed locally.
+
+## Execution Log
+| Date (UTC) | Update | Owner |
+| --- | --- | --- |
+| 2026-03-11 | Sprint created after direct manual probes on `/ops/operations/notifications` showed the product surface was healthy while the live ops/policy sweep still flagged `New channel` and `New rule` as missing. | QA |
+| 2026-03-11 | Added a notifications-shell wait gate to `live-ops-policy-action-sweep.mjs`, reran the full live sweep, and confirmed `flow:New channel` plus `flow:New rule` now pass. The completed run ended with `failedActionCount=0` and `runtimeIssueCount=0`. | QA / Developer |
+
+## Decisions & Risks
+- Decision: treat the notifications miss as a harness defect, not a product defect, because manual probes already proved the buttons existed and worked on the live stack.
+- Risk: the live sweep writes shared Playwright artifacts under `src/Web/StellaOps.Web/output/playwright`; clear them before the next iteration so stale evidence does not bleed into later audits.
+
+## Next Checkpoints
+- Archive on local commit, then clear Playwright output and start the next full scratch rebuild iteration.