test: record integrated navigation recovery qualification

The combined mainline recovery and narrow-screen repairs need independent integrated evidence after recent merges. Retain eight passing isolated Chromium results and representative screenshots, with explicit limits so source-only navigation success is not mistaken for candidate or notification-delivery qualification.

Change-source: pulse-maintainer

tests/qualification/navigation-reconnect/2026-09-05-integrated/README.md

@@ -0,0 +1,67 @@
+# Integrated navigation recovery check — 5 September 2026
+
+## Result
+
+At source `c8931787adb7bc3152f060a829ebe2a1aa2a2b9c`, all eight existing
+`96-navigation-socket-recovery.spec.ts` Chromium cases passed (132 seconds,
+starting 20:43:55 UTC). No application or test changes were made for this run.
+The backend was built locally with embedded production frontend assets and
+synthetic inventory; it was stopped by the runner afterwards.
+
+Local executable SHA-256:
+`a5aaf32fdba70744d1dfa4e0749f080dd8960dcd19d381ab6ec2fdc286d36b27`.
+This identifies this source-built test executable, not a published artifact.
+Locked dependencies were installed in root, frontend-modern and tests/integration;
+the runner used Playwright 1.56.1 and the Chromium project.
+
+Widths 1440 and 1100 (height 900), and 390 and 320 (height 844), each passed
+with and without injected admission HTTP failure. The checks exercise socket
+closure with 1013, retained navigation/inventory, recovery without document
+reload, and incident-control access at desktop/mobile widths. Narrow cases
+also exercise platform switching, More navigation, and Docker table keyboard
+access. Admission-failure cases require an observed failed request.
+
+`junit.xml` is the unmodified runner result. Two inspected screenshots retain
+representative desktop incident access during reconnect and 320px inventory.
+The narrow screenshot alone is not proof of connection state; the test asserts
+that state separately. Small update cells still wrap heavily at 320px, despite
+passing the existing no-clipped-text checks. No readability redesign is implied.
+
+Additional focused checks: 53 tests passed across websocket-resilience and
+websocket-unified; `TestAlertCharacterizationGetActiveAlertsExportsCanonicalIdentity`
+passed with `-count=1` in internal/alerts. No full repository suite was run.
+
+## Repeat
+
+From repository root after installing locked dependencies:
+
+```sh
+pulse-heavy-run -- env \
+  PULSE_E2E_USE_LOCAL_BACKEND=1 PULSE_E2E_SKIP_PLAYWRIGHT_INSTALL=1 \
+  PULSE_MOCK_MODE=true PULSE_E2E_NAVIGATION_RECOVERY=1 \
+  PULSE_E2E_LOCAL_BACKEND_PORT=18765 \
+  npm --prefix tests/integration test -- \
+  tests/96-navigation-socket-recovery.spec.ts --project=chromium
+```
+
+## Decision and limits
+
+Fresh retrieval of the [community comparison thread](https://www.reddit.com/r/Proxmox/comments/1lblkk8/anyone_else_switch_to_pulse_from_netdata_or_any/)
+on 5 September reinforced protecting quick-glance, low-overhead monitoring.
+This is historical, anecdotal feedback, not a new independent demand count or
+benchmark. The latest integrated source warranted checking the existing repairs
+together rather than creating another surface or repeating a withdrawn repair.
+
+This result narrows the integration regression uncertainty. It does **not**
+qualify acknowledgement persistence, off-host delivery, restart history,
+malformed alert snapshots, a reporter installation, or a release candidate.
+The REST recovery code still skips entries without usable IDs. The server's
+normal read model exports typed alerts and canonical identity has focused test
+coverage, but that is not proof every malformed-response path is impossible.
+Do not count that previously withdrawn repair as delivered. Revisit it with
+browser fault injection, genuine empty-response clearing, and the required
+same-content performance/contracts qualification if pursued.
+
+Next useful evidence is exact-candidate recovery and installed alert receipt,
+not another equivalent source-only navigation run without changed inputs.
+No release readiness or backport eligibility judgment is made here.

tests/qualification/navigation-reconnect/2026-09-05-integrated/junit.xml

@@ -0,0 +1,20 @@
+<testsuites id="" name="" tests="8" failures="0" skipped="0" errors="0" time="132.169269">
+<testsuite name="96-navigation-socket-recovery.spec.ts" timestamp="2026-09-05T20:43:55.759Z" hostname="chromium" tests="8" failures="0" skipped="0" time="131.012" errors="0">
+<testcase name="populated navigation survives socket loss at 1440px (admission failure: false)" classname="96-navigation-socket-recovery.spec.ts" time="9.231">
+</testcase>
+<testcase name="populated navigation survives socket loss at 1100px (admission failure: false)" classname="96-navigation-socket-recovery.spec.ts" time="7.962">
+</testcase>
+<testcase name="populated navigation survives socket loss at 390px (admission failure: false)" classname="96-navigation-socket-recovery.spec.ts" time="18.039">
+</testcase>
+<testcase name="populated navigation survives socket loss at 320px (admission failure: false)" classname="96-navigation-socket-recovery.spec.ts" time="16.376">
+</testcase>
+<testcase name="populated navigation survives socket loss at 1440px (admission failure: true)" classname="96-navigation-socket-recovery.spec.ts" time="11.781">
+</testcase>
+<testcase name="populated navigation survives socket loss at 1100px (admission failure: true)" classname="96-navigation-socket-recovery.spec.ts" time="24.992">
+</testcase>
+<testcase name="populated navigation survives socket loss at 390px (admission failure: true)" classname="96-navigation-socket-recovery.spec.ts" time="23.496">
+</testcase>
+<testcase name="populated navigation survives socket loss at 320px (admission failure: true)" classname="96-navigation-socket-recovery.spec.ts" time="19.135">
+</testcase>
+</testsuite>
+</testsuites>