fix(ai): discard proposals when section context changes

docs/ui-modernisation/AI_CONTEXT_VALIDATION.md

@@ -0,0 +1,13 @@
+# A3-IM08 bounded stale-proposal repair
+
+3 November2026. IM08 is authorised and remains IN PROGRESS. This subpackage fixes a reproduced section-assistant defect without changing any model/provider/API/schema or acceptance writer.
+
+The shared assistant sends source text, targetId, document/output language, context and sectionName when generating a proposal. Its existing invalidation effect omitted the last two inputs. CvBuilderEditor entry callers include edited entry title/subtitle in context, so a user could change facts while a request was pending, then see and apply the previous suggestion; a displayed review also survived those changes. Four controlled regressions reproduced both pending/visible cases for context and section name before the fix.
+
+Added context and sectionName to the same invalidation effect. Existing requestId guards reject late success/failure/finally work and clear the old proposal. Generation remains explicit, Apply still requires server token acceptance, and parent source flush/revision/epoch guards remain unchanged. No silent content write or automatic regeneration. A translated section label may conservatively discard a prompt-bound proposal; it never changes document/output language to match UI language.
+
+Checks PASSED:23 section-assistant cases plus59 parent studio cases,82 disjoint cases/2 suites; typecheck, scoped zero-warning lint and production build. Independent static review found no actionable issue and did not repeat tests. Initial58-case Settings/AI baseline overlaps the section cases and is not added. Parent suite emitted existing direct-focus act warnings at cv-builder-deep-link.test.tsx475/782; all59 cases passed, but this is not a warning-free test run. Fixture-only callbacks/tokens, no live model or semantic AI-quality claim.
+
+NOT RUN for this one-line dependency repair:backend rerun, new browser sweep, physical PDF, native device, live models/providers, final integrated security/release. Existing rendered document layouts are unchanged; controlled React regressions exercise the affected async behavior directly. Other IM08 Settings/automation presentation and final scope qualification remain open; no Gate F acceptance is requested for this subpackage alone.
+
+Native baseline/red/green/parent reports and source manifest are in evidence/ai-context. Red run:4 failed,4 passed,15 intentionally excluded by test-name filter. Final complete section suite23 passed,zero failed/pending; parent59 passed. Rollback reverts the effect-dependency change; no database/data rollback.

docs/ui-modernisation/UI-08.md

@@ -0,0 +1,14 @@
+# A3-IM08 Contextual AI and Settings automation
+
+3 November2026. Owner explicitly approved **"Approve calendar/documents/mail and A3-IM08"**. Isolated codex/a3-ui-08 starts at accepted UI07 fba80ef346bc021e675024638dacf9ccef7243db. Acceptance receipt e5a979386d6377138e7ea7fd44958adb22838886 is published and independently remotely verified. No live model activation; application automation controls are distinct from the paused, untouched backlog orchestrator.
+
+- [x] Inspect SettingsView, rules/notification owners, AI privacy/usage and contextual section-assistant callers.
+- [x] Initial selected baseline:58 cases/5 suites passed, not blanket AI quality or full roadmap acceptance.
+- [ ] Contextual assistance: preserve manual editing, source/document-language separation, explicit review/apply, stale-input discard, quota/consent/failure/retry/close. Investigate context and section-name changes while a proposal is pending or under review.
+- [ ] Settings automation: expose current supported controls and explain effects/timing/disable behavior under Settings; preserve existing rules drafts and notification/AI preference writers. Inspect actual worker/data contracts before promising independent per-rule switches not represented in current six numeric thresholds. No fake controls, worker activation or policy/infrastructure change.
+- [ ] Scoped EN/NB, responsive/focus/accessibility presentation; types/lint/build, controlled browser and applicable local backend tests; independent review.
+- [ ] Publish validated checkpoints and exact evidence; update durable handoff. Next Gate F before IM09 account/session/connections/admin/processing views.
+
+Current source findings:rules expose only six integer day thresholds1–365, not independent enable flags; email-follow-up notifications and AI/privacy preferences have actual existing boolean controls. Do not represent thresholds as disable switches or imply a general worker-control API. Contextual assistant invalidation currently covers source text, target ID, language and entitlement; generation also consumes context and sectionName, which need compatibility review.
+
+No schema/dependency/migration planned for presentation and bounded stale-proposal fix. Existing canonical Career Profile, CV/letter history/settings, RD21, per-owner data, quota/consent and server review/apply stay authoritative. Rollback reverts this UI increment; no database rollback. Full integrated AI/security/native/provider/release qualifications remain separate.

docs/ui-modernisation/evidence/ai-context/CHECKS.json

@@ -0,0 +1,20 @@
+{
+  "frontendPassed": 82,
+  "suites": 2,
+  "failed": 0,
+  "pending": 0,
+  "typecheckExit": 0,
+  "scopedLintExit": 0,
+  "buildExit": 0,
+  "independentReview": "No actionable static findings; tests not repeated",
+  "warnings": [
+    "Parent direct-focus tests emit act warnings;59 passed"
+  ],
+  "notRun": [
+    "backend rerun",
+    "new browser matrix",
+    "physical PDF",
+    "live models/providers",
+    "final integrated security/release"
+  ]
+}

docs/ui-modernisation/evidence/ai-context/SOURCE_MANIFEST.json

@@ -0,0 +1,23 @@
+{
+  "base": "e5a979386d6377138e7ea7fd44958adb22838886",
+  "recordedAtUtc": "2026-11-03T14:24:45.724564+00:00",
+  "policy": "LF-normalised text SHA256",
+  "sources": {
+    "job-tracker-ui/src/components/cv/AiSectionAssistant.tsx": "40475ae87da5a1966bdd52a5107b4c54e34a28a4be31687dc5471217105c4d7a",
+    "job-tracker-ui/src/ai-section-assistant.test.tsx": "bcf642a670c5b2741ba8dd46fbf1b1f231c4801b4f1c8bebec6eab6dabd31396",
+    "job-tracker-ui/src/views/CvBuilderEditor.tsx": "49537d71b577df7c564609c2cfcd239afe066fe4b4c550677544886a205e0e0b",
+    "job-tracker-ui/src/cv-builder-deep-link.test.tsx": "8d5ae8d6a07991856dd2b21bde332ad91762c46f7534c2154719cf8a41d81dd2",
+    "job-tracker-ui/src/cvBuilder.ts": "c021bafeabbc8ad3526d1b10b90ad8f7f0958ded8ed26eac8c1d28b53098b64d",
+    "job-tracker-ui/package.json": "3745394e406cefa40d30a87e840bccb359be147bac6ffba736e2ae90898a8626",
+    "job-tracker-ui/package-lock.json": "bf31e90d7af03e8af059db2aaf893925b9e8881a30a1688034ed88ef45d4fcab",
+    "docs/ui-modernisation/UI-08.md": "a9077d5bdce58e8f23d818564cbdb6853392444b29ff19063562f621732c2361",
+    "docs/ui-modernisation/AI_CONTEXT_VALIDATION.md": "f934cdf1f43b3c50026ce692e1fce8e0a754ba6373ca5039b9414308998f5adf"
+  },
+  "evidence": {
+    "a3-ai-context-green.json": "0a587c3a3cd5235b4e36b805805fc9d5eb6be89ef87ff40516c3ad818939cfc4",
+    "a3-ai-context-parent-tests.json": "c6ca2b8ffd4f2150a19fa3ce94c4c35cd72ea08833cf28b671218930a2c2e3d8",
+    "a3-ai-context-red.json": "8536fc3228bb8fa7cb7735f7e470998ca8bfaa003ed6c91a9d0b54f7e27c0e71",
+    "a3-ai-settings-baseline.json": "47ad0bc2aaff394e233fd7a69a33fe17b13443318635af8c806a09d69854f3a4",
+    "CHECKS.json": "76564d88813d9fc8b798916db2391038d57b4d9d87b5f802d82fb068202c7279"
+  }
+}