Files
jobtrackingapp/docs/ui-modernisation/INTERVIEW_AI_CONTEXT.md
T

2.8 KiB

Interview AI application-context lifetime

Baseline A3 348f4684, isolated branch codex/interview-ai-context at E:/Storage/Projects/Active/temp docs/jobjakt-local-build/next-security-patch. Source 0f8b17a. Uses the qualified locked Next16.4.6/follow-redirects1.16.3 installation. No live model.

Three deterministic tests failed on original runtime source: a late generation displayed application7 text under application8; a completed suggestion remained available after changing application; a late accepted write called the former board loader again after navigation. Synthetic model/API responses only. The first two are also a risk of offering the wrong application's content for explicit application8 acceptance; no real misdirected write is claimed.

The sole assistant caller now keys its instance by jobId, scoping instructions, generated-language draft, errors, busy state and suggestions to that application instance. Existing CV reader cleanup remains. A layout-lifetime ref invalidates on committed unmount; an already accepted write can finish at its original target, but cannot invoke the old board-refresh callback afterward. No write is redirected, automatically retried or newly authorised. Current-application explicit accept/reject and normal refresh still use existing paths. UI language and generated-content language remain independent within the current application.

PASSED: final55 cases/two suites (51 InterviewPrep/four daily connected route cases), non-incremental TypeScript, scoped two-file lint and source diff. Three red failures are included in55;48 skipped cases in the focused red selection are not final-suite skips. An earlier successful run used a passive cleanup; final55 was rerun after using synchronous layout cleanup and is the qualification result. Existing generation, explicit acceptance, failed-add retry, answer drafts, CV retry and follow-up cases remain passing. Receipts bind final source and readable/native evidence.

PASSED: final standard Next16.4.6 build on exact new source after the55-case run; compile, TypeScript and static export completed exit0. No warning bypass. This is current isolated E-drive export evidence, not the older C-drive export and not a fresh browser run.

NOT RUN: fresh full frontend/backend regression, browser, native device/AT, actual model/semantic quality, provider, security/recovery/release qualification. Parent preparation/follow-up outstanding reads, broader job/session lifetime, AI live fallback copy and other components remain separate proof-required work; this is not whole-application isolation acceptance. Existing PDF metrics/visual acceptance not reopened. No catalogue/schema/renderer/ownership/revision/provider changes, orchestrator, main merge or production action.