[rahuliitk]

File routes trusted the organization in the URL and accepted arbitrary storage buckets. They now verify organization access, restrict access to the configured evidence bucket, and require the exact organization key prefix. Questionnaire Auto-fill previously waited for all AI batches inside the HTTP request; it now returns 202 with a persisted job and runs in a background worker, with progress, saved answers and retry controls in the frontend.

Scope

• QT-001: retain existing JWT/configuration protections and add rejection regression tests; the reported issue was not reproduced locally.
• QT-002: fix file upload/download/delete authorization with mocked-storage regression tests. Preserve the explicit super-admin override.
• QT-003: add route and database search-isolation tests; no leakage reproduced in the tested search path.
• QT-004: add a PostgreSQL-backed queue with concurrent submission deduplication, exclusive worker claims, execution deadlines, interruption handling and preservation of saved answers. Prevent edits during processing and scope job status to the organization.
• Add progress polling, completion refresh and failure/retry UI, plus mobile navigation so questionnaire controls remain usable.
• Include schema reconciliation 0013, framework timestamp correction 0014, Docker development test dependencies and the frontend content-path correction.

Validation

• 29 targeted backend tests passed.
• Four frontend tests and TypeScript checking passed.
• PostgreSQL migration/seed regression tests passed; queue concurrency and saved-answer integration was rerun successfully against final head 0014.
• Playwright checks with mocked API responses covered desktop/mobile, progress after reload, completion, failure/retry controls and keyboard navigation. Screenshots reviewed.
• git diff --check passed.

Deployment and remaining checks

Deploy frontend and backend together: Auto-fill changes from a 200/message response to 202/job metadata. Apply migrations through 0014 first. Migration 0013 is forward-only; back up the database before applying it.

Real storage using two test accounts, real AI-provider behavior and production JWT configuration still require environment-specific verification. Local passing tests do not establish production readiness. See docs/qt-001-004-review.md for implementation details and repeatable test commands.