v3.8.3

Changed

• Repinned gpt-rag-ingestion from v2.7.2 to v2.7.3, which stops the data-ingestion administrative surface from being mounted in deployments that did not ask for it. The admin SPA and every admin API route were mounted unconditionally at import time, and the DEPLOY_ADMINISTRATIVE_PANEL setting was never read, so the surface was reachable regardless of deployment mode. Mounting is now resolved once from App Configuration inside the ASGI lifespan and gated on the resolved mode. Classic deployments are unaffected — the admin surface keeps mounting exactly as before — and hosted deployments without a panel now correctly return 404. Tracked under #592, the umbrella issue for hosted deployment-mode validation.
• The same pin also hardens the /api/panel/* API: startup validation fails closed on missing Entra ID configuration rather than only on missing Cosmos configuration, and neither reading nor writing feedback returns 500 because of a single malformed Cosmos document.
• UI, orchestrator, and AI Landing Zone pins are unchanged from v3.8.2. No GPT-RAG configuration, parameter, or infrastructure change is involved.

Component versions

Component	Version
gpt-rag-ui	v2.6.2
gpt-rag-orchestrator	v4.1.1
gpt-rag-ingestion	v2.7.3
infra / AI Landing Zone	v2.5.1

Validation

• unit-tests green on the pinned ingestion commit: 228 passed, 0 failed, on both the source branch and the release branch.
• UI and orchestrator pins are byte-identical to v3.8.2 and were not revalidated.
• AI Landing Zone pins verified to agree: manifest.json ailz_tag, .gitmodules infra.branch, and the recorded infra/ submodule gitlink all identify v2.5.1 / 9cc5859a.