End-to-end test pass vs a real Azure AI Foundry resource

[jongalloway]

Summary

Part of the converged workshop rewrite (milestone #1). The critical quality gate: every hands-on part must be run end-to-end against a real Azure AI Foundry resource before July 27. This is the primary timeline risk — budget for it explicitly.

Content

• Fresh-machine run of the pre-work checklist (Add mandatory pre-workshop Azure AI Foundry setup checklist #493) → provision Foundry + gpt-4.1-mini + text-embedding-3-small
• Part 2 console chat — runs, streams, maintains history
• Part 3 manual RAG — embed/search/augment produces grounded answers
• Part 4 template reveal — aichatweb scaffolds, Aspire + Qdrant run, chat works
• Part 5 providers — verify Azure primary; spot-check Foundry Local / Ollama / GitHub Models (legacy, pre-retirement)
• Optional: Products Page bonus still works
• Deployment + MCP parts unaffected by renumber
• Capture any doc fixes discovered during the run

Acceptance criteria

• A learner can complete Parts 1 → template reveal top-to-bottom on a clean machine using only Azure AI Foundry.
• Offline path (Ollama, and Foundry Local per Spike: verify Foundry Local embedding-model support for RAG #496) verified for at least the chat loop.

[jongalloway]

Add a security check to the test pass: verify no hardcoded API keys or connection strings anywhere in the workshop code — all credentials must come from user secrets / secure configuration via IConfiguration. This is an explicit acceptance gate (hardcoded keys are a common lab/sample mistake we must not ship).

[jongalloway]

Status update (2026-09-11): the July end-to-end report (#576) verified Parts 1-9 against live Microsoft Foundry and found no remaining failing parts, but Parts 10-11 were only config-tested and �zd up was not run. The original pre-July-27 deadline has therefore passed, but the full acceptance criteria are not completely closed; leaving this issue open until a real-resource deployment pass is recorded.