[AW Top 10] 04 Prevent Pi engine runs on chat-incompatible models #65772

Priority 4/10 | 5 source issues | Impact 4/5 | Confidence 4/5 | Effort 3/5

One assignment, one coherent fix

Pi-engine workflows fail because the selected model is not accessible through the chat completions endpoint and the proxy returns an unsupported_api_for_model error. Every provider request fails before any safe output is emitted.

Implementation scope

Add compile-time or preflight validation in the Pi provider path that checks the selected model against the endpoint capabilities from the proxy reflection data, and fall back to a supported endpoint or fail with an actionable message naming the model.

Done when

• A unit test shows a model flagged as responses-only is rejected or rerouted before the first inference request.
• The affected Pi-based smoke and daily workflows no longer fail with unsupported_api_for_model.

Why now

Five incidents across different workflows and models share the same HTTP 400 code, indicating a generic gap in model-to-endpoint validation rather than per-workflow errors.

Unchanged AW sources close only after this summary is completed. Newer source activity and not-planned retirement do not trigger source closure. Assigned summaries are frozen; unassign to allow reclustering.

Generated by AW Essential Issue Clustering · copilot · auto · 35.6 AIC · ⌖ 12.1 AIC · ⊞ 8.6K · ◷