Lab 02: compare routing decisions

Prerequisites: Lab 01, both configured models downloaded, and routing enabled. Artifact: a three-row routing comparison.

Exercise

1. Follow Choose the answering model to preview a coding request.
2. Preview a general explanation request by changing the URL-encoded query parameter.
3. Send both requests to /api/v1/query without a model override.
4. Send the coding request with an explicit qwen2.5-coder:3b override.
5. Record preview model, actual model, reasoning, and whether the selected model was available.

Request	Preview	Actual model	Explanation
Coding, automatic	Fill from response	Fill from response	Record evidence
Explanation, automatic	Fill from response	Fill from response	Record evidence
Coding, explicit model	Not applicable	Fill from response	Check availability

Acceptance criteria

You can distinguish rule scoring from runtime availability checks. You inspect the actual response rather than assuming that explicit selection always succeeds. A routing difference is an observation to investigate, not an automatic test pass.

If results differ

Refresh the registry after pulling models. If routing is disabled, restart the backend with the intended setting. The preview and actual route use different inputs, so document differences before diagnosing a regression.

Next: Lab 03: retrieval.