Compare local AI models for the explanation layer #408

As the AI-layer owner, I want a measured comparison of candidate local Ollama models for the explanation layer, so that "Warden" is a data-backed choice rather than an unexamined default.

Acceptance criteria

• At least two other locally-servable Ollama models are run against the same held-out set tools/explanation_quality.py already scores Warden against.
• Latency and approximate memory/resource footprint are measured for each candidate on real hardware, not estimated.
• A written comparison names which model is recommended and why, or confirms Warden stays, with the measurement as the reason rather than a preference.

Shape

AI layer, tooling and measurement only. No change to ai/explain.py's contract or to the F-1 finding shape.

Why this is worth doing

ai/Modelfile currently commits to one model with no comparison on record. #385 and #407 already built the held-out scoring infrastructure this would reuse directly, so the marginal cost of a real comparison is low.

Epic: Candidate idea (AI layer) - not yet agreed as a specialist pick
Size: not yet estimated
Assigned: @patelankeet2

Proposed 27 October 2026 as a candidate feature idea. Not yet discussed or agreed by the team; open for refinement or rejection at the next planning session, the same status every other idea in this batch carries.