Freeze the chat template date to preserve the reviewed inputs and baseline across daily runs. Weights, expected outputs, comparison rules and tolerances are unchanged. Promote after native full-fleet verification.
Use canonical model names for MIC baseline targets (#2)
