## Changes - Notebook covering the full calibration workflow: QA dataset → generate answers (`LiteLLMBackend`) → rate answers → fit detector → push weights to HF Hub via skops ## Acceptance criteria - Notebook executes cleanly via `uv run jupyter nbconvert --to notebook --execute` with a small synthetic dataset (no large model required) - Shows how to register a new model in the Registry
Changes
LiteLLMBackend) → rate answers → fit detector → push weights to HF Hub via skopsAcceptance criteria
uv run jupyter nbconvert --to notebook --executewith a small synthetic dataset (no large model required)