RL: response comparison to accepted evidence

00:00:00.000 — Compare the two responses
One response retains the compliance hold. The other would tender with approval still pending. Current telemetry satisfies only one condition.

00:00:09.926 — Inspect the source behind the judgment
Expand the retained source item, source digest and rubric version. The comparison stays bound to the evidence the reviewers judged.

00:00:21.758 — Preserve the improvement evidence
The accepted preference remains connected to the exact source and rubric. It does not train a model or publish an operating release.