Published October 22, 2026.

Select a sample that represents the work

Include conversations from each channel, tenure group, and complexity level. Exclude cases still under investigation unless the review is about the process.

Record the sampling rule and the date so the next calibration can be compared.

Score the same rubric independently

Reviewers score the sample before discussing it. Record each score and the evidence. The conversation starts with the gaps, not the average.

If reviewers disagree, update the rubric or the example, not the agent score.

Publish the outcome

Share the agreed standard and one or two examples with the team. Note any rubric change and the effective date.

Repeat the calibration on a regular cadence. A one-time session does not create consistency.

Source: reference.

Related guides: this workflow and this workflow.