Run a bias report

What this achieves

A bias report looks across a cycle’s submitted responses for rating patterns worth questioning — differences by department, or by the relationship a reviewer has to the person. It raises flags for a human to judge, and it does not decide anything.

Required role: Administrator.

Before you start

The report runs over responses that have been submitted. Run it on a cycle with enough responses to mean something; on a thin cycle it will find nothing and that absence proves nothing.

Steps

Go to Feedback & 1:1s → Cycles and open the cycle.
Select Bias Report.
Select Generate Bias Report.
Work through the Flags, reading the severity and how many people each affects.
Read Department Distribution and Rating by Relationship.
Read the Recommendations.
Select Mark as Reviewed on a flag you have investigated, adding notes on what you concluded.
Select Export to take the report away.

What the report shows

Option	Description
Flags	Detected patterns, each carrying a severity of high, medium, or low, and how many people it affects.
Department Distribution	Average rating by department.
Rating by Relationship	Mean rating from each reviewer relationship.
Recommendations	Suggested follow-up for what was found.

Rating by relationship is the pattern to read first

Reviewers of different kinds rate differently, and the table makes that visible. Managers rating consistently harder than peers, or direct reports rating consistently softer than everyone else, changes what a person’s overall score means depending on who happened to be asked.

Example: if HC Corp’s cycle shows peers averaging a point above managers, then two people with the same overall score are not comparable — the one with more peer reviewers scored higher for a reason that has nothing to do with their work.

Small departments identify people

Department distribution at HC Corp’s sizes needs care. Engineering at 34 is a group. Finance at 9 and People at 6 are a handful of identifiable individuals, and an average across six people alongside a named department is close to publishing their ratings.

Decide before you circulate this report which departments you will show broken out, and apply the same rule to the departments that look good as to the ones that do not.

No finding is not a clean bill

An empty flag list means the analyser did not detect a pattern in what was submitted. It does not mean the cycle was fair. Bias that every reviewer shares does not show up as a difference between reviewers.

The report also states on the page which analyses it did not generate for a cycle — read that line before treating the report as complete.

Marking a flag reviewed

Mark as Reviewed records that a flag was investigated, with optional notes on the conclusion. That note is the useful artefact: a flag reviewed and dismissed with a reason is a defensible decision, and a flag reviewed with no note is indistinguishable from one nobody looked at.

The interface warns that regenerating re-runs the analyser and that review state on individual flags may not survive it. Do the investigation and record the outcome before you regenerate, or take an Export first.

What happens next

The report stays with the cycle and can be exported. Where a flag points at a real pattern, the fix is in cycle design rather than in the report — the reviewer mix, the minimum peer count, or the calibration you run before ratings are finalised. Regenerate after more responses arrive if the cycle is still collecting.