Anchoring Bias
Prior conclusions anchor later review — independence is lost before the second bind
Bias mechanism
The first number, score, or conclusion becomes the reference point; subsequent reviewers adjust from it instead of assessing evidence independently.
Cognitive biasCoordination as architectureLocal optimisation
Anchoring bias at the commitment boundary is what turns “two-person review” into one judgment twice. The first reviewer’s conclusion — or the model score shown above the approve button — becomes the anchor. The second reviewer asks do I agree? not what does the evidence say?
Term entry: anchoring bias.
What it produces
- Rubber-stamp approval — override rates near zero because the second actor ratifies the anchor.
- Compliance theatre — workflows that look like dual control but preserve the anchor (first score visible, first review in the ticket subject line, ensemble score pre-merged).
Counter direction
- Blind peer review — second reviewer sees the same evidence, not the first conclusion, before forming their own.
- Disclosed model disagreement — separate model outputs before aggregation; disagreement is signal, not noise to collapse.
- Override-rate monitoring — Article 19 Rule 3 and Article 20: track how often humans and second reviewers actually change outcomes.