Anchoring Bias
Early numbers or conclusions disproportionately pull later judgment — even when irrelevant
Anchoring bias is the tendency for an initial value, label, or conclusion to exert undue pull on later judgment — even when the anchor is arbitrary, irrelevant, or known to be unreliable. Tversky and Kahneman’s classic work showed that unrelated anchors (e.g. a spun wheel) shift estimates; in organisations the anchors are rarely random: they are first drafts, model scores, senior opinions, or a colleague’s prior review.
The failure is not disagreement. It is loss of independence: the second mind assesses agreement with the anchor, not the evidence afresh.
At the commitment boundary
In Strengthening the Commitment Boundary, anchoring is why blind peer review matters. A reviewer who sees the first conclusion before forming their own will anchor to it; override rates collapse; the second signature becomes rubber-stamp approval with extra latency.
Judgment infrastructure counters anchoring by:
- Blindness before independent assessment — blind peer review and separated model outputs before human bind.
- Override-rate telemetry — sustained near-zero overrides signal decorative review, not independent judgment.
- Explicit challenge records — AI stress testing and disclosed disagreement so anchors are surfaced, not hidden.
Related biases
Anchoring stacks with confirmation bias (seeking evidence for the anchored view) and with groupthink when the first voice in the room becomes the anchor for everyone else.
Corpus stance
A2 — working context: Standard cognitive-psychology construct; cited here for commitment-quality design, not as epistemic canon. See the applied pattern entry: anchoring bias (pattern).