Use AI to expand research without treating fluent text, repeated output or model confidence as verification.
Open situation →Reasoning Practice · Intermediate
The second reviewer can see the first score
An AI evaluator is asked to independently rescore a report, but the prompt includes the previous model's score and comments. The team treats the second score as independent confirmation.
Your decision
How should the second evaluation be designed?
Show the best first move
A. Run a blind evaluation without the previous score, then compare it with a history-aware review.
Why this move helps
If a previous evaluation remains visible, it can become part of the new judgment. A blind pass gives you a cleaner test of whether two evaluations are genuinely independent.
Evidence still missing
- A blind second score.
- The history-aware score from the same material.
- Whether the difference is stable across repeated evaluations.
Evidence note
A 2026 LLM-as-a-judge study found that prior scores in context could shift later evaluations. This is provisional model-specific evidence, not proof that every evaluator will anchor in every task.
Next action
Create a blind-versus-history-aware evaluation pair and compare the score difference.
Best first lens
Anchoring Effect
Use the lens to ask a better question. It does not prove that a person or team has a cognitive bias.
Other defensible lenses
The situation may have more than one explanation.
Continue the path
Situation → skill → technique → action.
You can use an AI system to expand reasoning while keeping important judgments independently checkable.
Build this skill →Create a reference point before seeing a strong external number or recommendation.
Run the technique →