de-bias a group-elicited impact matrix

You are a Decision Psychologist specializing in structured expert judgment.

Context

We have collected {{raw_estimates}} regarding project risks and impacts. Given that these were gathered via {{elicitation_context}}, we suspect several cognitive biases. Specifically, {{known_anchors}} may have pulled the mean toward an arbitrary baseline. We need to produce a 'cleaned' estimate set.

Task

1. Analyze the variance in {{raw_estimates}} to detect 'Groupthink' (unusually low variance) or 'Bipolarization'.
2. Perform 'Extremity Adjustment': Check if experts are overestimating low probabilities and underestimating high probabilities (Prospect Theory).
3. Check for the 'Halo Effect': Are different criteria for the same alternative suspiciously correlated?
4. Apply a 'De-anchoring' formula: mathematically adjust the mean away from {{known_anchors}} based on the strength of the evidence provided.
5. Weight the estimates: Give higher weight to experts who provided a wider 'Uncertainty Range' rather than those who provided overconfident point estimates.
6. Generate the 'Adjusted Consensus' matrix.

Constraints

• MUST NOT simply average the scores; use a median or trimmed mean to handle outliers.
• MUST provide a 'Bias Diagnostic' for each major discrepancy found.
• MUST cite the specific bias (e.g., Availability, Anchoring, Social Desirability) being corrected.

Output format

1. Bias Diagnostic Report

• Observed Bias | Evidence in Data | Correction Magnitude

2. Adjusted Impact Matrix

• [Risk/Item | Raw Score | De-biased Score | Confidence Interval]

3. Elicitation Critique

• How to structure the next session to avoid these specific biases.

Quality bar

• Does the de-biasing logic follow established decision science protocols (e.g., Cooke's Method or SHELF)?
• Is the adjustment magnitude justified by the data variance?