cognitive-bias-audit

Audits a decision, plan, forecast, or analysis for cognitive biases. Biases are systematic, not random: the same pressures bend reasoning the same way every time, which is why you cannot introspect them away. Find where this specific decision is exposed and change the process that produced it, rather than labelling biases generically.

The bias families, each with its tell (how it shows up in actual reasoning) and a countermeasure, plus the debiasing toolkit, live in references/biases.md. Load that file before auditing and work the decision against each family exactly.

When to use

The user is about to commit to a high-stakes decision or forecast, or wants to pressure-test an analysis they or someone else produced. The unit of analysis is one decision and the reasoning behind it. Distinguish from a pre-mortem, which imagines the decision has failed and works backward from the failure; this skill inspects the reasoning as it stands now. Distinguish from a weighted-decision-matrix, which scores options against criteria; this audits how the choice between them is being reasoned about. A bias audit often prescribes those two as countermeasures.

Language

Conduct the session and write the report in the language the user is writing in. The reference file is English; keep the framework keyed to it regardless of the output language.

Inputs

• DECISION (required): the decision, plan, forecast, or claim under audit. If only a topic is given, pull out the actual call being made before starting.
• REASONING (required): the reasoning and evidence behind it, the key claims, and how the conclusion was reached. Without the reasoning there is nothing to audit; if it is missing, ask for it.
• STAKES (optional): what rides on the decision and what being wrong would cost. Sharpens the severity ranking, since a bias matters in proportion to how much it could distort a consequential call.