Cognitive BiasesPublic knowledge library

For AI assistants and agents

A decision review should expose uncertainty, not invent a diagnosis.

The Decision Review contract gives assistants a small structured output for a real decision: observed facts, possible reasoning risks, missing evidence, questions, practical techniques, the next action and canonical references.

JSON Schema 8 worked examples

Output contract

Keep facts and inference separate.

Required fields:

Candidate lenses use only low or medium confidence. The format intentionally has no high-confidence psychological diagnosis.

Agent rules

What a useful review should do.

  1. Restate the decision without choosing for the user.
  2. Put only supplied or verified information under observed facts.
  3. Describe cognitive-bias lenses as possible reasoning risks, not traits of a person.
  4. Ask for evidence that could change the conclusion.
  5. Recommend a concrete technique when it fits the situation.
  6. Preserve uncertainty and cite canonical project records.
  7. Return no lens when the evidence does not support one.

Worked cases

Eight examples across real decisions.

vendor-tool-selectionWe should migrate to technology X because several competitors are using it.

Next move: Define the criteria and run the same evidence check on technology X and one serious alternative.

View structured example →
product-prioritizationThe project is late, but we have already spent too much to stop now.

Next move: Recalculate the decision using future costs, future value and explicit continuation criteria.

View structured example →
ai-assisted-researchThe AI says the migration should take 40 days. Help me refine that estimate.

Next move: Create an independent estimate, then trace the AI estimate to evidence before updating the range.

View structured example →
production-incidentAfter the outage it is obvious that the team should have known this dependency would fail.

Next move: Reconstruct the timeline using only information available at each decision point before changing the process.

View structured example →
checking-repeated-claimsI have seen the same claim on six sites, so it is probably well supported.

Next move: Trace the six mentions to their original sources and count independent evidence rather than repetitions.

View structured example →
career-changeI have spent twelve years in this field, so changing direction now would waste all that time.

Next move: Compare the next twelve months of each option and design one small test of the strongest alternative.

View structured example →
performance-reviewThe employee missed the last delivery, so the annual performance rating should be low.

Next move: Review the full period against predefined criteria before assigning the rating.

View structured example →
forecastingGive me one exact launch date. We need a number for the plan.

Next move: Publish a forecast range with assumptions and an update trigger before choosing a planning date inside it.

View structured example →

Example JSON

Machine-readable and inspectable.

{
  "schemaVersion": "1.0",
  "decision": "Whether to migrate to technology X.",
  "situation": "vendor-tool-selection",
  "observedFacts": [
    "Several competitors are reported to use technology X."
  ],
  "inferredRisks": [
    {
      "lensId": "cognitive-bias-confirmation-bias",
      "rationale": "Competitor adoption may be used mainly as support for an option that is already preferred.",
      "confidence": "low"
    },
    {
      "lensId": "cognitive-bias-anchoring-effect",
      "rationale": "The competitor choice may become an early reference point before our own criteria are fixed.",
      "confidence": "low"
    }
  ],
  "missingEvidence": [
    "Our required capabilities and constraints.",
    "Evidence that competitor conditions are comparable to ours."
  ],
  "questions": [
    "Which criteria were fixed before looking at competitors?",
    "What evidence would make us reject technology X?"
  ],
  "techniques": [
    {
      "id": "predefined-criteria",
      "why": "Compare options against the same requirements."
    },
    {
      "id": "red-team-review",
      "why": "Build the strongest evidence-based case against the preferred migration."
    }
  ],
  "nextAction": "Define the criteria and run the same evidence check on technology X and one serious alternative.",
  "uncertainty": "Competitor adoption alone does not show whether technology X fits this organisation.",
  "canonicalReferences": [
    {
      "type": "situation",
      "id": "vendor-tool-selection",
      "url": "https://cognitive-biases.github.io/situations/vendor-tool-selection/"
    },
    {
      "type": "bias",
      "id": "cognitive-bias-confirmation-bias",
      "url": "https://cognitive-biases.github.io/biases/cognitive-bias-confirmation-bias/"
    },
    {
      "type": "technique",
      "id": "predefined-criteria",
      "url": "https://cognitive-biases.github.io/techniques/predefined-criteria/"
    }
  ]
}

Browse situations Browse techniques