npbuilds Bias Detector — The Mirror
Hold up a mirror to thinking and show the distortions. Cognitive biases are not character flaws — they're predictable patterns in how human cognition processes information under time pressure, complexity, and uncertainty. Identifying them is the first step to correcting for them.
Read references/bias-catalog.md for the complete catalog with examples and debiasing strategies.
Input
From the decision-theory director or directly:
- The decision, judgment, or reasoning to scan for biases
- Context: what was the situation? What information was available? What was the time pressure?
- Mode: comprehensive (scan for all major biases) or targeted (check for a specific suspected bias)
Process
Step 1 — Understand the Reasoning
Before scanning for biases, understand the reasoning being evaluated:
- What conclusion was reached (or is being reached)?
- What information was considered?
- What information was available but not considered?
- What was the process? (Deliberate analysis, gut feeling, group discussion, etc.)
Step 2 — Scan by Bias Category
Check each category systematically. Within each, focus on the biases most relevant to the specific situation.
Information Processing Biases
How we select, interpret, and weight information
| Bias | Pattern | Diagnostic Question |
|---|---|---|
| Confirmation bias | Seeking/favoring information that confirms existing beliefs | Was disconfirming evidence actively sought? |
| Anchoring | Over-relying on the first piece of information encountered | Would a different starting point yield a different estimate? |
| Availability heuristic | Judging probability by how easily examples come to mind | Are vivid/recent examples distorting frequency estimates? |
| Framing effect | Responding differently to the same information based on presentation | Would the conclusion change if the same facts were framed differently? |
| Base rate neglect | Ignoring population-level statistics in favor of specific cases | What's the base rate? Has it been considered? |
| Survivorship bias | Drawing conclusions only from visible successes | What about the cases that failed/disappeared? |
Decision-Making Biases
How we evaluate options and make choices
| Bias | Pattern | Diagnostic Question |
|---|---|---|
| Sunk cost fallacy | Continuing because of past investment rather than future value | If starting fresh today, would you make the same choice? |
| Status quo bias | Preferring the current state disproportionately | Is inaction being held to a lower standard than action? |
| Loss aversion | Weighting losses ~2x more than equivalent gains | Are potential losses being overweighted relative to potential gains? |
| Endowment effect | Overvaluing what you already have | Would you acquire this at the price you'd need to sell it? |
| Present bias | Overweighting immediate outcomes vs. future ones | How would this look from a year / five years out? |
| Omission bias | Preferring harm from inaction over equal harm from action | Is doing nothing really less harmful, or does it just feel that way? |
Social Biases
How others' behavior and opinions affect our thinking
| Bias | Pattern | Diagnostic Question |
|---|---|---|
| Groupthink | Conformity pressure suppressing dissent | Was there a genuine devil's advocate? Were dissenting views penalized? |
| Authority bias | Deferring to authority regardless of their domain expertise | Is the authority actually an expert in this specific area? |
| Bandwagon effect | Believing something because many others do | Would you hold this view if no one else did? |
| In-group bias | Favoring ideas/people from one's own group | Would this idea be evaluated differently if it came from an outsider? |
Retrospective Biases
How we evaluate past decisions and events
| Bias | Pattern | Diagnostic Question |
|---|---|---|
| Hindsight bias | "I knew it all along" after learning the outcome | What was actually knowable at the time of the decision? |
| Outcome bias | Judging decisions by results rather than process | Was this a good process that got unlucky, or a bad process that got lucky? |
| Attribution error | Attributing others' behavior to character, own to circumstances | Would you explain your own similar behavior the same way? |
| Narrative fallacy | Constructing a coherent story that oversimplifies causal chains | Is this narrative capturing the actual complexity, or smoothing it? |
Step 3 — Assess Impact
For each bias detected, assess:
| Dimension | Assessment |
|---|---|
| Confidence | How confident are you this bias is operating? [high/medium/low] |
| Direction | Which way is it pushing? [toward option X / away from option Y / inflating estimate / etc.] |
| Magnitude | How much is it distorting the judgment? [large/moderate/small] |
| Correctable | Can the user adjust for it? [yes, by doing X / partially / difficult to correct] |
Step 4 — Suggest Debiasing Strategies
For each significant bias, recommend a specific debiasing technique:
| Strategy | Works For | How |
|---|---|---|
| Consider the opposite | Confirmation bias, anchoring | Deliberately construct the case against your current view |
| Reference class forecasting | Planning fallacy, base rate neglect | Find similar past cases; use their distribution as your baseline |
| Pre-mortem | Overconfidence, planning fallacy | Imagine the decision failed; explain why |
| Red team | Groupthink, confirmation bias | Assign someone to argue against the consensus |
| Cooling period | Present bias, emotional reasoning | Delay the decision by a set time; revisit with fresh eyes |
| Outside view | Anchoring, narrative fallacy | Ask: what would a neutral advisor say about this situation? |
| Kill criteria | Sunk cost, status quo bias | Pre-commit to conditions under which you'll abandon the path |
Step 5 — Distinguish Bias from Legitimate Reasoning
Not every heuristic is a bias. Flag potential false positives:
- Gut feeling based on expertise: An experienced firefighter's "something's wrong" is pattern recognition, not a bias. Domain expertise produces valid intuitions.
- Reasonable risk aversion: Loss aversion in high-stakes irreversible situations may be rational, not biased.
- Prior beliefs based on evidence: Updating slowly on weak evidence isn't confirmation bias — it's appropriate Bayesian reasoning.
- Social information: Following expert consensus isn't authority bias if the experts are genuinely qualified in the relevant domain.
The test: Can the reasoning be justified on its own merits, independent of the psychological mechanism that produced it?
Output
BIAS SCAN
─────────
Subject: [the decision/judgment/reasoning analyzed]
Context: [situation, information available, time pressure]
Biases Detected:
1. [Bias name] — Category: [information/decision/social/retrospective]
Evidence: [specific feature of the reasoning that suggests this bias]
Confidence: [high/medium/low]
Direction: [how it's distorting the judgment]
Magnitude: [large/moderate/small]
Debiasing: [specific recommended strategy]
2. [Bias name] — Category: [category]
...
False Positive Check:
[Any heuristics that look like biases but may be legitimate reasoning]
Most Critical Bias: [the one distorting the judgment most]
Debiased Perspective:
[What the judgment/decision looks like after accounting for detected biases]
Caveat: [Limitations — e.g., "This scan was based on the information provided; biases in the information itself may not be detectable"]
Error Handling
Everything looks biased: Step back. If every reasoning pattern triggers a bias label, the scan is too sensitive. Focus on biases with high confidence and large magnitude. Some apparent biases are just good heuristics.
No biases detected: Possible but uncommon for complex decisions. Report honestly, but note that some biases (especially confirmation bias) are hard to detect from the inside. Suggest the user seek an external perspective.
User is defensive about biases: Normalize: "Everyone has these biases — they're features of human cognition, not personal failings. The goal is awareness, not judgment." Focus on the debiasing strategies, not the diagnosis.
Retrospective analysis (was the past decision biased?): Apply hindsight-bias correction first. Evaluate the reasoning based on what was knowable at the time, not on what's known now. A decision can be unbiased and still produce a bad outcome.
Related Skills
- behavioral-game-theory — Cognitive biases (anchoring, framing, loss aversion) catalogued here recur as strategic patterns in behavioral-game-theory. Use behavioral-game-theory when biases interact across multiple agents.