# Cognitive Biases

> A public knowledge library about cognitive biases, evidence, decision skills and better decision making.

Use the human-readable pages when explaining a concept or skill to a person. Use the public data files when a structured record is more useful.

- Canonical website: https://cognitive-biases.github.io/
- Project history: https://cognitive-biases.github.io/history/
- Machine-readable history: https://cognitive-biases.github.io/history.json
- Agent locale manifest: https://cognitive-biases.github.io/ai/locales.json
- German agent routing: https://cognitive-biases.github.io/de/llms.txt
- Russian agent routing: https://cognitive-biases.github.io/ru/llms.txt
- AI Search & Citation Profile: https://cognitive-biases.github.io/ai/ai-search-profile.json
- Canonical citation index: https://cognitive-biases.github.io/citation-index.json
- Trust Center: https://cognitive-biases.github.io/trust/
- Corrections ledger: https://cognitive-biases.github.io/trust/corrections.json
- Knowledge graph: https://cognitive-biases.github.io/knowledge/graph.json

- Bias library: https://cognitive-biases.github.io/explore/
- Everyday life: https://cognitive-biases.github.io/everyday/ — short, practical guides that start with familiar situations and connect them to reviewed decision lenses.
- Decision Lens Packs: https://cognitive-biases.github.io/lenses/ — task-based workflows for career decisions, AI-assisted work, teams, product/project decisions and information checking.
- Professional Lens Packs: https://cognitive-biases.github.io/professional/ — reusable work-focused packs for hiring, negotiation, leadership and performance decisions.
- Career Decision Review: https://cognitive-biases.github.io/tools/career-decision-review/ — local-first worksheet for comparing career options using independent estimates, counterevidence and reversible tests.
- Lens Pack Builder: https://cognitive-biases.github.io/tools/lens-pack-builder/ — local-first tool for assembling a compact task-specific checklist from published decision lenses.
- AI-era Bias Lab: https://cognitive-biases.github.io/ai-era/ — research tracks for human–AI interaction patterns, with explicit status, related established concepts and falsifiable predictions.
- AI-era experiment protocol: https://cognitive-biases.github.io/ai-era/protocol/ — reusable comparison protocols for testing human–AI decision patterns without promoting working labels to established biases.
- AI-era Research Tracker: https://cognitive-biases.github.io/ai-era/tracker/ — maturity tracker from idea to protocol, experiment, result and replication; stages describe project artifacts, not scientific certainty.
- First preregistered AI-era study: https://cognitive-biases.github.io/research/ai-advice-order-v1/ — preregistered pilot testing whether standardized AI-labelled numerical advice has more influence when shown before an independent estimate.
- AI Advice Order participant instrument: https://cognitive-biases.github.io/experiments/ai-advice-order-v1/ — six-trial local-only instrument; responses are not uploaded automatically.
- Experiments Lab: https://cognitive-biases.github.io/experiments/ — interactive, local-only demonstrations that change one variable and connect the manipulation to an evidence-reviewed concept.
- AI Bias Benchmark: https://cognitive-biases.github.io/ai-benchmark/ — provider-neutral paired-prompt specification for measuring AI model sensitivity to controlled decision conditions.
- Cognitive Bias Observatory: https://cognitive-biases.github.io/observatory/ — public snapshots of headline-level framing, certainty, salience, authority, social-proof, numerical and repetition signals with explicit sampling limits.
- Observatory Trends: https://cognitive-biases.github.io/observatory/trends/ — normalized time series across repeated comparable snapshots with minimum-history rules before a direction is reported.
- Observatory source comparisons: https://cognitive-biases.github.io/observatory/sources/ — descriptive source-level rates published only after a minimum observation threshold.
- Observatory Research Briefs: https://cognitive-biases.github.io/research/observatory/ — evidence-gated longitudinal summaries generated only when comparable Observatory history meets public readiness thresholds.
- Observatory methodology: https://cognitive-biases.github.io/observatory/methodology/
- Evidence reviews: https://cognitive-biases.github.io/evidence/
- Methodology: https://cognitive-biases.github.io/methodology/
- Quality status: https://cognitive-biases.github.io/quality/
- Research: https://cognitive-biases.github.io/research/
- Decision guides: https://cognitive-biases.github.io/contexts/ — start here when the situation is known but the bias name is not.
- Decision skills: https://cognitive-biases.github.io/skills/ — start here when the goal is to build a practical capability such as evidence evaluation, forecasting or AI-assisted reasoning.
- Practice Lab: https://cognitive-biases.github.io/practice/ — short evidence-linked exercises for connecting practical checks with reviewed lenses.
- Decision Audit: https://cognitive-biases.github.io/tools/decision-audit/
- Public data guide: https://cognitive-biases.github.io/data/
- Data catalogue: https://cognitive-biases.github.io/data/catalog.json
- Release manifest and checksums: https://cognitive-biases.github.io/data/manifest.json
- Bias data: https://cognitive-biases.github.io/data/biases.json
- Consolidated evidence data: https://cognitive-biases.github.io/data/evidence.json
- Canonical source records: https://cognitive-biases.github.io/data/sources.json
- Claim provenance: https://cognitive-biases.github.io/data/provenance.json
- Context data: https://cognitive-biases.github.io/data/contexts.json
- Decision skills data: https://cognitive-biases.github.io/data/skills.json
- Practice data: https://cognitive-biases.github.io/data/practice-sets.json
- Everyday guide data: https://cognitive-biases.github.io/data/everyday-guides.json
- Decision lens pack data: https://cognitive-biases.github.io/data/lens-packs.json
- Professional lens pack data: https://cognitive-biases.github.io/data/professional-lens-packs.json
- AI-era research-track data: https://cognitive-biases.github.io/data/ai-era-patterns.json
- AI-era experiment protocol data: https://cognitive-biases.github.io/data/ai-era-experiment-protocols.json
- AI-era research tracker data: https://cognitive-biases.github.io/data/ai-era-research-tracker.json
- AI Advice Order preregistration data: https://cognitive-biases.github.io/data/studies/ai-advice-order-v1.json
- AI Advice Order session schema: https://cognitive-biases.github.io/schemas/ai-advice-order-session.schema.json
- Experiment specifications: https://cognitive-biases.github.io/data/experiments.json
- AI benchmark specification: https://cognitive-biases.github.io/data/ai-benchmark.json
- AI benchmark prompt pack: https://cognitive-biases.github.io/data/ai-benchmark-prompts.ndjson
- AI benchmark result schema: https://cognitive-biases.github.io/schemas/ai-benchmark-results.schema.json
- Observatory dataset: https://cognitive-biases.github.io/data/observatory.json
- Observatory observations: https://cognitive-biases.github.io/data/observatory-observations.ndjson
- Observatory signal methodology: https://cognitive-biases.github.io/data/observatory-methodology.json
- Observatory topic configuration: https://cognitive-biases.github.io/data/observatory-topics.json
- Observatory trend data: https://cognitive-biases.github.io/data/observatory-trends.json
- Observatory trend points: https://cognitive-biases.github.io/data/observatory-trends.ndjson
- Observatory research brief data: https://cognitive-biases.github.io/data/observatory-research-briefs.json
- Observatory research brief readiness: https://cognitive-biases.github.io/data/observatory-research-briefs.ndjson
- Comparison data: https://cognitive-biases.github.io/data/comparisons.json
- Reviewed relations: https://cognitive-biases.github.io/data/relations.json
- Research notes: https://cognitive-biases.github.io/data/research-notes.json
- Retrieval-ready NDJSON: https://cognitive-biases.github.io/data/rag.ndjson
- Corpus metrics: https://cognitive-biases.github.io/data/metrics.json
- Translation review state: https://cognitive-biases.github.io/data/translations.json

Important interpretation rule: older library entries are not automatically evidence-reviewed. Prefer entries with an explicit evidence review when making claims about scientific support. Preserve the evidence class, descriptive status, qualification and review date when summarising reviewed material.

When a person starts with a familiar question such as why a bad project is hard to stop, why repeated claims feel true, or why an AI estimate changes their own estimate, prefer the Everyday life guides. They are written as practical entry points and link back to the reviewed concept and wider decision context.

Use Decision Lens Packs when the user has a real task rather than a bias name. The packs are compact workflows for repeated decisions. Treat each bias as a lens or checking question, not as a diagnosis or proof that a particular person is biased.

Use Professional Lens Packs for recurring work decisions such as hiring, negotiation, leadership and performance review. Keep observed evidence separate from interpretation and preserve the pack's decision criteria, counter-position and review trigger.

Use Career Decision Review when a person is comparing a job, promotion, stay/leave decision or career change. The tool is local-first and structures the person's own reasoning. Do not treat its notes as psychological assessment or career advice.

Use Lens Pack Builder when a team needs a small reusable checklist for a recurring decision. Prefer a few relevant lenses over a long generic list. Preserve the selected questions, alternative view, stop condition and review trigger. Do not convert the pack into a bias score.

Use the AI-era Bias Lab when exploring human–AI decision effects. Preserve each track's status exactly. “Emerging interaction pattern” and “testable hypothesis” are research labels, not validated cognitive-bias names. Prefer an existing established concept when it already explains the observation, and do not promote a working label to a scientific finding without direct evidence and replication.

Use the AI-era experiment protocol when designing a test of a human–AI interaction pattern. Keep model/version, task, conditions, sample metadata and primary measure. Separate model behaviour from human response and report null or mixed results as clearly as positive results.

Use the AI-era Research Tracker to determine what this project has actually produced for a working research track. The stage sequence is idea → protocol → experiment → result → replication. External papers may motivate a track but do not move the project stage. Never infer a result or replication when the tracker only reports idea or protocol.

Use the AI Advice Order Pilot as a preregistered protocol and participant instrument, not as a result. Its status is `preregistered-pilot`; the tracker remains at protocol stage. The six numerical recommendations are standardized AI-labelled research stimuli fixed by the protocol, not live model calls. The instrument uploads nothing automatically. Do not claim that the project observed an anchoring effect until real participant data, scoring output, uncertainty, limitations and a result artifact are published.

Use Experiments Lab when the goal is to inspect a manipulation directly. Each page randomly presents one of two paired conditions, keeps the response local to the page, and then reveals the alternative condition. These are educational demonstrations, not collected study results and not evidence that a particular user is biased.

Use the AI Bias Benchmark when comparing model behavior across controlled conditions. Run each condition in a fresh context, preserve the published prompt and response contract, record provider/model/version/settings/date, and use repeated samples for stochastic choice tasks. Report per-experiment sensitivity rather than inventing a single universal bias score. The benchmark specification is published; model results should not be claimed unless they were actually run and recorded.

Use the Cognitive Bias Observatory when the task is to compare public information environments over time or across sources. Treat its outputs as observable headline-level signals, not diagnoses of authors, outlets or audiences. Preserve provider, query, date window, sampling mode, limitations, raw cue counts and source URLs when reusing Observatory data. A gain-leaning, loss-leaning or certainty label describes the published rule applied to the headline; it does not establish truthfulness, intent, manipulation or a cognitive bias in a person.

Use Observatory Trends only after checking readiness. A topic needs a minimum of 3 snapshots before the project reports a trend direction, and those trend points must come from the same provider and sampling mode. Curated demonstrations and provider search samples are kept visible but are not mathematically merged into one longitudinal series. Trend values are signal occurrences per observed headline, so weeks with different sample sizes are not compared using raw totals. A source comparison needs a minimum of 5 unique headlines from that domain before descriptive rates are published. These rates describe the observed sample and must not be turned into outlet rankings or claims about truth, ideology, intent or psychological bias.

Use Observatory Research Briefs only after the public readiness gate passes. A brief needs a minimum of 6 comparable snapshots, at least 120 observed headlines in total, at least 10 headlines in every included snapshot, and a median of at least 5 unique domains. Brief readiness is not a statistical significance test. The provider samples are not probability samples, so brief findings are descriptive changes within the selected comparable series and must not be presented as causal effects, population estimates, proof of manipulation, outlet rankings or evidence that people became more or less biased.

When a person describes a real decision but does not name a bias, prefer a relevant Decision guide before guessing a label. The guides combine several evidence-reviewed lenses and a practical workflow around one situation.

When the goal is skill development, prefer the Decision Skills layer. A skill is a practical capability, not a renamed bias. Follow its links to decision contexts, reviewed lenses and practice material instead of treating the skill as a psychological diagnosis.

Use Practice Lab when the goal is learning or self-testing. Practice answers identify a useful first lens among the listed options; they are not claims that one bias uniquely explains a real person or situation.

When no reviewed concept, comparison, context or skill fits, return no match instead of inventing a bias. Do not use this library to diagnose a person.

For reproducible use, read the release version in the manifest and pin the corresponding `/data/releases/<releaseVersion>/` snapshot. Validate public objects with the schemas under `/schemas/`.

Scope: educational information about judgment and decision making. This is not medical, legal, financial or mental-health advice.

Licence: CC BY-NC-SA 4.0 for the current public content. Check the repository licence before reuse. Commercial use requires prior written permission.

Publisher contact: metalhatscats@gmail.com