Heuristics are mental shortcuts that trade completeness for speed. Herbert Simon called it bounded rationality: with limited time and information, good enough is the rational target.

Whatever comes to mind easily feels frequent and likely. Tversky and Kahneman showed we judge risk by mental retrievability — which is why plane crashes loom larger than heart disease.

We judge category membership by resemblance to the stereotype, not by the odds. If it looks like a librarian, we call it a librarian — even when farmers outnumber librarians twenty to one.

Estimates start from whatever number is lying around and adjust insufficiently from it. Even absurd, random anchors drag judgements — a finding among the most replicated in the field.

Paul Slovic showed that risk and benefit judgements run through one emotional dial: things we like feel safe and useful; things we dread feel dangerous and useless. Feeling substitutes for analysis.

When you recognise one option and not the other, you pick the recognised one — and Gigerenzer showed this often beats deep analysis. Sometimes knowing less genuinely predicts more.

Statements that are easy to process — clear fonts, rhymes, familiar accents — get rated truer, safer, and more likeable. The brain misreads processing ease as a signal about the world.

Likelihood gets judged by how easily a scenario plays in the mind's cinema. Kahneman and Tversky found near-misses hurt more — missing a flight by five minutes is easier to 'undo' than by an hour.

Gigerenzer's fast-and-frugal flagship: rank your cues, decide on the single best one that discriminates, ignore the rest. In noisy environments it rivals full regression models.

Simon's word for how choice really works: set an aspiration level, take the first option that clears it. Not laziness — under real constraints, endless optimising is the irrational move.

Bartlett showed memory is not playback but rebuild: each recall reassembles the event from fragments, beliefs, and mood — then saves the edited version over the original.

Once the outcome is known, Fischhoff showed, the past silently reorganises to make it look inevitable — 'I knew it all along'. It quietly destroys our ability to learn from decisions.

Elizabeth Loftus showed a leading question can rewrite a memory: witnesses who heard 'smashed' recalled broken glass that never existed. Recall plus suggestion equals revision.

Kahneman found experiences are remembered by their most intense moment and their ending — duration barely counts. A longer procedure with a gentler end is remembered as better.

Ordinary discomforts fade from memory faster than pleasures, so the past keeps brightening in storage. The good old days are partly an artefact of what pain forgets.

Sparrow and colleagues found we remember where information lives better than the information itself. Externalised memory is powerful — as long as you know what you've outsourced.

Unfinished tasks intrude on attention more than finished ones — the open loop keeps humming. Replications are mixed, but the practical fix holds: write it down and the nagging quiets.

Lists are remembered at the edges: strong start (primacy), strong finish (recency), soggy middle. Whatever you bury mid-list — in a speech, agenda, or CV — is what gets forgotten.

Ideas return without their source tags and feel freshly minted. Honest people plagiarise this way — the memory of reading it decays faster than the idea itself.

Vivid memories of shocking days feel photographic, but Neisser's Challenger study showed they drift like any other — only confidence stays high. Vividness is not accuracy.

Thorndike found one salient trait — looks, confidence, success — spills into unrelated judgements. We rate the well-dressed as smarter and the successful as wiser, on no extra evidence.

Tajfel's minimal group studies split strangers by coin toss — and loyalty appeared instantly. The tiniest 'us' triggers favouritism; real tribes just turn up the volume.

In Asch's line studies, people denied their own eyes to match the group — but note the myth-bust: most participants resisted most of the time. Pressure is real; surrender is not inevitable.

Titles, uniforms, and lab coats inflate the credibility of whatever follows them. Milgram's obedience work — for all its ethical baggage — showed how far borrowed authority can push.

Ross showed we project our own habits and views onto the population: whatever we do feels normal, so 'everyone agrees'. Your social bubble is a terrible opinion poll.

Gilovich found people massively overestimate how much others notice them — the embarrassing shirt, the stumble, the bad hair. The audience is busy starring in its own show.

Others fail because of who they are; we fail because of what happened. Notably cultural: the character-first default runs stronger in Western samples than East Asian ones.

Janis's term for cohesion curdling into bad consensus: dissent self-censors, doubts stay private, and the room agrees its way off a cliff. The evidence base is messier than the fame.

Darley and Latané: the more witnesses, the more responsibility diffuses. Myth-bust: the Kitty Genovese story that launched the field was largely inaccurate — the effect is real, the legend isn't.

Under uncertainty we copy the crowd — usually a decent heuristic, which is exactly why it fails hard in bubbles, panics, and laugh tracks. Cialdini made it persuasion's workhorse.

Ask people for 90% confidence intervals and the truth lands inside barely half the time. Confidence is a feeling about ourselves; accuracy is a fact about the world. They correlate weakly.

The real finding: low performers overestimate modestly — not fools declaring genius. Everyone self-rates near 'above average'; part of the graph is a statistical artefact. Still: the gap bites hardest at the bottom.

Successes are credited to skill; failures are billed to luck, timing, or other people. It protects the ego beautifully and the judgement not at all.

Ellen Langer found people bet more on dice they throw themselves. Rituals, lucky objects, and 'systems' flourish wherever randomness is doing the actual work.

Pronin's finding: people readily see biases in others and claim exemption for themselves — and the smarter the subject, the better the story of exemption. Reading this book grants no immunity.

Tali Sharot's work: we expect divorces, diseases, and disasters — for other people. The base rate applies to everyone who isn't us, which is how the uninsured stay cheerful.

Kahneman and Tversky: plans run on best-case scripts while reality runs on base rates. Your inside view says two weeks; the record of similar projects says two months. Believe the record.

Aronson and Mills: the harsher the initiation, the more members value the group. Suffering demands a justification, so the mind inflates the prize rather than admit the cost was waste.

Svenson's classic: 93% of drivers rated themselves above the median — a mathematical impossibility. On easy, vague, self-defined skills, nearly everyone lives above average.

Quoidbach, Gilbert, and Wilson: every age admits how much it changed — and predicts it's now done changing. You are the current draft, not the final one. Decide accordingly.

Give people a vivid case description and the population statistics quietly leave the room. Kahneman's cab problem: the witness feels decisive; the base rate is decisive.

Tversky and Kahneman's Linda problem: 'feminist bank teller' gets ranked likelier than 'bank teller' — impossible, since the specific can never beat the general. Detail adds plausibility while subtracting probability.

Five heads in a row and tails feels 'due' — but the coin has no memory and no debts. Independent events stay independent, however strongly the pattern-hungry brain objects.

Gilovich called the hot hand an illusion; thirty years later Miller and Sanjurjo found the original maths was itself biased — real streaks may exist. Editorial lesson: even bias research has biases.

'Ten died' lands harder than '1 in 100,000'. We count dramatic numerators and simply forget to ask the only clarifying question that matters: out of how many?

Extreme results drift back toward average on the next try — no cause required. Kahneman's flight instructors credited praise and blame for what was just statistics returning home.

Moving from 99% to 100% safe feels worth a fortune; 60% to 61% feels like nothing. Prospect theory: probability is felt on a warped scale that overprices certainty at both ends.

Tversky and Kahneman's hospital problem: small samples swing wilder, yet intuition treats a clinic of 15 births like a city of 15,000. Variance shrinks with size — feelings don't.

We prefer erasing a tiny risk completely over halving a huge one — the psychic reward of 'zero' beats the arithmetic of lives. Budgets follow the feeling, not the maths.

The category's thesis: numbers arrive wrapped in stories, and the story is what gets judged. The counter-move is boring and reliable — when stakes are real, do the actual arithmetic.

Kahneman and Tversky's headline: losses hurt roughly twice as much as equal gains please. Recent work debates the boundaries, but the tilt is real — we defend far harder than we reach.

'90% survive' and '10% die' are the same statistic — and produce different surgery decisions, even among physicians. The frame does the choosing when the numbers are identical.

Whatever the form pre-selects, most people keep — organ donation rates track the checkbox, not the culture. Inertia is a policy instrument; someone is always choosing your defaults.

Add a third option nobody wants — worse than B, similar to it — and choices swing toward B. Ariely's Economist pricing made the print-only decoy famous. Menus are engineered.

Resources already spent argue for spending more — 'we can't quit now' — though the past is unrecoverable either way. Escalation of commitment is this fallacy wearing a suit.

Samuelson and Zeckhauser: the current arrangement wins by default because change carries salient risk while staying carries hidden cost. Doing nothing is also a decision.

In the famous mug experiments, owners demanded roughly double what buyers offered — for a mug owned minutes. Possession alone inflates value; 'mine' is a price multiplier.

Norton, Mochon, and Ariely: labour breeds love. Self-assembled furniture, half-made cake mixes, and your own mediocre code all get valued above their market worth.

Iyengar's jam study: 24 varieties drew crowds, 6 drew buyers. Honest caveat — meta-analyses show the effect depends on context. But when options paralyse, cutting them is the kindness.

Prospect theory's S-curve, the category's spine: outcomes are felt as gains and losses from a reference point, losses loom steeper, and both ends flatten. It won Kahneman the 2002 Nobel.

Simons and Chabris's gorilla: focused on counting passes, half of viewers miss a gorilla strolling through. Attention is a spotlight, and everything outside it may as well not exist.

Swap the actor mid-conversation behind a moving door and most people don't notice. Vision keeps a sketch, not a recording — change the scene between glances and the sketch papers over it.

The pattern engine runs hot: faces in clouds, figures in shadows, a man in the moon, omens in noise. Better to see a tiger that isn't there than miss one that is — so the false alarms stay.

Random data clumps — genuinely random sequences look streaky. Londoners mapped Blitz bomb clusters onto spy theories; the statistics said scatter. Randomness rarely looks random.

Learn a word, buy a car, hear a name — suddenly it's everywhere. Nothing changed but your attention filter. Also called Baader-Meinhof; the world stayed the same size.

Nothing is judged alone: the £90 bottle looks cheap beside the £900 one, the average candidate shines after a weak one. Whoever controls the neighbours controls the judgement.

Zajonc showed repeated exposure alone breeds liking — no reward, no argument, just familiarity. It's why the chorus repeats, the logo recurs, and the tenth listen wins you over.

In a list of similar items, the one oddity is what memory keeps. Isolation buys recall — the red word in black text, the pause in the patter, the one slide with no bullets.

Watch lips say 'ga' while audio plays 'ba' and you hear 'da' — perception fuses the senses and hands you the compromise as raw reality. Seeing is inferring, even in hearing.

Helmholtz's old truth, the category's thesis: the senses deliver fragments and the brain ships its best guess as finished reality. The dress was blue and gold because guessing is seeing.

Wason named the master bias: belief bends the search, the reading, the memory, and the test. We don't examine evidence so much as recruit it. Every other entry here reports to this one.

Kunda, via Gilovich's two questions: for conclusions we like we ask 'can I believe it?'; for ones we hate, 'must I believe it?'. Different standards of proof, chosen by preference.

Ross's debriefing studies: tell people the evidence was fabricated and the induced belief survives anyway. Beliefs grow roots the evidence never had. Retractions chase; they rarely catch.

The famous claim — corrections strengthen false beliefs — mostly failed to replicate. Wood and Porter found corrections usually work, if modestly. A bias about biases, duly corrected.

Scan all the evidence, keep the convenient slice, present it as the whole orchard. It convinces others and — run silently — convinces the picker too.

Shoot the barn, then paint the target around the holes: gather data first, draw the hypothesis wherever the clusters landed, claim marksmanship. Prediction before data, or it's decoration.

Hasher's finding, advertising's engine: repetition breeds felt truth. It works on the informed too — knowing better doesn't switch it off, because the fluency still accrues.

Semmelweis showed handwashing saved mothers; the establishment mocked him. New evidence that indicts current practice triggers rejection first, reflection later — often much later.

Big events demand big causes — a lone gunman or a virus from a market feels insultingly small. Conspiracy theories thrive in that gap between the size of the effect and the cause.

The category's thesis: once a belief merges with the self, counter-evidence stops being information and starts being attack. That's why the smartest defend hardest — more armour, same wound.

Abraham Wald's bombers: armour where the returning planes weren't hit — the shot-down carried the real lesson. Every visible success conceals a graveyard of identical strategies.

Conclusions inherit the flaws of who got sampled: phone polls miss the phoneless, online surveys sample the online, exit interviews sample the leaving. Ask who never made the dataset.

A trend can hold in every subgroup and reverse when they're combined — Berkeley's admissions looked biased overall, fair by department. Aggregation is an editorial act.

The drunk searches under the lamp because the light is better there. Research, metrics, and dashboards all drift toward what's easy to measure — not where the keys are.

Vietnam by body count: quantify the measurable, ignore the unmeasurable, then declare the unmeasurable unimportant. The spreadsheet was winning while the war was lost.

Positive, novel results get printed; null results go in the file drawer. The literature you read is the survivors' newsletter — meta-analyses now correct for it, imperfectly.

Group averages don't describe individuals: a wealthy postcode contains poor households; a high-crime area, mostly law-abiding people. Statistics about groups are not verdicts about members.

Test a hundred variables and five will 'hit' at p<0.05 by chance alone. P-hacking turns noise into publications — and Nicolas Cage films really do track pool drownings.

Reclassify borderline cases from one group to another and both averages rise — no one improved. Stage migration haunts medical statistics: better scans 'improve' every survival curve.

The category's thesis: numbers inherit every flaw of their collection, then arrive dressed as objectivity. A chart is testimony, not truth — cross-examine it like any witness.

Gigerenzer's standing correction to the bias literature: shortcuts are adaptations, not defects. In their native environments, fast-and-frugal rules beat elaborate models. The sin is habitat mismatch.

Knowing a bias by name grants little immunity — awareness fades exactly when the bias fires. What works is changing the setup: checklists, defaults, and rules that decide before you're tempted.

The single best debiasing move on record: leave your special case and consult the reference class. How did this go for everyone who tried it? Start there; adjust reluctantly.

One of the few techniques with real experimental support: after judging, ask why the judgement might be wrong. The forced counter-search interrupts the confirmation machinery mid-cycle.

Thaler and Sunstein's move: since biases are predictable, design around them — defaults, friction, salience, timing. Honest caveat: 2022 meta-analyses found smaller effects than the bestsellers promised.

Kahneman's two systems — a metaphor, he stresses, not brain anatomy. The practice stands: match the mode to the stakes. Gut for the reversible; deliberation for the doors that lock.

Tetlock's superforecasters aren't bias-free — they're bias-aware and scored. Forecast, attach a probability, record it, grade it. Accuracy improves the way any skill does: with feedback.

Kahneman's favourite institution: rivals co-design the test that could settle their dispute, predict separately, publish jointly. Disagreement becomes instrumentation instead of noise.

Hundreds of biases have names; collecting them all is trivia, not wisdom. Find the three or four that repeatedly cost you, and build your personal pre-flight checks around those.

The thesis of this collection: every entry describes the reader. The catalogue fails as a weapon for judging others and works only as a mirror — same shortcuts, same speed, wiser habitats.