Thinking, Fast and Slow

1. 1Treat fluency as a signal, not a fact: when a claim or judgment feels easy, familiar, and low-effort, flag it as a candidate for System 2 review rather than accepting it as true. Fluency raises acceptance and the illusion of truth independent of accuracy.
2. 2Separate the two inputs behind any availability-based judgment: the number of instances retrieved versus the ease with which they came. Ask yourself which one you are actually going by, because ease usually wins by default.
3. 3Deliberately increase retrieval difficulty to test a judgment. If asked to list six supporting instances feels easy but twelve feels hard, notice that the harder task lowers your confidence — this steep, unexpected drop in fluency is doing the work, not the content.
4. 4Provide a spurious external explanation for the difficulty (background music, screen color, font, curved text boxes) to disrupt the heuristic. Once low fluency is 'explained' by an irrelevant cause, it stops feeding into the judgment — you can use the same trick on yourself by attributing effort to fatigue or distraction.
5. 5Raise System 2 engagement to force content over ease: make the judgment personally consequential. People with real stakes count instances rather than trusting the feel of retrieval.
6. 6Reset expectations in advance so surprise (the trigger for the availability inference) does not fire. Warn yourself that the last few instances will be hard to produce, the way being told the boy wears a top hat reduces surprise when you see it.
7. 7When you cannot suppress the fluent impression, learn to recognize the pattern and recall what you know about it — like measuring the Müller-Lyer lines. You will still 'see' the illusion, but you can refuse to act on it.

Schwarz asked people to list six instances in which they behaved assertively, then rate their own assertiveness. A second group was asked for twelve instances. The first three or four instances came easily to everyone; producing the last few for a set of twelve was a struggle, so fluency dropped. Counterintuitively, the people who listed twelve rated themselves as LESS assertive than those who listed only six — the experience of effortful retrieval trumped the larger number of examples retrieved. The effect reversed direction depending on the trait: people who couldn't easily list twelve instances of meek behavior concluded they were quite assertive. When experimenters told subjects that background music would make retrieval harder, the difficulty was no longer surprising, so it stopped feeding the judgment: those people rated themselves equally assertive after twelve instances as after six. A UCLA professor exploited the same bias in reverse — students asked to list MORE ways to improve a course rated it higher, because struggling to find complaints felt like evidence there wasn't much wrong.

• If a judgment feels easy and the answer arrives without effort, then treat that ease as a cue about the retrieval process, not proof about the world.
• If asking for more instances makes you feel worse about a trait rather than better, then you are being driven by unexpectedly low fluency, not by content — discount the shift.
• If the difficulty of retrieval has a plausible irrelevant explanation (fatigue, distraction, awkward format), then do not let that difficulty influence your conclusion.
• If the judgment matters to you personally and you have real stakes, then deliberately count and weigh the actual instances, because engaged System 2 replaces the ease heuristic.
• If you still 'see' the illusion after knowing the facts (like the measured lines), then govern your action by the known rule, not by the impression you cannot switch off.

• Reading number-of-examples as strength when ease is actually driving the rating — the tell is that a bigger list produces a weaker conclusion.
• Mistaking the feel of fluent retrieval for evidence about frequency: estimating that K is more common as a first letter because such words come to mind faster, though every letter is easier to recall in first position.
• Accepting a claim as true because it is familiar and easy to process — the illusion of truth — without ever engaging System 2 to check it.
• Assuming a warning removes the impression itself: you correct the belief but still experience the pull, then act on the pull. The tell is 'I know it's not longer but I still trust that it looks longer.'
• Being casual, low-stakes, and unengaged, which leaves you maximally exposed to availability — the tell is that you never bothered to count the actual instances at all.

• ▸Judgment being made: state the conclusion you are about to accept.
• ▸Ease rating (1-5): how effortless did reaching this feel?
• ▸Source of ease: is it familiarity, quick retrieval, or actual verified content?
• ▸Instance count vs. instance ease: which am I actually going by?
• ▸Steep-drop check: did adding more required instances make me feel worse? (Y/N)
• ▸Spurious-explanation test: is there an irrelevant cause (fatigue, format, mood) for the difficulty?
• ▸Stakes level: is this personally consequential enough to force System 2 counting?
• ▸Known-illusion flag: is this a pattern I've been fooled by before? What's the rule to apply?
• ▸Verdict: accept, discount for fluency, or re-do with counted content.

1. 1Name the decision and ask: what would I need to know to answer this well? List the required evidence categories before looking at what you actually have.
2. 2Compare your list against the information present. Explicitly mark every category that is missing or thin rather than letting System 1 build a story around only what is activated.
3. 3Ask whether your confidence is tracking the coherence of your story or the completeness of your evidence. If the story feels clean and effortless, treat that ease as a warning, not a confirmation.
4. 4Generate the opposing case yourself. In the Thornton one-sided-evidence study, participants could easily have produced the other side but did not — force yourself to argue against your first impression.
5. 5Check for base rates and quantities that resemblance suppresses. With Steve, resemblance to a librarian dominates until you ask about relative population sizes (20+ farmers per librarian).
6. 6Decorrelate inputs: before any group discussion, have each person write a brief independent summary, so early, assertive voices don't collapse everyone onto one coherent-but-partial story.
7. 7Deliberately seek information you don't yet have, the way you would when buying a home — run a checklist of boxes that must be filled before deciding, so lazy System 2 doesn't just ratify System 1.

Amos and two Stanford graduate students gave participants legal scenarios like the David Thornton / Thrifty Drug Store case. Some participants, like a jury, heard both sides; others heard only one. The lawyers added no facts you couldn't infer from the background story, and one-sided participants knew they had heard only one side and could have generated the other argument themselves. Yet one-sided presentation produced pronounced effects on judgment — and those who saw one side were more confident than those who saw both. This is exactly what WYSIATI predicts: confidence tracks the coherence of the story you can build from available information, not its completeness. Knowing little actually made it easier to fit everything into a clean pattern.

• If your judgment feels highly confident but rests on a single source or report, then stop — treat confidence as evidence of coherence, not adequacy (the consultant-report tell: 'WYSIATI, they didn't realize how little information they had').
• If a description resembles a stereotype, then explicitly retrieve the base rate before predicting (Steve: more farmers than librarians).
• If you have heard only one side, then generate the missing side yourself before forming a view — you can, and one-sided exposure still biases you if you don't.
• If a decision matters and effort is warranted, then run an explicit checklist of information that must be present, because System 1's input never ceases and will otherwise fill the gaps.
• If a group must decide, then collect independent written positions first — open discussion overweights whoever speaks early and assertively.

• Mistaking a coherent story for a complete one — the tell: you can't name a single piece of missing evidence when asked.
• Reacting to resemblance while statistical facts stay inactivated — the tell: base rates 'don't come to mind' until someone forces the question.
• Being more confident on less evidence — the tell: your certainty rose as your information narrowed, because a smaller set is easier to make consistent.
• Accepting one-sided input despite knowing it's one-sided — the tell: 'they didn't want more information that might spoil their story.'
• Letting a lazy System 2 rubber-stamp the first impression — the tell: no waiting, no discomfort, and a persistent bias toward whatever registered first.

• ▸Decision at hand: state the specific judgment or choice in one line.
• ▸Evidence I actually have: list the sources and facts currently activated.
• ▸Evidence I'd need for a full answer: list required categories regardless of availability.
• ▸Missing / thin items: mark every gap between the two lists above.
• ▸Source independence: are my inputs decorrelated or echoing one origin?
• ▸Base rate check: what statistical fact does resemblance make me ignore?
• ▸Opposing case: write the strongest argument against my first impression.
• ▸Confidence source: is my confidence from story coherence or evidence completeness?
• ▸Next action: what missing information will I seek before deciding?

1. 1Separate confidence from evidence: before acting on a judgment, write down explicitly what you actually observed versus what you inferred, because subjective confidence depends on the coherence of your story, not the quantity or quality of evidence.
2. 2Ask what's missing: apply WYSIATI in reverse — list the critical evidence that should inform this judgment but is absent, since we fail to allow for the possibility that what we see is all there is.
3. 3Force System 2 activation on the judgment. The Harvard frowning study shows base-rate neglect is often laziness, not ignorance — invest deliberate effort, treat the task as demanding, and instruct yourself to 'think like a statistician' rather than 'think like a clinician.'
4. 4Test by trying to refute, not confirm. Deliberately search for evidence that would break your belief, because the default positive test strategy of System 1 (and even scientists) seeks confirming data and exaggerates the likelihood of the story.
5. 5Decorrelate error: gather independent judgments before people influence each other, so errors cancel rather than compound (the Wisdom of Crowds penny-jar mechanism).
6. 6Discount hindsight. When reviewing a past outcome, note what you actually predicted beforehand rather than what now feels inevitable, since overconfidence is fed by the illusory certainty of hindsight.
7. 7Weight chance and small samples. Explicitly combine any surprising result with information about sample size — a task System 2 must do because System 1 cannot hold and rule-combine several ideas at once.

Kahneman describes grading exams and noticing he was contaminated by the halo effect: the first answer he read shaped how he scored the rest, sparing him the discomfort of seeing one student do well on some questions and badly on others. When he switched to grading one question across all students before moving to the next, an uncomfortable inconsistency emerged — the same student scoring high and low. That inconsistency was real; his prior smooth, confident impression had been manufactured coherence. The fix conformed to a principle: decorrelate error, grading each item independently so no single first impression propagated. This mirrors the penny-jar case from Surowiecki's Wisdom of Crowds, where individuals estimate poorly but averaged independent judgments are accurate because errors cancel — the same discipline that tames overconfidence in a single mind.

• If your confidence is high but you can name little evidence, then treat the confidence as a signal of story-coherence, not truth, and downgrade it.
• If you can construct only one coherent narrative from the facts, then assume disconfirming evidence is missing rather than absent, and go find it.
• If a surprising result comes from a small sample, then hold both facts in working memory together and refuse to let the vivid individual case override the base rate.
• If multiple people must judge the same thing, then collect their views independently before discussion so their errors decorrelate.
• If an outcome now seems obvious in hindsight, then recover what you predicted in advance and grade your foresight against that record, not the outcome.

• Mistaking fluency for accuracy — the tell: the judgment came fast and felt right, and you accepted it without checking what evidence you actually had.
• Base-rate neglect from laziness, not ignorance — the tell: you 'know' the statistic is relevant but only apply it when prompted; your System 2 endorsed the intuition without effort (the frowners caught it, the cheek-puffers didn't).
• Positive test strategy — the tell: you searched for cases that fit your belief ('Is Sam friendly?') and never asked the refuting question, so you exaggerated an improbable event's likelihood.
• Halo-driven coherence — the tell: your view of a person or situation is uniformly positive or negative across traits you never observed, like assuming personable Joan is also generous.
• Hindsight certainty — the tell: you feel you 'knew it all along' and cannot reconstruct the genuine uncertainty you faced before the outcome.

• ▸Judgment/prediction: state it in one sentence.
• ▸Evidence actually observed: list only what you directly saw.
• ▸Inferences added: what you filled in beyond the evidence.
• ▸Missing critical evidence: what should inform this but is absent (WYSIATI check).
• ▸Refuting test: one piece of data that would break this belief — did you look for it?
• ▸Base rate / sample size: the relevant statistic and how large the sample is.
• ▸Independent judgments: names of others asked before any discussion.
• ▸Hindsight check: what did I actually predict beforehand?
• ▸Revised confidence: high / medium / low after the audit.

1. 1State the frequency or probability judgment you are about to make out loud (e.g., 'adultery is more common among politicians').
2. 2Notice the retrieval: catch yourself searching memory for instances and ask whether ease-of-recall is standing in for actual frequency — this is the availability substitution.
3. 3Separate numerator from denominator. Write down the vivid instances that came to mind (numerator) and then deliberately ask about the base rate you never see (denominator). System 2 must count what System 1 skips.
4. 4Audit the source of your instances: are these examples abundant in your memory because they are frequent, or because they are dramatic, celebrity-linked, and heavily media-reported? Journalists' choice of topics inflates certain categories.
5. 5Check for a cascade: is a single nonevent being amplified by media repetition and public reaction feeding back into more coverage? Trace whether coverage reflects real prevalence or reflects prior public interest.
6. 6For risk decisions, name the pattern explicitly — probability neglect (treating small risks as zero or as huge, nothing between) combined with availability cascade — and force a comparison to a mundane baseline risk (e.g., traffic deaths).
7. 7Correct by anchoring to statistical facts rather than resemblance or recalled exemplars, and demote issues that are underweighted only because they are undramatic (declining educational standards, overinvestment in end-of-life care).

Kahneman held a long-standing impression that adultery is more common among politicians than among physicians or lawyers, and had even built explanations for it: the aphrodisiac effect of power and the temptations of life away from home. The 'fact' felt solid because politician transgressions came to mind so readily. On examination he realized the transgressions of politicians are far more likely to be reported than those of lawyers and doctors. The abundance of recalled instances was a product of journalists' topic choices, not of any real difference in frequency. Once he separated ease-of-retrieval from actual base rate, the impression collapsed — it 'could be due entirely' to the availability heuristic. The tell was that he had a strong belief with confident causal stories but no statistical grounding.

• If a frequency estimate rests on examples that spring to mind rather than counted data, then treat it as an availability judgment and discount it until you find the base rate.
• If your instances are dramatic, celebrity-linked, or recently in the news (e.g., Michael Jackson's death dominating channels), then assume media salience — not real prevalence — is driving the estimate.
• If you feel either near-zero worry or overwhelming dread about a small risk with nothing in between, then flag probability neglect and force a numerator/denominator comparison.
• If public alarm is escalating around a single incident with repeated coverage feeding more coverage, then label it an availability cascade and check casualty or prevalence figures against mundane baselines (e.g., terror deaths vs. traffic deaths).
• If an issue is critical but undramatic and you keep forgetting it, then deliberately raise its weight — its low availability is precisely why it is neglected.

• Manufacturing causal stories to justify an availability-driven belief (the 'aphrodisiac of power' explanation) — the tell is confident reasons attached to a claim you never checked against data.
• Imagining the numerator (the tragic story on the news) while ignoring the denominator — the tell is intense concern that is insensitive to the actual probability of harm.
• Mistaking media coverage for reality — the tell is that your sense of what matters tracks what channels report, and coverage in turn just mirrors prior public interest, a closed loop.
• Treating small risks as all-or-nothing — the tell is you either dismiss a risk entirely or give it far too much weight, like the parent waiting up who knows there is nothing to worry about yet cannot stop images of disaster.
• Assuming you are immune because you know the statistics — the tell is that even Kahneman notes his own example choices were guided by availability and that it is difficult to reason oneself into complete calm; terrorism speaks directly to System 1 regardless of knowledge.

• ▸Judgment being made: state the frequency/probability claim in one sentence.
• ▸Instances that came to mind: list the specific examples you recalled.
• ▸Ease-of-retrieval check: were these easy to recall because frequent, or because dramatic/celebrity/media-covered?
• ▸Numerator: the vivid story or count you are actually imagining.
• ▸Denominator: the base rate or total population you cannot easily see — find a real number.
• ▸Cascade check: is coverage growing off a single nonevent in a feedback loop with public reaction? (yes/no + evidence)
• ▸Mundane baseline comparison: name a comparable everyday risk and its rate (e.g., traffic deaths).
• ▸Corrected estimate: your judgment after anchoring to statistics instead of resemblance/recall.
• ▸Neglect flag: name any important-but-undramatic issue you are underweighting and raise its priority.

1. 1Identify the anchor before you estimate: name every number that entered your mind first — an asking price, an opening offer, a prior figure, even an obviously irrelevant one like a spun wheel. It will pull your estimate toward it whether or not you consent.
2. 2Separate the two mechanisms. Ask: did I consciously adjust from a starting number (System 2), or did the number just make a compatible impression 'come to mind' (System 1 priming)? Different defenses apply to each.
3. 3When adjustment is the process, notice you stop at the near edge of the region of uncertainty — the first point where you're no longer sure you should move farther. Deliberately push past that edge and keep generating reasons to move.
4. 4When priming is the process, don't try to 'correct'; the anchor already selectively evoked compatible evidence. Instead, generate the opposite — force your associative machinery to construct arguments as if the anchor were false.
5. 5Protect adjustment from depletion. People stay closer to the anchor when memory is loaded or resources are drained; do anchored estimates when fresh, not tired or distracted.
6. 6Physically reject the anchor. Epley and Gilovich found head-shaking (rejecting) moves estimates farther from the anchor; nodding enhances anchoring. Adopt a stance of active rejection while you reconsider.
7. 7Refuse to be first-anchored by opponents: when someone throws out a value 'to consider,' recognize any number offered as a candidate solution induces an anchor, and generate your own independent estimate before reacting to theirs.

Kahneman and Tversky spun a wheel of fortune rigged to stop at 10 or 65 in front of University of Oregon students, then asked each to write the number down and estimate the percentage of African nations in the UN. The wheel's number carries zero information — participants should have ignored it entirely. They did not: those who saw 10 averaged 25%, those who saw 65 averaged 45%. The 20-point gap came purely from an uninformative random number acting as an anchor. This was the first demonstration of anchoring's sheer absurdity — no one could call the influence of a wheel of fortune 'reasonable.' The lesson: if an obviously meaningless number moves judgments this much, a meaningful-looking one (an asking price, a rival's offer) moves them more, and your determination to resist it does not protect you.

• If a number was offered to you 'to consider' as a possible answer, then treat your estimate as contaminated and build a fresh figure from independent evidence.
• If you produced an estimate without any conscious sense of adjusting, then assume priming (System 1) is at work — deliberately generate anchor-incompatible evidence rather than trying to nudge the number.
• If you did consciously adjust from a starting value, then assume you stopped too early and push farther until you're clearly past the region of uncertainty.
• If you're tired, distracted, or cognitively loaded when making an anchored estimate, then defer the judgment — a depleted System 2 stays closer to the anchor.
• If you're setting a listing price, opening offer, or first figure, then recognize you are handing the other side an anchor — set it deliberately, not casually.

• Believing willpower defeats anchoring — the same house 'appears more valuable' at a high listing price even for someone 'determined to resist'; the tell is confidence that you weren't influenced.
• Mistaking all anchoring for adjustment and 'correcting downward' when the effect was actually priming; the tell is that with the Gandhi-144 question you never felt yourself adjusting, yet your estimate still rose.
• Stopping adjustment at the first plausible number; the tell is a feeling of 'I'm no longer sure I should go farther' — that near edge of uncertainty is exactly where you quit too soon.
• Making anchored estimates while depleted (loaded memory, slightly drunk, distracted) and staying glued to the anchor; the tell is that easy or fatigued conditions produce estimates suspiciously close to the given number.
• Nodding along — literally or figuratively accepting the anchor before reasoning; the tell is agreement or head-nodding when a figure is named, which enhances anchoring rather than resisting it.

• ▸Estimate needed: state the unknown quantity you must judge.
• ▸Anchor(s) present: list every number that came to mind or was offered first.
• ▸Anchor relevance check: is this number actually informative, or uninformative (like a wheel spin)?
• ▸Mechanism: adjustment (I consciously moved from it) or priming (it just shaped my impression)?
• ▸Anti-anchor evidence: write 3 reasons the true value differs from the anchor's direction.
• ▸Push-past-edge estimate: give a value beyond the first point where you felt unsure.
• ▸Depletion status: am I fresh, or tired/loaded/distracted right now? Defer if depleted.
• ▸Independent estimate: the figure you'd give if the anchor had never appeared.
• ▸Final judgment: the estimate you commit to after rejecting the anchor's pull.

1. 1When you face a probability or likelihood question, notice the substitution: you are being asked 'how likely,' but your mind answers 'how similar to the prototype.' Name the swap explicitly before you answer.
2. 2Retrieve the base rate first. Before rating fit, ask what fraction of the population actually belongs to each category — e.g., more than 20 male farmers exist for each male librarian in the US.
3. 3Rate representativeness separately and label it as such. It is legitimate for a similarity judgment to ignore base rates and evidence quality; the error is letting that number stand in for probability.
4. 4Assess the quality of the evidence. Ask whether the description is trustworthy; recall that Tom W's source was 'not highly trustworthy.' Discount fit when the source is weak.
5. 5Combine, don't substitute: anchor on the base rate, then adjust toward the description only to the degree the evidence is diagnostic and reliable.
6. 6Check for the anti-base-rate trap: a description engineered to fit a small category (computer science) and mismatch large ones (humanities, education) will pull you toward the rare answer. Distrust vivid fits to tiny groups.
7. 7For conjunction-type framings, remember that adding a specific detail raises representativeness but can only lower probability. A more detailed, more 'fitting' story is never more probable than its simpler component.

Kahneman typed up Tom W as an 'anti-base-rate' character — a good fit to small fields like computer science and a poor fit to the largest fields like humanities and education. He handed it to Robyn Dawes, a sophisticated statistician and skeptic about intuitive judgment, the one person who should have flagged the base rate. Dawes smiled slyly and guessed 'computer scientist?' — he had not spontaneously thought of the base rate at all, recognizing his mistake only when Kahneman said the words. Then 114 graduate students in psychology, all trained in statistics, ranked the nine fields by probability, and those rankings did not differ from rankings by similarity to the stereotype: substitution was perfect. They knew computer science was a small field and knew the description was untrustworthy, yet ranked it highly because it scored highest on representativeness. The tell was that the probability question got answered with the answer to the easier similarity question.

• If the question asks 'how probable' but your gut delivers a vivid match to a stereotype, then stop and treat that gut answer as a similarity score, not a probability.
• If a category is rare in the population, then require strong, trustworthy evidence before ranking it above common categories, no matter how well the description fits.
• If the source of a description is unreliable, then shrink your judgment toward the base rate proportionally.
• If adding detail to a scenario makes it feel more likely, then flag a conjunction error — the more specific version cannot exceed the probability of its parts.
• If the description seems designed to fit a small group perfectly, then suspect an anti-base-rate setup and weight base rates harder.

• Answering the easy question for the hard one: reporting similarity when asked probability. Tell — your ranking by likelihood is identical to your ranking by resemblance.
• Ignoring group size: deciding Steve is a librarian because he's 'meek and tidy,' forgetting there are 20+ farmers per librarian. Tell — you never asked how many people are in each category.
• Assuming expertise protects you: statisticians like Dawes and 114 trained students all failed. Tell — you 'know' base rates matter but did not spontaneously use them.
• Being seduced by coherent detail: a richer, more specific story feels more probable. Tell — you rate the detailed scenario above its own broader component.
• Trusting a description's fit without checking its source. Tell — you never questioned whether the evidence was diagnostic or accurate.

• ▸Question as asked: write the exact likelihood/probability question
• ▸Prototype match: what stereotype does the description resemble, and how strongly (0–100)?
• ▸Candidate categories: list every plausible outcome, including the boring common ones
• ▸Base rate per category: estimate the population share of each (e.g., farmers vs. librarians)
• ▸Evidence source & trust: who gave the description and how reliable is it (high/med/low)?
• ▸Diagnosticity: how much does this evidence actually distinguish the categories?
• ▸Adjusted ranking: start from base rate, shift toward fit only by trust × diagnosticity
• ▸Substitution check: does my probability ranking merely copy my similarity ranking? (yes = redo)
• ▸Conjunction check: did any detail make a specific option feel more likely than its parts?

1. 1When you form a judgment about category membership or probability, pause and name the base rate first: ask 'how common is this category in the population before I saw any description?' Steve is more likely a farmer than a librarian because there are more male farmers, full stop.
2. 2Treat the vivid description as evidence to be weighed against the base rate, not as a replacement for it. The personality sketch of Steve as 'meek and tidy' is salient, but salience is not frequency.
3. 3Run the WYSIATI check: ask 'what critical information am I missing, and would it change my answer?' The associative machine builds a coherent story from what is activated and suppresses the ambiguity that a missing base rate would introduce.
4. 4Force the neglected fact into memory deliberately, the way remembering Detroit-is-in-Michigan raises murder-rate estimates. Slow down and search: which relevant statistical facts do I know but have not retrieved?
5. 5Estimate the answer two ways — once from the story, once from the base rate — and reconcile the gap. If they diverge sharply, the story is probably overriding the statistics.
6. 6For sum-like or quantity questions, check that your judgment actually scales with the numbers. If your answer to 'save 2,000 birds' equals your answer to '200,000 birds,' you have reacted to a prototype, not a quantity.
7. 7Recognize the social/effort cue: an intuitive answer that arrives instantly and effortlessly (like '10¢' for the bat and ball) is exactly the kind that skipped the base-rate check. Ease is a warning sign, not a confirmation.

Recall Steve, described as meek, tidy, and helpful — a soul often judged to be a librarian. The personality description is salient and vivid, and it maps onto the stereotype of a librarian, so System 1 delivers 'librarian' with cognitive ease. But you surely know there are far more male farmers than male librarians in the population; that statistical base rate almost certainly did not come to mind when you first answered. What you saw was all there was. To correct it, you first state the base rate (farmers vastly outnumber librarians), then treat 'meek and tidy' as weak evidence that shifts, but does not overturn, that prior. The reconciled judgment is that Steve is still probably a farmer — the vivid match cannot outweigh a lopsided base rate.

• If a judgment arrives instantly and feels certain from a vivid description, then stop and retrieve the base rate before committing — ease signals the base rate was skipped.
• If the description is stereotypical or emotionally gripping, then discount it: prototypes drive the Exxon Valdez bird result where 2,000 and 200,000 birds got the same $80.
• If your answer does not change when the underlying quantity changes tenfold, then you are matching a prototype, not the number — recompute.
• If you cannot state what population or frequency your case is drawn from, then you lack the base rate and should treat your confidence as an illusion.
• If remembering one additional fact (like Detroit's crime rate) would move your estimate, then you have not searched memory adequately — slow down and search.

• Substituting representativeness for probability: judging Steve a librarian because he fits the type. Tell: your answer is driven entirely by how well the case matches a stereotype.
• Letting a coherent story suppress the missing statistic: System 1 settles on a pattern and buries the doubt. Tell: you feel no discomfort and asked no 'what am I missing' question.
• Quantity neglect in emotional contexts: reacting to the prototype image rather than the count, as with the drowning oil-soaked bird. Tell: willingness to act is flat across wildly different magnitudes.
• Knowing the base-rate-relevant fact but failing to retrieve it (Michigan/Detroit). Tell: you'd get a different answer if the fact were spoon-fed to you.
• Trusting effortless intuition and skipping the check because checking is 'unpleasant.' Tell: you gave the first answer that came to mind, like the 50%+ of elite students who said 10¢ on the bat-and-ball.

• ▸Question asked: state the probability/category judgment you are making.
• ▸Base rate first: what is the population frequency before any description? (state it as a rough number or ratio)
• ▸Vivid/description evidence: what specific details are pulling your intuition, and how stereotypical are they?
• ▸Instant answer: write the first answer that came to mind, and note how effortless it felt.
• ▸Missing information (WYSIATI): what critical fact, if present, would change the answer?
• ▸Memory search: name one relevant fact you know but did not initially retrieve.
• ▸Quantity check: if a number is involved, does my answer scale when the number changes?
• ▸Reconciled judgment: weigh base rate against description and state the corrected answer.

1. 1When you observe an extreme outcome followed by a more average one, first ask whether luck played any role in the first outcome. If it did, expect regression before inventing any cause.
2. 2Estimate how extreme the original score is relative to the average. The more extreme it is, the more regression you predict, because an extreme score signals an unusually lucky (or unlucky) occasion.
3. 3Make your prediction of the next outcome more moderate than the evidence you started from. Never predict a repeat of an extreme performance; shift it toward the mean.
4. 4Run the arrow-of-time test: predict the earlier event from the later one. If you still see regression going backward, the effect cannot be causal, so drop the causal story.
5. 5Separate the astute observation from the false inference. The pattern (extreme, then average) is real; the causal claim attached to it (my praise, my punishment) is what to discard.
6. 6Before crediting an intervention, check whether you selected the cases by their extremity. If you only praised top performers or only shouted at the worst, you built regression into your data.
7. 7Substitute a chance mechanism to test the intuition: model the outcome as skill plus luck. If a pure luck process reproduces the same before/after pattern, no causal explanation is needed.

An Israeli Air Force flight instructor objected that praise makes cadets worse and screaming makes them better, contradicting the principle that reward beats punishment. His observation was correct: praised cadets did do worse next time, punished ones did better. But he had praised only cadets whose performance was far above average—probably lucky that attempt—so they were likely to deteriorate regardless of praise, and screamed only at unusually bad performances that were likely to improve regardless of the shout. To demonstrate this, Kahneman had every officer throw two coins backward at a chalk target without looking; ranked from best to worst on the first try, most top performers deteriorated and most poor performers improved on the second try, with no reward or punishment involved. The pattern the instructor saw matched pure random fluctuation. He had attached a causal interpretation to the inevitable fluctuations of a random process.

• If you selected cases because they were extreme, then expect the next measurement to move toward the mean before attributing the shift to your treatment.
• If regression appears when you predict an earlier event from a later one, then reject any causal explanation for the change.
• If an outcome depends partly on luck, then predict the next outcome to be closer to the average than the current evidence—more so the more extreme the current score.
• If a pure chance model reproduces the observed before/after pattern, then no cause is required and adding one is an error.
• If a story sounds satisfying but rests only on the change between an extreme and a normal outcome, then treat the satisfaction as a warning, not evidence.

• Reading punishment as effective because bad performances improve afterward—the tell is that you only intervened after the worst attempts.
• Reading reward as harmful because good performances decline afterward—the tell is you only praised the best attempts.
• Explaining the 'Sports Illustrated jinx' by pressure or overconfidence, when the simpler account is that cover athletes had a lucky peak season that luck won't repeat—the tell is a psychological narrative for a purely statistical dip.
• Inventing motivational stories like 'he's tense protecting his lead, so he'll do worse' or 'nothing to lose, so he'll relax'—the tell is a confident causal claim about a second attempt following an extreme first one, with no measurement of the supposed cause.
• Preferring any causal account over 'it's a mathematically inevitable consequence of luck'—the tell is discomfort with the non-story making you fill the gap.

• ▸Outcome observed: describe the extreme result and the follow-up result.
• ▸Selection basis: was this case chosen because it was extreme? (yes/no)
• ▸Luck component: does luck plausibly influence this outcome? (yes/no)
• ▸Extremity rating: how far from average was the first result (mild/strong/extreme)?
• ▸Regression prediction: state a follow-up estimate moved toward the mean.
• ▸Arrow-of-time test: does regression also appear predicting the earlier event from the later? (yes/no)
• ▸Proposed causal story: write the tempting explanation you want to attach.
• ▸Verdict: given the checks above, is the causal story warranted or is this regression?

1. 1Catch the moment a decision resolves fast and feels good. When an answer comes quickly and you feel satisfied with yourself, treat that ease as a flag, not a verdict — it usually signals the affect heuristic is operating.
2. 2Name the hard question you actually face (e.g., 'Is Ford stock currently underpriced?'). Write it out in full so it cannot be quietly swapped.
3. 3Identify the easy question your mind answered instead (e.g., 'Do I like Ford cars?'). The tell is that your justification refers to liking and feelings rather than to the relevant metric.
4. 4Check for the substitution: list your reasons and mark each as either a feeling of liking/disliking or a piece of decision-relevant evidence. If liking dominates, you answered the easy question.
5. 5Test whether you have genuine expertise for this situation. A chess master's fast intuition is reliable because he recognizes the position; if you lack that recognition, the fast answer is affect, not skill.
6. 6Deliberately engage slower, effortful thinking to answer the original question with relevant facts and statistics, the way an economist would ask about price.
7. 7Guard against mood contamination: recognize that a good mood loosens System 2's control and makes you more intuitive and less vigilant, so postpone consequential judgments made while feeling especially good or especially bad.

An executive decides to invest a large sum in Ford stock. He explains that he likes the cars, likes the company, and likes the idea of owning its stock, and he is satisfied with his gut feeling. The one question an economist would call relevant — is Ford stock currently underpriced? — never entered his mind. Faced with a difficult question (should I invest?), his intuitive machinery answered an easier related one (do I like Ford cars?) and let that determine the choice, without his noticing the substitution. Because stock picking accuracy is low and he has no relevant expertise here, it is reasonable to conclude he did not know what he was doing. His feelings of liking guided judgment directly, with little deliberation or reasoning — the essence of the affect heuristic.

• If your reasons for a risk/benefit judgment are all statements of liking or disliking, then you have answered the easy question — stop and answer the hard one.
• If the answer arrived quickly and effortlessly AND you lack demonstrated expertise in this exact situation, then treat the intuition as affect, not skill.
• If you are in a notably good mood, then defer consequential judgments, because good mood increases reliance on System 1 and lowers vigilance.
• If you can state feelings but not the decision-relevant metric (price, base rate, probability), then you are missing the input that would actually resolve the question.
• If your fast answer is compelling and you feel no urge to check it, then apply Cognitive Reflection discipline — the reluctance to verify is itself the warning.

• Mistaking liking for evidence: the executive was satisfied with himself while never asking the relevant question. Tell: your justifications name affections, not facts.
• Assuming your intuition is expertise: only recognized situations yield reliable fast answers. Tell: you cannot articulate what pattern you 'recognized.'
• Deciding in a good mood and calling it clarity: happiness feels like confidence but reduces vigilance and raises logical errors. Tell: unusual ease and creativity paired with no double-checking.
• Impulsive acceptance of the first idea: low Cognitive Reflection scorers answer with the first thing that comes to mind and won't invest effort to check. Tell: impatience and craving for immediate resolution.
• Never noticing the substitution at all: the swap from hard to easy question happens without awareness. Tell: you feel you answered the original question but your evidence maps only to the easier one.

• ▸Decision at hand: state the choice you are about to make in one line.
• ▸Hard question (the relevant one): write the question an expert/economist would insist on (e.g., is it underpriced?).
• ▸Easy question my mind answered: name the liking/disliking question that came readily.
• ▸My stated reasons: list them and tag each as FEELING or EVIDENCE.
• ▸Expertise check: do I recognize this exact situation from real experience? yes/no.
• ▸Current mood: note if unusually happy/sad and whether it may be loosening vigilance.
• ▸Missing decision-relevant metric: name the number or fact I have not consulted.
• ▸Verification step: write the one effortful check I will perform before deciding.
• ▸Verdict: proceed on evidence, or pause because affect is driving.

1. 1Before trusting any gut feeling, name the environment it formed in: ask whether the domain is regular and predictable, or noisy and irregular. Intuition only becomes valid where regularities exist to be learned.
2. 2Check for adequate practice: Simon's chess masters needed thousands of hours before they came to see the board differently. Ask how many repetitions the intuition rests on, not how confident it feels.
3. 3Verify the feedback loop was timely and accurate. Firefighting and medicine give quick, clear outcomes; stock picking gives delayed, noisy feedback. No reliable feedback means no reliable learning.
4. 4Reframe intuition as recognition, not magic: 'the situation has provided a cue; this cue has given the expert access to information stored in memory, and the information provides the answer.' Locate the actual cue.
5. 5Screen for substitution: when a hard question comes to mind with an easy, confident answer, suspect you answered a related easier question instead (affect heuristic — 'do I like it?' standing in for 'is it a good decision?').
6. 6When the environment fails the regularity/feedback test, do not rely on the felt confidence. Switch to slow, deliberate System 2 work — for statistical questions, 'replace impression formation by computation whenever possible.'
7. 7Treat confidence and validity as separate: 'We are often confident even when we are wrong.' Use an outside check because an objective observer detects your errors more readily than you do.

A chief investment officer told Kahneman he had just put tens of millions of dollars into Ford stock. His stated reason: he had attended an automobile show and been impressed — 'Boy, do they know how to make a car!' He trusted his gut and was satisfied. But the relevant question was 'Is Ford stock currently underpriced?' — and he never asked it. Instead his mind answered an easier question, 'Do I like Ford cars?', and let that determine the decision. This is the affect heuristic and the substitution it produces: a difficult question quietly swapped for a related easy one, felt as insight. Stock picking is not the regular, high-feedback environment that breeds valid intuition, so his confidence carried no diagnostic weight.

• If the environment is regular and gives fast, accurate feedback and you have thousands of hours in it, then trust the recognition-based intuition.
• If the answer arrives instantly and feels emotionally 'liked' but the question was genuinely hard, then suspect substitution and force the original question back into view.
• If the domain is irregular or feedback is delayed and noisy (stock picking), then discount confidence entirely and compute or seek an outside view.
• If you can identify the specific cue the situation provided and trace it to stored knowledge, then the intuition is likely recognition; if you cannot, then it is likely a heuristic guess.
• If you're planning something with a known method (like sample size), then 'replace impression formation by computation whenever possible' rather than trusting tradition and gut.

• Mythologizing intuition as magical insight — the tell is treating an expert's judgment as unexplainable rather than as recognition of a familiar cue.
• Trusting confidence as a proxy for accuracy — the tell is being 'satisfied with yourself' about a decision in a domain where outcomes are actually unpredictable.
• Answering the easy question and not noticing the swap — the tell is a fluent, immediate answer to a question that should have required real deliberation.
• Applying intuition across domains — expertise in cars becomes felt authority over stock valuation; the tell is confidence that outruns the domain's learnability.
• Belief in the law of small numbers: even statisticians 'exaggerated the likelihood' of replication and chose samples too small — the tell is trusting a strong impression from thin evidence and skipping the computation you know how to do.

• ▸Decision/judgment: state the intuitive call you're about to trust.
• ▸Original question: write the actual question that must be answered.
• ▸Substituted question: what easier/liked question might your mind have answered instead?
• ▸Environment regularity: is this domain predictable enough to learn? (yes/no + why)
• ▸Practice depth: roughly how many reps/hours of real exposure back this feel?
• ▸Feedback quality: was outcome feedback timely and accurate, or delayed and noisy?
• ▸Cue identified: name the specific cue triggering recognition (blank = red flag).
• ▸Confidence vs. validity: rate felt confidence 1-10, then note whether the environment justifies it.
• ▸Fallback action: if screen fails, specify the computation or outside check you'll run instead.

1. 1Identify whether you are in a low-validity environment: outcomes are noisy, feedback is delayed or absent, and expert intuition shows no consistent edge (the passages cite clinical prediction, radiologists, corporate audits, and unstructured interviews as such settings).
2. 2List the small number of predictor variables that actually correlate with the outcome, rather than collecting everything the mental shotgun computes; more resemblance-based cues do not add validity.
3. 3Assign each predictor a weight — even equal weights work, per Dawes's 'robust beauty of improper linear models'; convert each candidate's raw data to a standardized score.
4. 4Sum the weighted scores into a single formula output and rank candidates strictly by that number.
5. 5Suppress the intuitive matching operation: when your System 1 supplies a stereotype match (Steve 'must' be a librarian), treat it as a signal to return to base rates, not as evidence.
6. 6Anchor every prediction to the base rate first, then adjust toward the individuating evidence only as far as its real validity justifies — do not translate one scale directly onto another.
7. 7Commit in advance to follow the formula's output, because the whole point of the William Grove meta-analysis (Clinical Versus Mechanical Prediction) is that mechanical rules beat holistic judgment even for trained experts.

Take the Steve problem exactly as posed: a neighbor describes Steve as shy, withdrawn, tidy, orderly, with a passion for detail, and asks whether he is more likely a librarian or a farmer. Intuition matches the description to the librarian stereotype and answers 'librarian' immediately. A statistical approach starts with the base rate: there are more than 20 male farmers for every male librarian in the United States. Even if a much higher fraction of librarians than farmers are 'meek and tidy,' the sheer 20-to-1 population difference means more meek and tidy men sit on tractors than at information desks. So the formula-driven answer is 'farmer.' The passage's participants ignored this and relied exclusively on resemblance — the exact error a weighted, base-rate-anchored rule eliminates.

• If the environment is noisy with poor feedback (low validity), then use a formula and distrust expert holistic judgment — Grove's meta-analysis shows mechanical prediction wins.
• If a vivid personality or resemblance cue makes one answer feel obvious, then stop and retrieve the base rate before answering.
• If you find yourself translating one scale onto another (reading age at four → predicted GPA), then you are prediction-by-matching, which is statistically wrong; regress toward the mean instead.
• If a candidate's data comes from an unstructured interview, then down-weight it — Dana and Dawes document 'the persistence of an illusion' in believing such interviews.
• If adding another predictor does not demonstrably raise validity, then leave it out rather than let it dilute the formula.

• Relying on representativeness: judging Steve by his fit to a stereotype while ignoring the 20-to-1 base rate. Tell: the answer feels instant and certain.
• Availability substitution: estimating a rate (divorce among professors, adultery among politicians) by how easily instances come to mind rather than by data. Tell: you can name examples but not counts.
• Extreme prediction from weak evidence: matching age-four reading directly to a top GPA. Tell: your forecast is as extreme as the input, with no regression to the mean.
• Overriding the formula with a gut 'special case' exception. Tell: you built the rule, then quietly ignored it when it disagreed with you.
• Assuming intelligence protects you — Stanovich shows high IQ does not make people immune to these biases; rationality is a separate ability. Tell: 'I'm too smart to fall for that.'

• ▸Outcome to predict: state the single quantity or category you are forecasting.
• ▸Validity check: is feedback noisy/delayed? (yes → use formula).
• ▸Base rate: record the population frequency before looking at the individual.
• ▸Predictors (max 4): list only variables with demonstrated correlation to the outcome.
• ▸Weight per predictor: assign a value (equal weights are acceptable per Dawes).
• ▸Standardized score: convert each raw input to a comparable scale.
• ▸Formula output: weighted sum, ranked against other cases.
• ▸Intuition gap: note your gut answer and the distance from the formula — do not close it.
• ▸Commitment: sign that you will act on the formula output, not the intuition.

1. 1Name the forecast you actually need (time, cost, or risk) and state your inside-view estimate first, before defenses form. Kahneman's team said 'about two years'; write your number down so it can be tested.
2. 2Recognize the inside view for what it is: you are extrapolating from the information in front of you (WYSIATI), the chapters you wrote first, your peak commitment. Flag that your early data is unrepresentative of the whole.
3. 3Switch to the outside view: identify a reference class of comparable projects that were actually completed, and ask what the distribution of outcomes looked like for them, ignoring your specific circumstances.
4. 4Get the base-rate number from that class before you elaborate your own case. In Kahneman's episode, the expert knew comparable curriculum teams finished in seven-plus years, and 40% never finished at all.
5. 5Adjust your inside-view estimate toward the reference-class base rate rather than treating the base rate as a footnote. Do not add 'a few months as a margin of error' to a fundamentally optimistic number.
6. 6Explicitly budget for 'unknown unknowns'—the succession of unforeseeable events (Rumsfeld's phrase) that inside-view planning cannot list. The reference class already contains them; your plan does not.
7. 7Strip added detail from persuasive scenarios: richer, more coherent stories feel more probable but are less likely. When a plan gets more detailed and more convincing, treat that as a reason to distrust its probability, not trust it.
8. 8Guard against irrational perseverance separately from the estimate: decide in advance the condition under which you abandon the project, because facing the choice you will otherwise 'give up rationality rather than give up the enterprise.'

Kahneman's textbook team estimated about two years to finish a curriculum. That figure was the inside view: they knew the chapter count and how long the first two chapters had taken, and cautious members added a few months. Their own expert, asked to think as an outside observer, recalled that comparable teams took seven to ten years and about 40% never finished. Instead of revising the two-year plan toward that base rate, the team 'gathered ourselves together and carried on as if nothing had happened.' The book was completed eight years later, by which point Kahneman had left the team and the Ministry of Education no longer wanted it. The planning fallacy set the estimate near a best-case scenario; irrational perseverance kept them going after the outside view had already predicted the outcome.

• If your estimate comes only from your own plan and progress so far, then you are on the inside view—stop and find a reference class before committing the number.
• If you cannot name a set of comparable completed projects and their actual outcomes, then you do not yet have a defensible forecast, only a hope.
• If adding detail makes a scenario more persuasive, then lower—not raise—your probability estimate, because plausibility is being confused with probability.
• If your early data comes from the easiest part of the work or your period of peak motivation, then treat it as unrepresentative and do not extrapolate it linearly.
• If the outside view contradicts your inside-view number and you feel the urge to 'carry on as if nothing happened,' then set an explicit abandonment condition now, before sunk cost hardens.

• Treating the reference-class number as interesting trivia and reverting to the inside-view estimate. Tell: you heard '40% never finish' and still wrote 'two years' anyway.
• Confusing a margin of error with the outside view—adding a small buffer to an optimistic base. Tell: your revised number is your original plus a few months, not anchored to comparable projects.
• Being seduced by detailed scenarios. Tell: the plan that specifies exactly how each phase unfolds feels the safest, when specificity should reduce your confidence.
• Extrapolating from unrepresentative early progress. Tell: you cite how fast the first, easiest chapters went, at the moment commitment was highest.
• Irrational perseverance after the forecast turns bad. Tell: you would rather abandon rational assessment than abandon the project, and no one has defined what would make you quit.

• ▸Forecast needed: state whether you are predicting time, cost, or risk—one target only.
• ▸Inside-view estimate: your first number from your own plan, written before you argue with yourself.
• ▸Reference class: name the set of comparable completed (and abandoned) projects.
• ▸Base-rate outcome: the typical result for that class, including failure/non-completion rate.
• ▸Unrepresentativeness flags: note where your early data is easiest-first or peak-motivation.
• ▸Unknown-unknowns budget: added time/cost for unforeseeable events the class already contains.
• ▸Adjusted estimate: your number moved toward the base rate, not the base rate footnoted.
• ▸Abandonment condition: the specific trigger at which you stop, decided before sunk cost grows.

1. 1Frame every risky decision as a set of simple gambles with explicit money outcomes and probabilities, the way Kahneman and Tversky did: e.g., 'win $130 on heads, lose $100 on tails.' Reducing a fuzzy decision to a gamble is what makes the evaluation testable.
2. 2Locate the reference point before evaluating anything. Ask what state the chooser treats as the baseline (current wealth, an expectation, a prior offer). Outcomes are not evaluated in absolute terms but as gains or losses relative to that point, so misplacing the reference reverses the whole analysis.
3. 3Recode each outcome as a gain or a loss from that reference point, not as a final wealth state. The same $100 is a different psychological object depending on whether it is subtracted from or added to the baseline.
4. 4Apply loss aversion: weight losses more heavily than equal-sized gains. The coin bet illustrates it — many people refuse to risk $100 to win $130 even though the expected value is positive, because the pain of losing $100 outweighs the pleasure of winning $130.
5. 5Apply diminishing sensitivity: the felt difference between $100 and $200 is larger than between $1,100 and $1,200 in either direction. Evaluate marginal changes, not totals.
6. 6Test your own intuitive preference first, then check it against the logic of choice. Kahneman and Tversky's method was to make up choice problems, record the first answer that came to mind even when they knew it was wrong, and hunt for where that answer violates the rules of rational choice.
7. 7Strip inconsequential features from the problem statement and re-present it, because decisions get shaped by framing — how gains and losses are worded rather than what they are. If preference flips when only the wording changes, System 1's reaction, not the substance, is driving the choice.

Consider the passage's own gamble: you win $130 if a coin shows heads and lose $100 if it shows tails. By the logic of expected value the bet is favorable — the average outcome is +$15. Yet the intuitive preference of most people is to refuse. Prospect theory explains why: the two outcomes are coded as a possible gain of $130 and a possible loss of $100 against a reference point of current wealth. Under loss aversion the $100 loss looms larger than the $130 gain, so the psychological value of the bet is negative even though its monetary value is positive. This is exactly the kind of 'intuitive preference that consistently violated the rules of rational choice' that Kahneman and Tversky recorded in their own decisions. The demonstration is powerful precisely because it is simple — you feel the reluctance yourself before any theory is invoked.

• If the expected value is positive but you still recoil, then suspect loss aversion is inflating the weight of the potential loss — name the reference point and recompute.
• If a person's choice flips when the same options are reworded, then the driver is a framing effect, not the outcomes; treat the reversal as evidence, not preference.
• If two options differ only in whether outcomes are described as gains or as losses, then expect risk-averse behavior in the gain frame and risk-seeking in the loss frame.
• If you cannot state the reference point, then you cannot predict the choice — resolve the baseline before applying loss aversion or diminishing sensitivity.
• If a marginal change is added to a large existing amount, then discount its felt impact because diminishing sensitivity flattens the value curve away from the reference point.

• Evaluating outcomes as final wealth states instead of gains and losses. The tell: your analysis gives the same answer regardless of the chooser's starting point, when real preferences clearly shift with it.
• Ignoring the reference point entirely and treating the gamble in pure expected-value terms. The tell: you predict acceptance of the +$15 coin bet and are surprised when people refuse.
• Treating problems in isolation and missing that framing has already skewed the choice. The tell: you accept a stated preference at face value without checking whether a reworded version reverses it.
• Assuming symmetry between gains and losses. The tell: you model a $100 loss and a $100 gain as equal-and-opposite, so you cannot explain the reluctance to bet.
• Overreaching — attributing every deviation to prospect theory. Kahneman notes their focus on biases was criticized as an unfairly negative view of the mind; the tell is that you stop checking whether the intuition is sometimes just accurate.

• ▸Decision framed as a gamble: state each outcome with its probability in money terms
• ▸Reference point: what baseline state does the chooser treat as neutral?
• ▸Outcome coding: label each outcome as a gain (+) or loss (−) relative to the reference point
• ▸Expected value: probability-weighted sum of the money outcomes
• ▸Intuitive preference: record the first answer that comes to mind, before any calculation
• ▸Loss-aversion check: does the felt weight of the loss exceed the felt weight of the equal gain?
• ▸Diminishing-sensitivity check: are outcomes far from the reference point, flattening their marginal impact?
• ▸Framing test: reword the problem (gain frame vs loss frame) and note whether the preference flips
• ▸Rationality gap: does the intuitive preference contradict the expected-value logic? Name the violation

1. 1Locate the reference point before you assess any outcome: ask what baseline (status quo, prior wealth, or expectation) the person is comparing against, because the same absolute state feels different depending on the point it is measured from.
2. 2Separate the absolute level from the recent change. Bernoulli's mistake was treating utility as a function of the level of wealth alone; instead, code the outcome as a gain or loss relative to the reference point.
3. 3Identify each party's distinct reference point separately — two people at the identical current state can carry opposite reference points and therefore opposite experiences.
4. 4Reframe the outcome as a directional move: is this a gain (state above reference) or a loss (state below reference)? The label, not the level, predicts the felt experience.
5. 5Test whether the reference point is the status quo or an expectation. When someone anticipated a higher state, disappointment comes from falling below the expected reference, not below zero.
6. 6Use the sensory analogy as a diagnostic: just as loudness depends on the preceding sound, judge whether the outcome was preceded by a 'whisper' (low reference, so it feels large) or a 'roar' (high reference, so it feels small).
7. 7When predicting satisfaction, forecast the change relative to reference — not the endpoint — because subjective experience tracks the move, not the destination.

Today Jack and Jill each have a wealth of 5 million. Yesterday, Jack had 1 million and Jill had 9 million. Bernoulli's theory says equal wealth means equal utility, so they should be equally happy — but Jack is elated and Jill despondent. The reason is reference dependence: Jack's reference point is 1 million, so 5 million is coded as a large gain; Jill's reference point is 9 million, so the same 5 million is coded as a heavy loss. The pull is so strong that Jack with only 2 million today would still be happier than Jill with 5 million, because his change from 1 million is a gain while her change from 9 million is a loss. The absolute level (2 vs. 5) loses to the direction of the move. To predict who is happier, you must know the reference point, not the endpoint — exactly as predicting loudness requires knowing the reference sound, not just the absolute energy.

• If two parties hold the same current state, then check yesterday's state (or their expectation) before assuming equal satisfaction — the reference point, not the level, decides.
• If an outcome sits above the reference point, code it as a gain; if below, code it as a loss — and expect the felt experience to follow the direction.
• If someone anticipated a specific result, then treat that expectation as the reference and measure the outcome against it, not against zero or against absolute wealth.
• If you cannot predict a subjective experience from the absolute number alone, then find the comparison baseline it is automatically measured against (as loudness requires knowing the preceding sound).
• If a gain and a loss produce the same endpoint, then still expect different reactions, because the reference point makes the paths, not the destination, decisive.

• Assuming utility is a function of the level of wealth (Bernoulli's error): the tell is predicting equal happiness for two people at the same current state, ignoring where they came from.
• Using one reference point for multiple parties: the tell is being surprised that the 'same' outcome delights one person and crushes another.
• Confusing the endpoint with the change: the tell is forecasting satisfaction from the final number rather than from the move relative to baseline.
• Missing an expectation-based reference: the tell is calling someone irrational for being disappointed with a good absolute result they had expected to exceed.
• Ignoring the automatic comparison: the tell is treating an outcome's magnitude as fixed, when it is actually experienced as large or small only relative to what preceded it.

• ▸Current state: the actual outcome or level being evaluated
• ▸Reference point: the baseline it is compared against (status quo, prior state, or expectation)
• ▸Source of reference: mark whether it is status quo, recent prior state, or an expectation
• ▸Direction: gain (above reference) or loss (below reference)
• ▸Magnitude of change: the size of the move from reference to current state
• ▸Preceding context: the 'whisper or roar' that set the baseline
• ▸Other party's reference point: fill separately if a second person is involved
• ▸Predicted experience: the felt reaction implied by direction + magnitude, not the endpoint

1. 1State the choice as a change from a reference point, not as a state of wealth. System 1 is more sensitive to changes than to states, so the way you set the baseline decides whether an outcome reads as loss or gain.
2. 2Name the reference point explicitly. The current holding, the status quo, or an expectation each anchor a different sense of what counts as losing — move the anchor and the same outcome flips valence.
3. 3Weight the prospective pain of a loss roughly twice the pleasure of the matching gain, since System 1 responds more strongly to losses than to gains. Ask whether a decision is being made to avoid a loss rather than to secure a value.
4. 4Widen the frame. Prospect theory frames problems narrowly, in isolation; deliberately group the decision with the other decisions it belongs to so a single loss does not dominate the whole set.
5. 5Separate the risk aversion that protects you from the risk aversion imposed by accountability. Anticipated hindsight and outcome bias push you toward bureaucratic, conventional, over-tested choices that protect the decider more than the beneficiary.
6. 6Test the endowment: ask what you would pay to acquire the thing you now own. If the price to sell far exceeds the price to buy, loss aversion — not value — is setting the number.
7. 7Decorrelate the emotional pull. Because losses recruit an automatic bodily recoil (as with the bananas/vomit reaction), write the numbers down and evaluate them cold before letting the felt aversion decide.

As malpractice litigation became more common, physicians changed their procedures: they ordered more tests, referred more cases to specialists, and applied conventional treatments even when those were unlikely to help. Read through loss aversion, each act is a move to avoid a scrutinized loss rather than to add patient value — the physician weighs the vivid pain of a lawsuit far more heavily than the diffuse gain of a cleaner, cheaper course of care. Accountability and hindsight sharpen this: a decision maker who expects to be second-guessed retreats to standard operating procedures and extreme reluctance to take risks. The passage names the result plainly — these actions protected the physicians more than they benefited the patients, creating a conflict of interest. Notice the reference point at work: the doctor is defending against a downside, not choosing toward an upside. The tell is that the conventional treatment is applied even when unlikely to help — pure loss avoidance with no expected gain.

• If a choice is framed as avoiding a loss, then re-frame it as a state of final outcomes and re-decide — narrow framing overweights the loss.
• If your selling price for something you own is far above what you'd pay to buy it, then the gap is the endowment effect, not the asset's worth.
• If you're taking a defensive action that is unlikely to produce any gain, then you are protecting yourself from blame, not serving the objective — flag the conflict of interest.
• If you expect the decision to be judged in hindsight, then assume you're biased toward over-conservative, over-tested, bureaucratic choices and correct upward.
• If a lucky risk-taker is being praised as prescient, then discount the halo — a few lucky gambles crown reckless leaders, and it says nothing about the risk they actually took.

• Setting the reference point unconsciously so that a normal outcome registers as a loss — tell: you feel the decision as 'not losing' rather than 'gaining,' and the pain is disproportionate to the stakes.
• Evaluating each decision in isolation — tell: you refuse a small favorable gamble you'd happily accept as one of many, because the single potential loss looms alone.
• Mistaking accountability-driven risk aversion for prudence — tell: you can list the tests, referrals, and conventional steps that protect you but not the beneficiary.
• Letting the embodied recoil settle the choice before the numbers are read — tell: you've already decided against the loss and are now assembling reasons, System 1 first, System 2 rationalizing.
• Rewarding survivors of reckless bets and punishing sensible doubters — tell: outcome, not process, is driving your judgment of who took the right risk.

• ▸Reference point in use: state the baseline against which this reads as a loss.
• ▸Alternative reference point: restate the same outcome as a final state of wealth/health.
• ▸Felt loss vs felt gain: rate each 0-10; note if the loss side runs ~2x.
• ▸Wider frame: name the set of decisions this one belongs to.
• ▸Endowment check: buy price vs sell price for the thing at stake.
• ▸Defensive-only test: does this action add expected value, or only avoid blame?
• ▸Accountability pressure: who scrutinizes this in hindsight, and which way does that push me?
• ▸Cold-number decision: the choice made after writing the outcomes down, before the recoil.

1. 1Separate probability from decision weight for any risky prospect. The probability is the objective chance; the decision weight is how much that chance actually moves your System 1 evaluation. Write both down side by side.
2. 2Locate the prospect on the probability scale, because weighting is nonlinear at the ends. Ask where a 5% change sits: 0→5% and 95→100% are qualitative changes (creating possibility, delivering certainty); 5→10% and 60→65% are merely quantitative. Treat the endpoints as high-distortion zones.
3. 3Test for the certainty effect: check whether a sure outcome is pulling disproportionate weight. Compare the sure thing against a near-sure gamble of equal or higher expected value and watch which one your gut prefers.
4. 4Test for the possibility effect: check whether a tiny probability (a lottery ticket, a rare catastrophe) is being overweighted relative to its objective size. The tell is a strong emotional reaction to a near-zero chance.
5. 5Map the prospect onto the fourfold pattern using two axes—gains vs. losses, and high vs. low probability—to predict the risk attitude that will emerge before you decide.
6. 6Recompute the expected value with objective probabilities (average of outcomes each weighted by its true probability, e.g. 20%×$1,000 + 75%×$100 = $275) and compare it to the choice your weighting pushed you toward. The gap is the cost of distorted decision weights.
7. 7Reframe the endpoints deliberately: convert an overweighted small risk into a frequency or a repeated-exposure statement, and treat 'certainty' as just one more point on the probability line rather than a special category.

Take the four cases where your chance of receiving $1 million improves by exactly 5%: A) 0→5%, B) 5→10%, C) 60→65%, D) 95→100%. The expectation principle says each raises your utility by precisely 5% of the utility of $1 million—identical news four times. Your experience says otherwise: A and D feel far more impressive than B and C. Moving 0→5% is qualitative because it creates a possibility of winning that did not exist before—a hope—while 5→10% is only a quantitative nudge. Moving 95→100% is equally special because it removes all doubt and delivers certainty, which System 1 pays a premium for. B and C sit in the flat middle of the weighting curve where the same 5% barely registers. This shows decision weights are nonlinear: the ends of the probability scale carry more psychological force per percentage point than the middle.

• If an outcome is a near-certainty and you are being asked to pay a premium to make it fully certain, then flag the certainty effect and check whether the premium exceeds the true 5% (or smaller) probability gap.
• If a probability is very small but the emotional pull is large, then assume you are overweighting it—overweighting low probabilities is a documented System 1 tendency—and re-anchor on the objective number.
• If the prospect involves gains at high probability, expect risk aversion (a preference for the sure thing); if it involves gains at low probability, expect risk seeking (the lottery pull).
• If the prospect involves losses at high probability, expect risk seeking (fighting to avoid a near-certain loss); if losses at low probability, expect risk aversion (buying insurance).
• If your gut choice and the computed expected value diverge, then treat the divergence as caused by decision weights and require an explicit reason before overriding the expected value.

• Treating a 5% change as equally meaningful everywhere on the scale. The tell: you rank 60→65% as good news equal to 0→5%, ignoring that one creates possibility and the other doesn't.
• Paying a large premium for certainty without noticing it. The tell: you accept a much lower expected value just to remove the last sliver of risk.
• Confusing overweighting a rare event with rational caution. The tell: your fear or hope scales with vividness, not with the actual probability, and you can't state the number.
• Forgetting that losses flip the risk attitude, so you apply gain-domain intuitions to loss-domain choices. The tell: you're surprised that people 'gamble' to avoid a near-certain loss when they were risk-averse over gains—it's the asymmetry of losses and the fourfold pattern, not irrationality.
• Letting System 1 deliver the global evaluation while you remain a passive observer. The tell: you can state your preference instantly but cannot decompose the weights you assigned to each outcome.

• ▸Prospect: state the outcome(s) and their objective probabilities in one line
• ▸Domain: mark gain or loss
• ▸Probability zone: mark low (near 0), mid, or high (near 100)
• ▸Predicted risk attitude: read from the fourfold pattern (risk-averse or risk-seeking)
• ▸Objective expected value: compute sum of each outcome × its true probability
• ▸Gut preference: write the choice System 1 delivers before analysis
• ▸Distortion flag: name the effect at work (certainty effect / possibility effect / overweighting low probability)
• ▸Gap: describe the difference between gut preference and expected value, and its cost
• ▸Reframe: restate the prospect as a frequency or as one point on the probability line to reduce distortion

1. 1Write out both formulations of the same fact before you decide anything — the survival frame ('90% survive') and the mortality frame ('10% die'). If you cannot state both, you have not seen the choice, you have seen a pitch.
2. 2Identify which system is answering. The emotional pull toward the reassuring frame is System 1 producing a fast reaction; forcing the equivalence onto paper recruits System 2 to check it.
3. 3Test logical equivalence explicitly: convert both statements to the same units (percentages, absolute numbers, dollars). If they reduce to identical content, any preference difference between them is a framing effect, not information.
4. 4Neutralize the emotional coloring by restating the option in the frame you find least attractive, then re-ask whether your choice holds. If it flips, your preference was driven by wording.
5. 5Apply the WYSIATI check: ask 'what would the other formulation make salient that this one hides?' Because you normally see only one formulation, and what you see is all there is, deliberately supply the missing one.
6. 6When presenting a choice to others (patients, teams, customers), give both frames or a neutral frame, so their decision tracks the outcome rather than your phrasing.
7. 7Log the decision and the frame you used. Systematic errors recur predictably, so a record lets you spot when your preferences keep bending to presentation.

A surgeon can describe the same operation two ways: 'the odds of survival one month after surgery are 90%' or 'mortality within one month of surgery is 10%.' The two statements are logically identical — 90% living is the same population as 10% dying. Yet the survival frame is more reassuring and the mortality frame more alarming, so patients (and physicians) lean toward surgery when they hear the first and hesitate when they hear the second. The equivalence is transparent the moment you write both down, but an individual normally sees only one formulation, and what she sees is all there is. The same mechanism makes cold cuts labeled '90% fat-free' more attractive than the identical product labeled '10% fat.' The remedy is not more data — it is forcing the second frame into view so the decision rests on the 90/10 reality, not the word chosen to carry it.

• If two descriptions reduce to the same numbers but produce different feelings, then treat the difference as a framing artifact and discard it from the decision.
• If you can only recall one formulation of a choice, then assume WYSIATI is operating and generate the opposite frame before committing.
• If a frame emphasizes a positive attribute (survival, fat-free), then restate it as its negative complement (mortality, fat content) and confirm your preference survives the switch.
• If your choice flips when the wording flips, then the preference is inconsistent and neither answer is trustworthy — return to the underlying outcome.
• If you are the one presenting the option, then present both frames or a neutral one, because your phrasing will otherwise author the other person's choice.

• Treating the reassuring frame as more informative than the alarming one. Tell: you feel more confident after the '90% survival' version even though no new fact arrived.
• Seeing only one formulation and never noticing a second exists — the pure WYSIATI trap. Tell: you can't state the complementary version when asked.
• Assuming transparency of equivalence protects you. Tell: you acknowledge '90% survive equals 10% die' yet still prefer the surgery framed positively.
• Attributing framing susceptibility to other, less capable people ('the fecklessness of undergraduates') while exempting yourself. Tell: you say experts are immune, forgetting philosophers and economists failed the same demonstrations.
• Handling each choice in isolation, so inconsistencies across differently-framed decisions never surface. Tell: you never compare two of your own decisions that rest on the same facts worded differently.

• ▸Decision at stake: state the outcome you are actually choosing about
• ▸Frame as given: copy the exact wording you were presented
• ▸Emotional reaction to this frame: reassuring / alarming / neutral
• ▸Complementary frame: rewrite the same fact from the opposite angle
• ▸Reduced to common units: express both frames as identical numbers
• ▸Equivalence confirmed?: yes/no — are the two statements logically the same
• ▸Does preference flip across frames?: yes = inconsistency detected
• ▸What the given frame hides: the attribute the other formulation makes salient
• ▸Decision on the merits: choice made after reading the least attractive frame

1. 1Identify whether ownership is present in the situation, because the endowment effect only fires when someone already holds the good; the reference point is 'do I own this now or not.'
2. 2Elicit two separate numbers, not one: the minimum price the owner will accept to sell (WTA) and the maximum price a non-owner will pay to buy (WTP). Standard theory says these should match; the gap is the effect.
3. 3Diagnose the good's category first — the effect is strong for goods held for use or enjoyment (wine, concert ticket) and near-zero for goods held as a proxy for money (a $5 bill exchanged for singles, a merchant's shoes).
4. 4Frame the transaction in loss-versus-gain terms: for the owner, selling registers as giving up (a loss); for the buyer, purchasing registers as getting (a gain). Losses weigh about twice as much as equal gains.
5. 5Estimate the size of the gap using the roughly 2:1 loss-aversion slope — the just-acceptable selling price will tend to sit well above the just-acceptable buying price for the same item.
6. 6When you want to neutralize the effect, reclassify the good as a routine medium of exchange rather than a possession — this collapses WTA toward WTP.
7. 7Predict a status-quo bias in any choice involving change: the disadvantages of switching loom larger than the advantages, so expect strong resistance even to favorable offers.

Professor R (Richard Rosett) was a sophisticated wine lover who refused to sell a bottle from his collection even at $100 in 1975 dollars, yet at auction would never pay more than $35 for a bottle of the same quality. Standard economic theory says he should have a single value for the bottle: if it is worth $50 to him, he sells above $50 and buys below $50, so buying and selling prices coincide. Instead his minimum selling price ($100) sat far above his maximum buying price ($35), and between those figures he neither bought nor sold. Owning the bottle raised its value. Prospect theory explains this by abandoning the idea of a unique utility: when he owns the bottle he feels the pain of giving it up; when he doesn't, he weighs the pleasure of getting it. Because the loss-averse value function is steeper in the negative domain, giving up nice wine hurts more than acquiring equally good wine pleases — producing the $35-to-$100 gap.

• If the person already owns the good, then expect their selling price to exceed a non-owner's buying price for the identical item.
• If the good is a routine proxy for money (change, standard retail stock), then expect little or no endowment effect — WTA ≈ WTP.
• If the good is held for use or is rarely traded, then expect a large WTA/WTP gap and price it accordingly.
• If a favorable opportunity requires giving something up, then anticipate resistance biased toward the status quo unless the benefit clearly exceeds the roughly doubled weight on the loss.
• If you observe someone quoting one number for selling and refusing to buy at a much lower number, then diagnose ownership reference points, not irrationality about market value.

• Assuming a single 'true' value for an item — the tell is treating WTA and WTP as interchangeable and being puzzled when they diverge (as economic theory was with Professor R).
• Applying the effect universally — the tell is expecting loss aversion when changing a $5 bill for singles or buying shoes, where the owner feels no loss.
• Interpreting the seller's high price as greed or the buyer's low price as lowballing, missing that both are driven by the reference point of ownership.
• Ignoring that history matters — the tell is modeling utility as depending only on the end state (Bernoulli's and indifference-curve assumption) rather than on whether you already hold the good.
• Underestimating status-quo bias in decisions of change — the tell is being surprised when people 'turn down very favorable opportunities' because overweighted losses block the switch.

• ▸Good in question: name the specific item being valued.
• ▸Ownership status: does the person currently own it — yes/no.
• ▸Good type: proxy-for-money (routine trade) or held-for-use/rarely-traded.
• ▸Minimum selling price (WTA): lowest amount the owner will accept.
• ▸Maximum buying price (WTP): highest amount a non-owner will pay.
• ▸Gap size: WTA minus WTP; note if it approaches the ~2:1 loss-aversion ratio.
• ▸Loss framing: what does giving it up feel like to the owner?
• ▸Gain framing: what does acquiring it feel like to the buyer?
• ▸Neutralization step: can the good be reclassified as a medium of exchange to shrink the gap — yes/no and how.

1. 1Separate the two questions the two selves answer: the experiencing self answers 'Does it hurt now?'; the remembering self answers 'How was it, on the whole?' Ask each explicitly rather than collapsing them.
2. 2Capture real-time state directly, not retrospectively. In the cold-hand study, subjects used a free hand to move keyboard arrows for a continuous record of pain — a direct communication from the experiencing self. Build a moment-by-moment log for any experience you want to measure honestly.
3. 3Integrate over duration when you care about actual experience: sum or average the momentary states across the full length of the episode. The experiencing self accumulates every second; 40 minutes of musical bliss counts even if the last minute is ruined.
4. 4When you catch yourself scoring a whole episode by its ending or peak, flag it: that is the remembering self substituting for the experiencing self. Name the substitution.
5. 5Decide which self the decision should serve before choosing. If minimizing lived suffering, protect the experiencing self (shorter is better even at higher peak). If a memory must be lived with, shape the peak and the ending instead.
6. 6Watch the tyranny of the remembering self in your own choices: you naturally choose the episode with the better memory, exposing your experiencing self to unnecessary pain. Override by consulting the duration-integrated record, not the recalled grade.

In the cold-hand experiment participants immersed a hand to the wrist in painfully cold 14° water. The short episode ran 60 seconds. The long episode ran 90 seconds: the first 60 were identical, then a valve let slightly warmer water raise the temperature about 1° for the final 30 seconds. For the experiencing self the long episode is strictly worse — it contains all the pain of the short one plus 30 extra seconds of (milder) pain. Yet when told a third trial would repeat one episode exactly, participants tend to choose the long one, because its gentler ending leaves a better memory. Their remembering self, guided by the mild end rather than the total duration, condemns the experiencing self to more pain. The experiencing self has no voice in the choice, so it loses.

• If the goal is to reduce pain actually experienced, then minimize duration even if that raises peak intensity and leaves an awful memory (do the procedure swiftly).
• If the goal is to reduce the memory of pain, then lower the peak and end on relatively mild pain — gradual relief beats abrupt relief.
• If someone says an ending 'ruined the whole experience,' then treat it as a memory judgment, not an experience judgment — the earlier good time already happened and cannot be undone.
• If a decision is being made about repeating or scheduling an experience, then the remembering self is at the wheel; deliberately pull the duration-weighted record of the experiencing self into the choice.
• If you have only a retrospective rating and no moment-by-moment data, then assume it over-weights peak and ending and under-weights duration.

• Confusing experience with the memory of it — the compelling cognitive illusion that a past experience can be ruined. Tell: you assign a failing grade to an episode that was mostly good because it ended badly.
• Duration neglect: judging an episode by peak and end while ignoring how long it lasted. Tell: a longer, strictly worse experience gets the better rating.
• Letting the remembering self govern choices by default. Tell: you pick the option you 'remember better' and end up signing your experiencing self up for more actual suffering.
• Optimizing memory when you meant to optimize experience (or vice versa) without ever deciding which you wanted. Tell: you rushed a painful procedure but complain the patient hates the memory — you served the wrong self.
• Treating the experiencing self's silence as absence of a stake. Tell: you say 'does the actual experience count for nothing?' only rhetorically, then act as if it does count for nothing.

• ▸Episode: the experience being evaluated or chosen
• ▸Experiencing-self log: moment-by-moment record of 'Does it hurt now?' / 'How good is now?'
• ▸Duration: total length of the episode (the number the memory tends to ignore)
• ▸Peak: the most intense moment (good or bad)
• ▸Ending state: how the episode felt in its final seconds
• ▸Remembering-self verdict: the single 'How was it, on the whole?' grade
• ▸Substitution check: is the verdict driven mainly by peak+ending? (yes/no)
• ▸Goal for this decision: reduce experienced pain OR shape the memory (pick one)
• ▸Chosen action: swift/short (serves experiencing self) or mild peak+gentle end (serves remembering self)

1. 1Separate the two questions the selves answer: 'Does it hurt now?' (experiencing self) and 'How was it, on the whole?' (remembering self). Label which one a given judgment is coming from before you act on it.
2. 2Recognize that only the remembering self makes decisions and keeps score, so any choice about repeating an experience is being made on memory, not on the sum of moments actually lived.
3. 3Diagnose peak-end dominance: the memory of an episode is set mainly by the intensity at its worst (or best) moment and by how it ended, not by how good most of it was.
4. 4Diagnose duration neglect: notice that the length of the experience barely registers in the remembered rating, so a longer episode of pain can leave a better memory if it ends more mildly.
5. 5When designing or choosing an experience, decide the objective explicitly — minimize experienced pain, or minimize remembered pain — because these two goals point to opposite procedures.
6. 6Test any 'it ruined the whole experience' claim by asking whether the experience itself was ruined or only the memory of it; the bad ending happened after the good part had already been lived and cannot undo it.

After a lecture, a member of the audience described listening raptly to a long symphony on a disc that was scratched near the end, producing a shocking sound; he said the bad ending 'ruined the whole experience.' It did not. The experiencing self had had roughly 40 minutes of near-total musical bliss, and the shocking sound arrived after that bliss had already occurred, so it could not reach back and cancel it. What was ruined was only the memory. His remembering self, dominated by the end, assigned the entire episode a failing grade and effectively discarded those 40 minutes. This is the peak-end and duration-neglect pattern in action: the end and the worst moment set the score, while the length and the mass of good moments count for nothing.

• If the goal is to reduce experienced pain, then conduct the procedure swiftly even if it raises peak intensity and leaves an awful memory.
• If the goal is to reduce remembered pain, then extend the procedure with mild relief at the end, because gradual relief with a mild ending leaves a better memory than abrupt relief.
• If someone is choosing which of two experiences to repeat, then expect the remembering self to pick the one with the better peak-end profile even if it means more total suffering for the experiencing self.
• If a judgment weighs the ending far more than the middle and ignores duration, then treat it as the remembering self's verdict, not an accurate account of what was lived.
• If you catch yourself saying an experience was 'ruined,' then check whether any actual experienced good was undone — usually it was not, only the memory changed.

• Confusing experience with the memory of it — the tell is language like 'it ruined the whole experience' when the good part was fully lived before the bad ending.
• Letting the remembering self choose the repeat experience — the tell is exposing your experiencing self to unnecessary, longer pain because the shorter-but-worse-ending option is remembered as worse.
• Treating a remembered rating as if it summed all moments — the tell is that a 40-minute bliss and a 60-minute bliss get nearly identical scores (duration neglect).
• Optimizing the wrong objective — the tell is speeding up a painful procedure (good for experienced pain) while complaining patients hate it, because you ignored the peak-end memory they keep.
• Assuming the experiencing self will be heard — the tell is that nobody consults it, because it has no voice; only the remembering self makes and defends the decision.

• ▸Episode described: name the experience being judged or chosen
• ▸Experiencing-self account: rate roughly how it felt moment to moment ('Did it hurt / feel good now?')
• ▸Peak moment: the most intense point (best or worst) and its intensity
• ▸End moment: how the episode actually ended and how intense that was
• ▸Duration: total length of the episode
• ▸Remembered verdict: the one-line 'How was it, on the whole?' rating you'd give
• ▸Peak-end vs. total check: does the verdict track peak+end while ignoring duration and the middle? (yes/no)
• ▸Objective for next time: minimize experienced pain OR minimize remembered pain (pick one)
• ▸Design change: what to alter (speed up, extend with mild ending, soften the finish) to serve that objective

1. 1Separate the two selves before you design anything. Write down the experiencing self's question ('Does it hurt now?') and the remembering self's question ('How was it, on the whole?'). These produce systematically different measures: the hedonimeter total (duration-weighted area under the curve) versus the retrospective rating (peak-end average).
2. 2Decide which measure you are optimizing. If the goal is minimizing memory of a bad experience, target the peak and the end. If the goal is minimizing actual pain experienced moment-to-moment, target the total duration and integral, even at the cost of a worse memory.
3. 3To improve remembered value, compute the peak-end average: take the intensity at the worst (or best) moment and average it with the intensity at the final moment. Lower that average, not the duration.
4. 4Engineer the ending deliberately. Arrange for gradual relief rather than abrupt relief so the final rating is mild; a low ending rating drags the peak-end average down and rewrites the whole memory.
5. 5Do not add duration to improve a memory unless the added time is less intense than the peak — this is the trap that makes people willingly suffer more. Adding a mild tail can improve the remembered experience while worsening the total experience.
6. 6When forecasting people's choices, apply the intuitive rule they actually use: 'pick the option you like most or dislike least,' where liking is set by the peak-end memory. Predict against remembered utility, not total utility.
7. 7Audit for duration neglect explicitly: ask whether your evaluation changed at all when the length changed. If length made no difference, you are rating by peak and end, not by the full experience.

In the cold-hand experiment, participants immersed a hand in painfully cold water for two trials: a short one (60 seconds at a steady bad level) and a long one (the same 60 seconds followed by 30 more seconds during which the water was slightly warmed, so pain diminished at the end). The experiencing self clearly suffered more in the long trial — more total time in pain, a larger area under the curve. But the remembering self preferred it: the peak was the same, and the ending was milder, so the peak-end average was lower. When offered a third trial to repeat, fully 80% of those whose pain diminished at the end chose the longer episode, declaring themselves willing to endure 30 seconds of needless extra pain. They knew which trial was longer — they were asked — but did not use that knowledge; they simply chose the option they disliked least in memory. Applied to patient A (last rating 7, peak 8, average 7.5) versus patient B (last rating 1, peak 8, average 4.5), the same logic explains why A retains a far worse memory despite B's procedure lasting much longer.

• If the objective is to reduce the memory of pain, then lower the peak intensity rather than the duration.
• If the objective is to reduce actual pain experienced, then conduct the procedure swiftly even if that raises peak intensity and leaves an awful memory.
• If you want a better memory of an unavoidable ordeal, then end on a mild note — prefer gradual relief to abrupt relief.
• If you are predicting someone's choice to repeat an experience, then assume they pick by peak-end memory and ignore duration entirely.
• If changing the length of an experience does not change how it is rated, then you are dealing with duration neglect and should not use added time as a lever for actual welfare.

• Optimizing the total experience while people judge and choose by memory — you deliver less suffering but they remember it as worse, and refuse to return. Tell: your metric improves but satisfaction ratings and repeat choices drop.
• Letting an experience end at its worst moment. Like patient A whose procedure ended at a bad point, the whole episode is remembered as awful regardless of a decent middle. Tell: overall ratings track the final minutes, not the average.
• Adding a mild tail to improve a memory without noticing you added real suffering — the cold-hand mistake of accepting 30 needless seconds of pain. Tell: people prefer the objectively longer, more painful option.
• Assuming people weight duration because they know the duration. Subjects knew which trial was longer but did not use it. Tell: correct factual recall of length coexists with a choice that ignores it.
• Treating the retrospective rating as if it measures the integral of the experience. It weights only two moments, peak and end. Tell: two experiences with identical peak-end but very different totals get the same rating.

• ▸Objective: state whether you are minimizing remembered pain or actual experienced pain
• ▸Peak intensity: record the worst (or best) moment's level on a fixed scale
• ▸End intensity: record the level at the final moment before it stops
• ▸Peak-end average: average the peak and end values — this predicts the memory
• ▸Duration: note total length, then confirm it will NOT drive your evaluation
• ▸Ending design: specify whether relief is gradual or abrupt, and why
• ▸Trade-off check: does improving the memory add real suffering to the experiencing self? Quantify it
• ▸Prediction: state which option a person will choose to repeat, based on lowest peak-end average

1. 1When you catch yourself judging your life or forecasting your future happiness, name the single aspect currently occupying your attention — the raise, the diagnosis, the move, the argument. That named item is the suspect.
2. 2Ask directly: how much of my day will this item actually occupy once it stops being novel? System 1 runs ahead of the facts, constructing a rich image from scraps; force System 2 to check whether the item is one thing among many or the whole picture.
3. 3Treat the vividness of the item as a warning sign, not evidence. Like the fins on the Müller-Lyer lines, the felt magnitude is an illusory pattern — learn to recognize it and recall what you know about it.
4. 4Deliberately widen the frame: list the boring, unchanging parts of the affected life (commute, meals, chores, sleep) that the salient item will not touch. Attention had ignored them precisely because they lacked drama.
5. 5Because System 1 cannot be turned off, do not try to feel differently — instead refuse to act on the impression. Distinguish the impression ('this will change everything') from the belief you adopt after checking ('this changes little of most days').
6. 6Reserve this effortful monitoring for high-stakes judgments — a career move, a relocation, a purchase compared for 'overall value.' Continuous vigilance is impractical; apply it where the mistake is expensive.

Compare two washing machines for overall value — a task the passages list as demanding System 2 attention. Because you are focused on price at this moment, that single dimension fills your attention and you forecast that the cheaper machine will make you happier. But the focusing illusion overweights whatever is currently in view: price is one aspect among many. Force the frame wider and the ignored, undramatic dimensions surface — noise, cycle time, water use, repair frequency over ten years — none of which had come to mind because none was salient. Just as the viewers counting basketball passes fail to see a gorilla in view for nine seconds, intense focus on price makes you effectively blind to the attributes that dominate daily use. You cannot stop price from feeling decisive; you can measure the other dimensions and choose to believe the fuller comparison instead.

• If one aspect feels like it will 'change everything,' then treat that certainty itself as the fin on the line — measure before you believe.
• If you cannot state how many hours per week the salient item will actually occupy, then you are forecasting from attention, not from evidence.
• If the ignored aspects of the affected life are boring and unchanging, then they are exactly the ones the focusing illusion is hiding from you — list them explicitly.
• If the stakes are low and the choice routine, then let System 1 run; reserve effortful reframing for consequential judgments only.
• If you notice your examples all came easily to mind, then suspect that availability, not importance, selected them.

• Trying to make yourself feel the item is less important — it will not work, because System 1 cannot be turned off; the tell is repeated failed attempts and returning frustration.
• Confusing the impression with the belief: you 'know' the item is one factor among many yet still act as though it dominates, exactly as you still see the lines as unequal after measuring.
• Building a rich causal story from a single vivid detail ('this move will fix my life') — System 1 runs ahead of the facts producing a representation that makes too much sense; the tell is a narrative with no boring parts.
• Assuming the currently salient dimension explains the whole outcome, the way the rural lifestyle was wrongly credited with both high and low cancer rates; the tell is an explanation that would fit the opposite result too.
• Applying vigilance everywhere until it becomes impossibly tedious, then abandoning it entirely; the tell is oscillation between overthinking and no checking at all.

• ▸Salient item: the one aspect now filling your attention
• ▸Predicted effect: what you feel it will do to your happiness
• ▸Actual daily footprint: estimated hours/week it will occupy once novelty fades
• ▸Ignored dimensions: three unchanging parts of the affected life it will NOT touch
• ▸Vividness flag: is the certainty itself the warning sign? (yes/no)
• ▸Impression vs belief: state each separately in one line each
• ▸Stakes level: high (reframe now) or low (let it run)
• ▸Decision after widening the frame: what you will act on, not what you feel

1. 1Separate the two selves before you measure anything: ask one question for the experiencing self (what emotional states dominate your moments now) and a distinct question for the remembering self (how satisfied you are with your life on reflection). They will not match.
2. 2Catch the substitution. When someone asks 'How happy are you with your life these days?' notice that System 1 answers the easier question 'What is my mood right now?' instead. Flag any life-evaluation answer that tracks momentary mood too tightly.
3. 3Check for duration neglect and the peak-end pattern in remembered experiences: a longer episode of pain can leave a better memory if it ends better, so the remembering self will choose to repeat the objectively worse experience. Ask about total time exposed, not just the memory.
4. 4Test for quantity insensitivity. If the well-being judgment does not scale with the size of what is at stake (2,000 vs 200,000 birds all rated near $80), you are measuring an emotional prototype, not a real evaluation.
5. 5Use intensity matching deliberately rather than by accident. Map the felt intensity of the state onto a defined scale (low = 'defeated in the primary' to high = 'president') so the judgment is anchored to amounts, not to whatever prototype System 1 volunteered.
6. 6Watch context and recency: recent events and current situation carry the most weight, distant memories only fill gaps. Record when and where the evaluation was made because the setting is biasing the number.
7. 7Do not rely on continuous vigilance. You cannot turn System 1 off, so instead of policing every judgment, build a fixed procedure (like separating the scoring of experience from the scoring of memory) that removes the bias structurally.

Take the two-selves pain study. Expose a person to two painful episodes; one is strictly worse because it lasts longer, but its memory is engineered to end better. The experiencing self suffers more total pain in the longer episode. Yet when the person later chooses which to repeat, the remembering self picks the longer one, exposing the experiencing self to unnecessary pain. This is the exact structure of a well-being error: the self that decides is not the self that feels. Apply it to a life-evaluation survey and the same split appears — 'How happy are you with your life these days?' gets answered as 'What is my mood right now?', so a person in a good mood over-rates a mediocre life. The tell is that the number moves with the moment and ignores the accumulated duration and quantity of what actually happened.

• If a happiness rating tracks current mood, then treat it as a mood report, not a life evaluation, and re-ask at a different time or in a different frame.
• If the choice being made serves the remembering self (which experience to repeat), then check it against the experiencing self's total exposure before trusting it.
• If willingness to pay or stated value does not scale with quantity, then you are seeing an emotional prototype and the judgment cannot be summed or compared across sizes.
• If the judgment arrived instantly and effortlessly, then suspect substitution — an easy question was answered in place of the hard one.
• If a strong positive impression forms early and later information seems to confirm it, then suspect a halo effect and score components independently and in isolation.

• Treating one 'happiness' number as if it captures both selves — the tell is that the answer never changes whether you frame it as a moment or as a life, because only one self was consulted.
• Letting the remembering self dictate future choices — the tell is repeating an objectively worse experience because it 'felt better looking back.'
• Quantity neglect in emotional judgments — the tell is that saving 2,000 and 200,000 birds get roughly equal valuations ($78 vs $88).
• Mistaking mood for evaluation — the tell is that the life-satisfaction score swings with weather, recent news, or the room you are sitting in.
• Believing you can overcome the bias by willpower — the tell is fatigue and inconsistency, because System 2 is too slow and lazy to substitute for System 1 on every judgment; only a structural procedure fixes it.

• ▸Experiencing-self prompt: rate the emotional states dominating your recent moments (not your life)
• ▸Remembering-self prompt: rate satisfaction with your life on reflection
• ▸Current mood check: name your mood right now to expose substitution
• ▸Context note: time, place, and any recent event weighting the answer
• ▸Quantity test: does the value scale with the size of what's at stake? yes/no
• ▸Duration record: total time actually spent in the experience being judged
• ▸Peak-and-end note: how the experience ended vs how it mostly felt
• ▸Choice check: which self is this decision serving, experiencing or remembering?
• ▸Gap flag: circle where the two selves disagree and which one you are about to obey

1. 1Locate the decision where Humans systematically underperform Econs — one involving fine print, delayed benefit, or effortful opt-in — because that gap is where a nudge produces gains without coercion.
2. 2Set the default to the option that serves the person's long-term interest, since the default is perceived as the normal choice and deviating from it requires effortful System 2 deliberation, more responsibility, and risk of regret.
3. 3Preserve the exit: make opting out a single low-cost act (check a box) so no one can claim freedom is diminished. If the escape is hard, you have crossed from paternalism into compulsion.
4. 4Reframe any immediate loss as a foregone gain — tie the costly change to a moment when the person is gaining anyway (e.g., a raise), so loss aversion no longer resists it.
5. 5Build in automaticity so the laziness of System 2 works for the person over time rather than against them; the change should proceed without requiring repeated re-decisions.
6. 6Simplify the contract or disclosure so a Human can actually read and understand it — reject the Econ assumption that disclosure alone protects people; small print and complex language defeat the purpose.
7. 7Test the design for cross-spectrum appeal: a legitimate nudge involves no misdirection or artifice, so it should be defensible to both conservatives and liberals who value free choice.

Save More Tomorrow, proposed by Thaler and Benartzi in 2003, is the flagship case. Employees who sign on let the employer raise their savings contribution by a fixed proportion each time they get a raise, and the increase runs automatically until they opt out. It works on three psychological principles at once: it requires no immediate change, so there is no immediate loss to resist; it ties saving to pay raises, converting a loss into a foregone gain that is far easier to bear; and its automaticity aligns System 2's laziness with the worker's long-term interest. Nobody is compelled to do anything, and there is no artifice. The plan was sponsored in Congress by a coalition of extreme conservatives and liberals, and has improved the savings rate of millions of workers. Its appeal has since been recognized in the UK and South Korea.

• If a choice harms only the chooser and freedom to exit is fully preserved, then a nudge is defensible; if the design removes the exit, then it is compulsion, not libertarian paternalism.
• If a beneficial change triggers loss aversion, then reframe it as a foregone gain by attaching it to a moment of gain (a raise), rather than asking for an active present sacrifice.
• If the person is unsure what to do, then set the long-term-interest option as the default, because the default is read as normal and inertia will carry it.
• If protecting people requires firms to reveal information, then mandate simple, readable contracts — because disclosure without comprehension protects no Human.
• If observed behavior looks odd, then first check whether the person has a good reason (Becker's rule) before invoking a psychological bias to justify intervention.

• Making the opt-out costly or hidden — the tell: you have to explain why leaving is hard, which means freedom is actually diminished.
• Relying on disclosure alone while keeping print small and language complex — the tell: the contract is technically transparent but no ordinary customer can act on it, and profits depend on that gap.
• Asking for an immediate loss instead of a foregone gain — the tell: uptake collapses because loss aversion is fighting the design head-on.
• Designing a nudge that serves the institution rather than the person's long-term interest — the tell: it functions as exploitation of System 1 quirks, the very thing nudges are meant to guard against.
• Extending nudge logic to cases with no satisfying answer — the tell: you are subsidizing decisions likely to end badly (Coelho's small-business loan problem) as if it were as clean as auto-enrollment in a pension plan.

• ▸Target decision: the choice where Humans underperform Econs
• ▸Long-term interest: the outcome that serves the chooser over time
• ▸Default setting: the option pre-selected as 'normal'
• ▸Opt-out mechanism: the single low-cost action to leave (state it)
• ▸Loss-to-gain reframe: what immediate loss you convert to a foregone gain, and how
• ▸Automaticity: what proceeds without re-decision, and its trigger
• ▸Simplification check: is the contract/disclosure readable by a Human? Y/N
• ▸Freedom test: is anyone compelled or misdirected? Y/N — must be No
• ▸Cross-spectrum check: would both conservatives and liberals accept it?

1. 1Notice when an impression, feeling, or answer arrives without effort — the belief that a lamp is on the desk, the hint of irritation in a spouse's voice. That instant, unbidden arrival is System 1's signature.
2. 2Ask what question you actually answered. When faced with a hard question, check whether an easier related question supplied the answer instead — 'Should I invest in Ford stock?' becoming 'Do I like Ford cars?' This substitution happens without your noticing.
3. 3Name the heuristic in play. If your judgment rests on resemblance to a stereotype, you are running the representativeness heuristic; if it rests on how easily examples come to mind, you are running the availability heuristic.
4. 4Look for the ignored statistical fact. System 1 fixes on the vivid cue (Steve's 'meek and tidy' personality) and skips base rates (20+ male farmers per male librarian). Deliberately retrieve the number you skipped.
5. 5Trace whether ease of recall, not actual frequency, drove your estimate. When you judged divorce rates among professors, you searched memory for cases and judged category size by retrieval ease — audit that source.
6. 6Ask whether coverage or drama inflated the impression. Kahneman's belief that adultery is more common among politicians dissolved once he realized politicians' transgressions are simply reported more often.
7. 7Use a demonstration on yourself, not a lecture. Reading the actual question and watching your own mind get tripped up is what makes the bias recognizable — abstract description does not.

Kahneman held a long-standing impression that adultery is more common among politicians than among physicians or lawyers, and he had even manufactured explanations for it — the aphrodisiac effect of power, the temptations of life away from home. This is System 1 at work: a quick, confident judgment paired with a fluent causal story. On examination he realized the transgressions of politicians are far more likely to be reported than those of lawyers and doctors. His 'fact' was produced entirely by the availability heuristic — the ease with which reported political scandals came to mind — plus journalists' choices of topics. The tell was his own confidence combined with ready-made explanations; the correction came only from asking why those instances were so retrievable.

• If an answer arrives instantly and effortlessly with high confidence, then treat that confidence as a symptom to check, not evidence — we are often confident even when wrong.
• If your judgment relies on how a case resembles a stereotype, then stop and ask for the base rate before concluding.
• If your estimate of frequency or importance tracks how easily examples come to mind, then suspect the availability heuristic and ask whether media coverage or memorability, not reality, is the source.
• If you can produce a fluent story explaining an impression, then note that the story is not verification — System 1 supplies coherence, not truth.
• If the original hard question has no ready answer but a related easy one does, then flag the substitution before acting on the easy answer.

• Mistaking fluent confidence for accuracy — the tell is that you cannot trace how you arrived at the belief, yet feel certain of it.
• Answering the easier substituted question and never noticing — the tell is that your answer feels obvious while the actual question was genuinely difficult.
• Reading about a bias in the abstract and assuming immunity — the tell is attributing errors to 'the fecklessness of undergraduates' rather than testing your own thinking on the demonstration.
• Judging category size by retrieval ease — the tell is that your estimate rises for dramatic, well-publicized, or recently mentioned instances (letters at word-start, reported scandals).
• Overriding relevant statistics with a vivid resemblance cue — the tell is that a personality sketch or image dominates and you never asked how many of each type actually exist.

• ▸Impression / quick judgment: state the answer that arrived automatically.
• ▸Confidence level: rate how certain it feels (this is a symptom, not proof).
• ▸Original question: what were you actually asked or trying to decide?
• ▸Substituted question: what easier question may your mind have answered instead?
• ▸Cue used: resemblance, ease of recall, emotion, or fluent story?
• ▸Ignored fact: what base rate or statistic did you skip?
• ▸Source of ease: is retrieval driven by media coverage, drama, or recency rather than frequency?
• ▸Correction: what does the evidence say once you retrieve the skipped data?

1. 1Notice when a question is difficult and no expert answer arrives instantly — that gap is your signal to engage System 2 rather than accept the first impression System 1 offers.
2. 2Check whether you are answering the original question or a substituted easier one (e.g., 'Do I like Ford cars?' instead of 'Should I invest in Ford stock?'). Name the actual question in words before answering.
3. 3Load the relevant rule or statistical fact into working memory alongside the intuitive impression. System 1 gives you the resemblance; only System 2 can combine it with base rates like '20 male farmers per male librarian.'
4. 4Adopt a task set — an explicit instruction that overrides your habitual response (e.g., 'count every f on this page'). Set it up deliberately and hold it until the task is complete.
5. 5Verify plausibility by bracketing the answer's range before computing: recognize implausible outputs (12,609 or 123 for a multiplication) even when you can't yet confirm the true value.
6. 6Budget attention deliberately. System 2 is resource-limited, so run the effortful check only on high-stakes or bias-prone judgments; expect muscle tension, raised heart rate, and dilated pupils as the cost signature.
7. 7Anticipate the 'informed gossip' — how a knowledgeable colleague would evaluate your reasoning — and pre-empt the critique. This external check motivates self-correction more reliably than good intentions.

Consider the executive deciding whether to invest in Ford stock. System 1 instantly delivers a warm impression because he likes Ford cars, and that feeling silently determines his choice — he has answered the easy question without noticing the substitution. Engaging System 2 means first stating the original question in words: 'What is the expected return on Ford stock?' That reframing exposes that liking the cars is irrelevant to the answer. System 2 then must hold multiple distinct considerations in memory at once — earnings, sector conditions, valuation — and combine them by rule, which System 1 cannot do. This is effortful and orderly, the same strain you feel proceeding step-by-step through a two-digit multiplication like 17 × 24. If the executive skips this and stays with the affective impression, he commits the classic substitution error; if he engages System 2, he at least has a shot at correcting it.

• If an answer arrives instantly and effortlessly to a genuinely hard question, then suspect substitution and re-read the original question before acting.
• If a judgment requires combining two or more distinct facts — especially resemblance plus a base rate — then hand it to System 2, because System 1 integrates one thing well but cannot juggle multiple topics or statistics.
• If the task overrides a habitual response (counting f's, following an unfamiliar rule), then explicitly adopt a task set and expect it to feel effortful until practiced.
• If the stakes are low or the intuitive answer is an expert one (a chess master's move), then let System 1 stand rather than spending scarce attention.
• If your impression feels like a 'fact' you've explained to yourself, then check whether availability produced it — as with 'politicians commit more adultery,' which was really just journalists' reporting choices.

• Assuming you are the reasoning System 2 in control, when System 1 was the secret author of the choice. Tell: you can justify a conclusion but cannot reconstruct how you actually arrived at it.
• Letting System 2 stay lazy and endorse System 1's impression without checking. Tell: no felt effort — no strain, no concentration, no sense of holding material in memory.
• Answering the substituted easy question. Tell: your reasons are about something adjacent (liking the product) rather than the question asked (its investment value).
• Having the correcting fact but not applying it — knowing there are 20 farmers per librarian yet still judging by resemblance. Tell: you can recite the base rate afterward but it played no role in the judgment.
• Overloading working memory so the check collapses. Tell: you lose track of intermediate results, as in near-limit tasks like two-digit multiplication or Add-3.

• ▸Original question (verbatim): write the exact question you must answer, in words.
• ▸Easy-question test: what easier question might I be answering instead?
• ▸Intuitive impression: the fast answer System 1 supplied, and how it feels.
• ▸Relevant rule or base rate: the statistical fact or rule the answer requires.
• ▸Combination step: how I hold the impression and the fact together to decide.
• ▸Plausibility bracket: the highest and lowest answers I'd immediately reject.
• ▸Task set: the explicit instruction I'm adopting to override habit.
• ▸Effort check: am I actually straining (concentration, attention) or coasting?
• ▸Gossip test: how would an informed colleague critique this reasoning?

1. 1Treat mental effort as a finite budget, not a switch: before starting demanding work, note that self-control, monitoring, and effortful computation all draw from the same limited pool, so schedule the hardest cognitive task when nothing else is competing for it.
2. 2Reserve System 2 engagement for questions where no expert answer and no heuristic answer arrives quickly — the signal is that you feel you 'could choose whether or not to engage in the computation' rather than an answer surfacing automatically.
3. 3Serialize demanding tasks. Because effortful activities interfere with each other, never pair two high-load operations (e.g. computing 17 × 24 while turning into dense traffic). Pair a demanding task only with an easy, undemanding one.
4. 4Reduce the physical and self-control load surrounding the work: sit rather than stand, be still rather than move fast, remove the urge-resistance costs (like maintaining a fast walking pace) that quietly drain the same budget.
5. 5Assign explicit priority when overload threatens. System 2 protects the top-priority task and starves the rest second by second, so name the one activity that must not fail before you begin.
6. 6Monitor arousal as a gauge of load: pupil dilation, raised heart rate, tensed muscles, and blindness to obvious stimuli signal you are near capacity. When you notice the tells, stop stacking tasks.
7. 7Use System 2's programming ability deliberately — set attention and memory in advance (search for the white-haired woman, remember 'drive on the left') — but expect that holding such a set requires continuous effort and will decay if you overload.

Take the multiplication 17 × 24. A precise answer does not come to mind, and you sense you can choose whether to engage. When you commit, you retrieve the school-learned program for multiplication and implement it step by step. The strain is real: you hold intermediate results in memory while tracking where you are and where you're going. Your body participates — muscles tense, blood pressure rises, heart rate climbs, and your pupils dilate, then contract the instant you reach 408 or give up. Now try to run this while making a left turn into dense traffic: you can't, and you shouldn't — the two efforts compete for the same limited resources. This is why a walking companion asked to compute 23 × 78 immediately stops in his tracks.

• If an answer to a hard question arrives instantly, suspect System 1 substituted an easier question (do I like Ford cars?) for the real one (should I invest in Ford?) — then deliberately engage effort to check.
• If two tasks both demand attention, then do not run them together; if one is easy and undemanding (empty highway, routine pleasantries), then concurrency is safe.
• If you must construct an intricate argument under time pressure, then be still and sit — reserve the budget for thinking, not for movement or self-control.
• If overload is imminent, then declare one priority task in advance; the rest will be starved automatically, so choose which failures you can tolerate.
• If you notice dilating pupils, raised heart rate, or blindness to obvious stimuli, then you are at peak load — stop adding tasks rather than pushing through.

• Answering the easy question while believing you answered the hard one — the tell is a fast, confident answer to a genuinely difficult problem with no felt strain.
• Stacking two effortful tasks and assuming both are handled — the tell is degraded performance you don't notice, like missing a target you're staring straight at during peak effort.
• Trusting intuitive confidence as accuracy — the tell is being 'confident even when wrong,' with an outside observer catching errors you can't see.
• Ignoring the self-control cost of the surrounding situation — the tell is coherent thinking collapsing as you speed up or resist an urge, even though the intellectual task itself hasn't changed.
• Trying to maintain an attentional set while overloaded — the tell is the set silently lapsing (you stop scanning for the white-haired woman) without you registering the failure.

• ▸Task: name the single cognitive task requiring System 2
• ▸Trigger check: did an answer arrive automatically? (yes = suspect substitution / no = genuine effort needed)
• ▸Priority rank: is this the one task that must not fail? (yes/no)
• ▸Competing loads: list anything else demanding attention right now
• ▸Body posture: sitting / standing / moving — reduce to lowest-load option
• ▸Self-control drains present: urges being resisted, pace being maintained
• ▸Arousal tells to watch: pupils, heart rate, blindness to obvious cues
• ▸Concurrency verdict: safe to pair only with an easy, undemanding task? (yes/no)
• ▸Stop condition: what tell tells you you've hit capacity and must stop

1. 1When you form a fast judgment, name the target question you meant to answer (e.g., 'Is Ford stock underpriced?') and write it down explicitly.
2. 2Ask what easier question your mind actually answered instead (e.g., 'Do I like Ford cars?'). The technical definition of a heuristic is a simple procedure that gives adequate but often imperfect answers, so look for the substitution.
3. 3Identify which shortcut is running: substitution (easier question), representativeness (resemblance to a stereotype), availability (ease of recall), affect (liking/disliking driving the call), or anchoring.
4. 4Test for availability by asking whether your impression comes from actual frequency or from what is dramatic and easily reported—as with judging politicians' transgressions as common because media report them more than those of doctors or lawyers.
5. 5Test for affect by checking whether your conclusion tracks how much you like the object rather than the diagnostic evidence the question requires.
6. 6Attach a diagnostic label to the pattern (halo effect, availability, affect heuristic); a named category makes the bias easier to anticipate, recognize, and discuss.
7. 7Decide whether expertise is actually present. Accurate intuition comes from prolonged practice (chess master, firefighter), not from a shortcut—if you lack relevant expertise, treat the fast answer as a heuristic answer, not knowledge.
8. 8When a shortcut is detected on a consequential judgment, switch deliberately to slow, effortful System 2 reasoning and answer the original target question.

A chief investment officer told Kahneman he had just invested tens of millions of dollars in Ford stock. Asked how he decided, he said he had attended an auto show and been impressed: 'Boy, do they know how to make a car!' The target question—Is Ford stock currently underpriced?—is hard and no expert answer was available. So his mind substituted an easier related question: Do I like Ford cars? His feelings of liking the cars, the company, and the idea of owning the stock guided the decision directly, with little reasoning: the affect heuristic. He never noticed the substitution and was satisfied with himself, yet given what is known about stock-picking accuracy, he likely did not know what he was doing.

• If the target question is hard and no expert-recognized solution comes to mind, then assume any quick answer is a substituted heuristic answer, not a real answer.
• If your judgment of frequency rests on how easily examples come to mind rather than counted data, then suspect availability and ask what dramatic reporting is distorting the sample.
• If your conclusion moves in lockstep with how much you like or dislike the object, then flag the affect heuristic and separate liking from the diagnostic evidence.
• If a description resembles a stereotype (a 'meek and tidy soul with a passion for detail'), then check the base rates before letting resemblance drive the probability.
• If you have prolonged, relevant practice in the exact situation, then trust the fast intuition; if not, engage System 2.

• Not noticing the substitution—answering the easy question while believing you answered the hard one; the tell is fluent confidence with no reasoning you can defend.
• Mistaking availability for frequency—assuming reported events are common (politicians' transgressions vs. doctors'); the tell is that your examples are the ones the media covers most.
• Treating affect as analysis—liking the product and calling that an investment thesis; the tell is a gut feeling ('impressed at the auto show') standing in for the relevant metric.
• Overtrusting intuition where no expertise exists—assuming a firefighter's or chess master's kind of accuracy applies to you; the tell is confidence without prolonged practice in that specific domain.
• Confidence outrunning correctness—being sure while wrong, when an objective observer would spot the error; the tell is that you can neither explain nor defend the evidence behind the belief.

• ▸Target question: the assessment you actually intended to produce
• ▸Answer you gave: the quick judgment that came to mind
• ▸Heuristic question: the easier question you may have answered instead
• ▸Shortcut in play: substitution / representativeness / availability / affect / anchoring
• ▸Availability check: is this based on counted frequency or on ease of recall?
• ▸Affect check: does my answer track how much I like/dislike the object?
• ▸Expertise check: do I have prolonged practice in this exact situation? (yes/no)
• ▸Diagnostic label: name the bias so you can recognize it next time
• ▸System 2 action: the deliberate step needed to answer the target question

1. 1Name the bias before diagnosing it. Systematic errors recur predictably; acquiring a diagnostic label (halo effect, availability heuristic) makes the error easier to anticipate, recognize, and understand — treat vocabulary as the first tool.
2. 2Identify which heuristic is doing the work. When a hard question (probability, frequency, importance) gets answered instantly, check whether the answer came from resemblance (does X look like a stereotype?) or availability (how easily do instances come to mind?).
3. 3Surface the statistical fact the intuition ignored. For any judgment cued by resemblance, ask what the base rates are — e.g., the ratio of the categories — and force that number into view before accepting the intuitive answer.
4. 4Test whether ease of retrieval, not real frequency, is driving your impression. Ask: is this topic vivid because it's common, or because it's dramatic and heavily reported? Trace the retrieval path back to its source.
5. 5Accept that you cannot switch the bias off — build a recognition rule instead. Like the Müller-Lyer illusion, System 1 keeps producing the false impression even after you know better; the only defense is learning to distrust impressions in the specific circumstance that triggers them.
6. 6Run the judgment as if anticipating intelligent gossip. Imagine informed colleagues evaluating your choice using precise labels; the expectation of that critique is a stronger motive for self-criticism than a private resolution.
7. 7Distinguish heuristic-driven guesses from genuine expert intuition. Prolonged practice produces accurate intuitions (the firefighter commander's silent alarm); a rule of thumb applied to an unfamiliar problem produces bias. Ask whether real skill was actually available.

Consider the librarian question: Steve is described as meek, tidy, orderly, detail-obsessed, and you are asked whether he is more likely a librarian or a farmer. The stereotype resemblance strikes everyone immediately and points to 'librarian.' But the operative statistical fact is that there are more than 20 male farmers for each male librarian in the United States. Because the farmer category is roughly twenty times larger, it is almost certain that more meek and tidy souls sit on tractors than at library desks. Participants ignored this base rate and answered from resemblance alone — a predictable bias produced by the representativeness heuristic. The corrective step is not to feel the resemblance less (you can't) but to force the 20-to-1 ratio into the judgment before committing to an answer.

• If a question feels instantly answered but is actually hard (probability, frequency, importance), then assume a heuristic supplied the answer and check it against a base rate.
• If your estimate of how common something is came from how easily examples popped to mind, then suspect availability — and ask whether media coverage or drama, not frequency, seeded those examples.
• If a description resembles a stereotype, then find the size of the competing categories before judging which is more likely.
• If an intuition survives after you know the fact contradicts it (you still 'see' the lines as unequal), then stop trying to feel differently and instead apply a learned recognition rule to the triggering situation.
• If the judgment rests on genuine, practiced expertise rather than a shortcut applied to an unfamiliar case, then the intuition may be trustworthy — verify that real experience with this exact pattern exists.

• Attributing errors to other people's fecklessness — dismissing biases as the failing of undergraduates or the unintelligent. Tell: you feel immune while reading the demonstration instead of catching your own thinking tripping.
• Building post-hoc explanations for a biased impression. The author 'knew' adultery was commoner among politicians and even invented reasons (aphrodisiac of power) before realizing it was just reporting frequency. Tell: you have a confident causal story for a 'fact' you never checked.
• Confusing availability with importance in public or organizational judgment. Tell: the issues on your mind are the dramatic, celebrity-driven ones (Michael Jackson coverage), while unexciting but critical issues never surface.
• Believing that knowing about a bias removes it. Tell: you say 'I understand the halo effect' yet still rate the confident, handsome speaker more favorably than he deserves.
• Ignoring base rates because the individuating story is vivid. Tell: you answer 'librarian' and the 20-to-1 farmer ratio never entered your reasoning.

• ▸Judgment made: state the intuitive answer you reached in one line.
• ▸Question type: is this really about probability, frequency, or importance? Name it.
• ▸Heuristic suspected: resemblance/representativeness, availability, or other — pick one.
• ▸Retrieval source: where did the supporting instances or impression come from (memory ease, media, stereotype)?
• ▸Base rate check: what is the relevant statistical fact or category size you have not consulted?
• ▸Illusion persistence: does the impression remain even after you know the fact? Yes/No.
• ▸Recognition rule: write the specific 'distrust this impression when ___' rule for this trigger.
• ▸Gossip test: how would an informed colleague using the correct label critique this judgment?