Decision-making & Reasoning

Evidence

Claims data can bear on.

Essay: The catalogue of predictable errors

Attacking the person rather than the claim, which leaves the claim untouched.

"He's not a climate scientist, he's a geologist" leaves the argument untouched. It becomes legitimate only when the person's reliability is the actual question, as with a witness's record of perjury.

In practice

"You would say that, you work for an oil company." The claim about emissions data is still either right or wrong.
A reviewer dismisses a study because the author has a book to sell, without opening the methods section.

Attributed to: Classical rhetoric
Standing: Not fallacious where the person's credibility is itself the matter at issue.
Further reading: Wikipedia

Page for this entry

Treating wide agreement as evidence of truth.

A survey showing that four out of five dentists recommend a product is evidence about dentists' opinions; it does not by itself establish that the product is best. The fallacy is treating popularity as sufficient proof, not treating popularity as data.

In practice

Nine of the ten agencies use the same forecasting model, which is offered as evidence the model is sound. All ten licensed it from the same vendor.
A restaurant is chosen because the queue is longest, and the queue is longest because it was chosen.

Attributed to: Classical rhetoric
Also called: Appeal to popularity; bandwagon fallacy

Page for this entry

Treating the authority of the speaker as settling a question outside their competence.

A Nobel laureate in physics does not settle a nutrition question merely by virtue of the Nobel; authority carries most weight when it is relevant to the field and supported by evidence. Linus Pauling's vitamin C advocacy is the standard case: two chemistry-adjacent Nobels, and a claim the evidence never supported.

In practice

A cardiologist's opinion on a school funding formula is quoted at length because he is a doctor.
A committee defers to the most senior person present on a question about a software licence none of them has read.

Attributed to: Classical rhetoric
Also called: Appeal to authority
Standing: Legitimate deference to relevant expertise is not this fallacy.

Page for this entry

Judgements of risk and benefit follow how an option feels rather than what its statistics say.

People judge nuclear power as both high-risk and low-benefit, and vaccines as both low-risk and high-benefit. In reality risk and benefit are largely independent; the correlation lives in the feeling, not the world.

In practice

A board rejects a warehouse automation project as risky while approving a larger untested marketing spend, because one feels mechanical and the other feels like growth.
A parent bans a trampoline and drives the children forty minutes to sport four times a week.

Attributed to: Paul Slovic
Further reading: Wikipedia

Page for this entry

An initial number biases later numerical judgements even when it is known to be arbitrary.

Tversky and Kahneman had subjects spin a wheel rigged to stop at 10 or 65, then estimate the percentage of African countries in the UN. The wheel moved answers by 20 points. Asking price does the same work in negotiation.

In practice

A salary negotiation opens at £58,000 and settles near it, whatever either side privately thought the role was worth.
A builder's first quote sets the frame for every later conversation, including the one where you feel you have won a discount.

Attributed to: Tversky & Kahneman
Further reading: Wikipedia

Page for this entry

Frequency and probability are judged by how easily examples come to mind.

People rate air travel as more dangerous than driving because crashes are memorable. Vivid, heavily reported hazards can feel more frequent than mundane hazards that kill far more people.

In practice

A team estimates its outage rate from the two incidents it remembers, not the log, and is confident and wrong.
A homeowner insures generously against fire and not at all against the flood that is likelier on their street.

Attributed to: Tversky & Kahneman
Further reading: Wikipedia

Page for this entry

Vivid case detail displaces the underlying prevalence, inflating confidence in rare conclusions.

Given a description of a shy, tidy man, people judge him more likely a librarian than a farmer, ignoring that farmers vastly outnumber librarians. In medicine it produces the standard error of reading a 95%-accurate test on a rare disease as 95% likely to be right.

In practice

A screening test 99% accurate for a condition affecting one in ten thousand returns a hundred false positives for every true one, and every patient hears "99% accurate".
A fraud model flags an account as "highly likely" fraudulent, and the analyst forgets that almost no accounts are fraudulent.

Attributed to: Kahneman & Tversky
Also called: Base rate neglect
Further reading: Wikipedia

Page for this entry

Beliefs survive the discrediting of the evidence that produced them.

Ross and Lepper gave subjects fake feedback on a task, then fully debriefed them. Self-assessments remained shifted afterwards — the belief outlived the evidence that had been explicitly withdrawn.

In practice

A team is shown that the incident data behind their reliability estimate was miscoded, and the estimate persists in every subsequent conversation.
A candidate is cleared of the allegation that ended their shortlisting, and is not re-shortlisted.

Attributed to: Ross & Lepper
Further reading: Wikipedia

Page for this entry

People recognise bias in others' judgement while believing their own introspection to be accurate.

Emily Pronin's studies found people rated themselves less susceptible than average to biases just described to them. The blind spot survives being told about the blind spot.

In practice

A manager who has just read about confirmation bias uses it to explain her deputy's position and not her own.
Everyone on the panel agrees that interview scores are unreliable, and each believes their own scores are the exception.

Attributed to: Emily Pronin
Further reading: Wikipedia

Page for this entry

People frequently fail to notice when the option they receive is not the one they chose, and will then explain at length why they chose it.

Johansson and Hall showed subjects pairs of faces, took their choice, and returned the other face by sleight of hand. Most did not notice, and went on to justify the preference they had not expressed. The finding bears less on memory than on introspection: the explanation is constructed after the fact from whatever option is in hand.

In practice

A committee defends the option it believes it selected, having been handed the minutes of a different decision.
A customer given the wrong variant explains why that variant suits them.

Attributed to: Petter Johansson, Lars Hall and colleagues
Also called: Undetected choice substitution

Page for this entry

After choosing, people remember the chosen option as better and the rejected one as worse than they judged them at the time.

Mather and Johnson found people misremember their own past ratings in the direction of the choice they made. It makes retrospective accounts of a decision unreliable evidence about the decision.

In practice

Someone who agonised between two job offers remembers the rejected one as obviously worse within six months.
A buyer who chose between two houses recalls the other as having problems that were not in their notes at the time.

Attributed to: Mather & Johnson
Further reading: Wikipedia

Page for this entry

Random sequences are read as patterned, because true randomness contains more streaks than people expect.

In 30 coin flips a run of five is more likely than not. Cancer-cluster investigations repeatedly find that geographic clumping is what randomness looks like at small sample sizes.

In practice

Four customer complaints in one week from the same postcode prompts a regional investigation; over a year, some week and some postcode had to be first.
A trader has six good months and is promoted; the sequence is well within what a coin would produce.

Attributed to: Thomas Gilovich
Further reading: Wikipedia

Page for this entry

People shift cognitive work onto the environment whenever doing so is cheaper than doing it internally, and thereby lose the internal capability.

Risko and Gilbert's 2016 review treats note-taking, rotating the page, setting reminders and searching rather than recalling as one phenomenon governed by a cost comparison. The offloading is usually rational, and it reliably reduces unaided performance, so the measured decline in memory for offloaded material is a consequence of the strategy rather than evidence of damage.

In practice

A team that documents thoroughly cannot answer the same questions without the document, which is what documenting bought.
A driver who has used satellite navigation on a route twenty times cannot describe it.

Attributed to: Evan Risko & Sam Gilbert
Also called: External memory use

Page for this entry

Given three options, people disproportionately choose the middle one, so the extremes shape the choice they do not win.

Simonson's 1989 finding. Adding an expensive third option shifts sales to the middle, which is why menus and pricing tiers are built with an option nobody is expected to buy.

In practice

A consultancy quotes £120,000, £45,000 and £22,000 for the same problem, and the £45,000 option is what the pricing was designed to sell.
A wine list with one very expensive bottle sells more of the second-most-expensive.

Attributed to: Itamar Simonson
Also called: Extremeness aversion

Page for this entry

People seek, notice, and recall evidence that fits what they already believe, and set the rest aside.

In Wason's 1960 rule-discovery task, many participants generated tests expected to fit their current hypothesis rather than tests designed to rule it out; only six of twenty-nine reached the correct rule without first announcing an incorrect one.

In practice

An analyst convinced a supplier is unreliable notices every late delivery and codes the early ones as luck.
A manager who has decided a hire was a mistake reads their next four pieces of work as confirmation.

Attributed to: Peter Wason
Also called: Myside bias
Further reading: Wikipedia

Page for this entry

A detailed conjunction is judged more probable than one of its own components.

Tversky and Kahneman's Linda problem: told Linda is outspoken and concerned with justice, most respondents rate "bank teller and feminist" more probable than "bank teller." The conjunction cannot be more probable than its component.

In practice

A risk register entry reading "supplier insolvency during Q3 following the regulatory decision" feels more plausible than "supplier insolvency", and is necessarily less likely.
A forecast that names the mechanism in detail is believed more and should be believed less.

Attributed to: Tversky & Kahneman
Also called: The Linda problem
Further reading: Wikipedia

Page for this entry

Treating correlation as causation without ruling out reverse causation or a common cause.

Ice cream sales and drowning both rise in summer. The harder version is reverse causation: exercise and health correlate, and the causation runs both ways.

In practice

Firms with a wellbeing programme have lower absence, and healthier firms are also the ones that can afford the programme.
Students who attend revision classes score higher, and the students who attend are the ones who were going to revise anyway.

Attributed to: Classical logic
Also called: Correlation is not causation

Page for this entry

Logical rules for negating complex statements; useful for clear thinking about opposites.

De Morgan's 1847 laws: not (A and B) equals (not A) or (not B). Their practical use is in specification and negotiation, where negating a compound condition is where most reasoning errors occur.

In practice

A contract clause requiring "not late and not incomplete" is satisfied by being either on time or complete, unless the negation is worked through.
A policy prohibiting staff who are "not qualified and not supervised" permits the unqualified if supervised, which is what the drafter meant and not what most readers take.

Attributed to: Augustus De Morgan
Further reading: Wikipedia

Page for this entry

Judgement quality degrades across a long sequence of decisions, biasing towards defaults.

The 2011 Israeli parole study reported sharp session-order effects, but later analysis showed the case ordering was not random and identified confounds including legal representation and prison scheduling. The ego-depletion mechanism has also largely failed to replicate, so neither the field study nor the lab account is clean evidence.

In practice

A recruiter reviewing forty CVs picks more conventionally at 5pm than at 10am, and would deny it.
A shopper who has compared eleven washing machines buys the one the salesperson recommends, having stopped deciding an hour ago.

Attributed to: Roy Baumeister
Possible explanations: One explanation is depletion, in which the capacity for deliberate control runs down with use. Another is motivation, since people become less willing rather than less able. A third is scheduling, where what looks like fatigue tracks breaks, case order and workload.
In its favour: Defaults do dominate decisions made late in long sequences, and the pattern has been reported in several field settings.
Against it: Ego depletion, the mechanism behind it, has largely failed to replicate, and the parole study most often cited has been challenged for non-random case ordering and confounds.
Standing: Ego depletion has largely failed to replicate, and the parole field study has been challenged on non-random case ordering and confounds.
Further reading: Wikipedia

Page for this entry

Adding a deliberately inferior third option shifts choice between the original two.

Huber, Payne and Puto's 1982 finding. The Economist's famous subscription page offered print-only at the same price as print-plus-web; almost nobody chose it, and its presence shifted buyers to the bundle.

In practice

A consultancy offers a £90,000 package, a £150,000 package, and a £155,000 package that includes everything in the £150,000 one plus training.
A cinema's medium popcorn exists to make the large look reasonable.

Attributed to: Huber, Payne & Puto
Also called: Asymmetric dominance effect
Further reading: Wikipedia

Page for this entry

Defaults strongly influence choices, because many people accept the preselected outcome rather than actively switching.

Johnson and Goldstein's 2003 online experiment found about 42% donor consent under opt-in against 82% under opt-out. Their much-quoted cross-country figures were effective consent rates rather than a controlled estimate of actual donation, so the experiment carries the causal weight.

In practice

A pension scheme with automatic enrolment reaches 90% participation; the identical scheme requiring a form reaches a quarter of that.
A cookie banner where "accept all" is one click and refusal is four determines what most visitors choose.

Attributed to: Johnson & Goldstein
Possible explanations: One explanation is effort, since switching costs attention. Another is implied endorsement, as people read the preselected option as advice. A third is loss aversion, with the default acting as the reference point.
In its favour: The experimental evidence is strong and replicates across domains, including pension enrolment, where automatic entry raises participation sharply.
Against it: The cross-country organ donation figures usually quoted are effective consent rates rather than a controlled estimate, and actual donation depends on family consent and hospital infrastructure, so defaults do less than the headline suggests.
Standing: The experimental evidence is strong; the much-quoted cross-country organ-donation figures are effective consent rates, not a controlled estimate.
Further reading: Wikipedia

Page for this entry

People will act once an uncertainty is resolved either way, yet decline to act while it remains open, violating the sure-thing principle.

Shafir and Tversky's 1992 finding: students would buy a holiday if they passed the exam, and would buy it if they failed, but would pay to defer the decision until they knew. The reason differs in each branch, and no reason exists until one is chosen.

In practice

A buyer will commit if the funding round closes and will commit if it fails, and will not commit while it is open.
A team will proceed whether or not the reorganisation happens, and pauses anyway until it is announced.

Attributed to: Shafir & Tversky

Page for this entry

Merely owning something raises the price at which a person will part with it.

Thaler's mug experiments: people given a mug demanded roughly twice what people not given one would pay for it. Recent work suggests the effect is weaker than first reported and partly an artefact of the exchange procedure.

In practice

A team asked to give up a meeting slot they were assigned defends it far harder than they would have fought to get it.
A household prices its own car at what it feels worth to own, and every buyer prices it at what it is worth to buy.

Attributed to: Richard Thaler
Possible explanations: One explanation is loss aversion, with ownership setting the reference point so that giving up counts as a loss. Another is that owning a thing draws attention to its good features. A third is procedural, since unfamiliar exchange tasks can confuse participants.
In its favour: The finding replicates widely, and it fits field evidence on inertia in trading and on how hard people find it to sell inherited assets.
Against it: It is weaker than first reported, shrinks with market experience, and diminishes under procedures designed to remove confusion about the task.
Standing: Weaker than first reported, and partly an artefact of the exchange procedure used to measure it.
Further reading: Wikipedia

Page for this entry

Shifting the meaning of a term mid-argument so a true premise appears to support a different conclusion.

"Nothing is better than eternal happiness; a ham sandwich is better than nothing" is the schoolroom version. The serious cases turn on words like "freedom", "natural" and "discrimination" shifting sense between premises.

In practice

A contract defines "material" precisely in clause 3 and uses it loosely in clause 19, and the dispute is entirely about which meaning governs.
A policy about "flexible working" means location to one side of the table and hours to the other, and both sign it.

Attributed to: Classical logic
Further reading: Wikipedia

Page for this entry

Decision-makers increase investment in a failing course to justify their earlier choice.

Staw's 1976 studies. The mechanism is self-justification rather than sunk cost arithmetic: the effect is strongest when the same person made the original decision and must now be seen to have been right.

In practice

A programme director who chose the vendor authorises a further £2m rather than concede the selection was wrong.
A homeowner who has spent £30,000 on a renovation approves the next £10,000 because of the £30,000, which is gone either way.

Attributed to: Barry Staw
Also called: Throwing good money after bad
Further reading: Wikipedia

Page for this entry

People overestimate how widely their own views and habits are shared.

Ross, Greene and House's 1977 study: students asked to wear a sandwich board estimated that most others would make the same choice they had, whichever choice that was.

In practice

A hiring panel that all found the interview process reasonable concludes candidates find it reasonable.
A management team that likes the new office layout assumes the staff do, and the survey says otherwise.

Attributed to: Ross, Greene & House
Further reading: Wikipedia

Page for this entry

Presenting two options as exhaustive when others exist.

"Either we do this or we do nothing" is the standard institutional form. The test is whether the excluded middle actually exists, not whether the two named options are genuinely different.

In practice

"Either we ship on Friday or we lose the customer" survives until someone asks about shipping half of it on Friday.
A board is offered redundancies or insolvency, and the third option is a conversation nobody has scheduled.

Attributed to: Classical logic
Also called: False dichotomy; either-or fallacy
Further reading: Wikipedia

Page for this entry

Nothing matters as much as you think it does while you are thinking about it.

Kahneman and Schkade's finding that Californians and Midwesterners report similar life satisfaction, while both groups predict Californians are happier. The climate matters only while you are asked about the climate.

In practice

An employee weighing a job offer fixates on the commute and finds, a year in, that it barely features in how they feel about the work.
A buyer choosing between two flats agonises over the kitchen and, two years later, thinks about the neighbours.

Attributed to: Kahneman & Schkade
Further reading: Wikipedia

Page for this entry

Forecasters who hold many loose models beat those who apply one grand theory.

Tetlock's twenty-year forecasting tournament found the best predictors held many small models loosely, while single-framework experts did worse than chance on their own specialities.

In practice

A forecaster who tracks eleven weak indicators beats one who applies a single elegant theory of the cycle.
A doctor who holds four possible diagnoses loosely outperforms one committed to the most interesting.

Attributed to: Philip Tetlock

Page for this entry

Logically equivalent descriptions of the same choice produce different decisions depending on gain or loss framing.

Tversky and Kahneman's Asian disease problem: a programme saving 200 of 600 and one letting 400 die are the same programme, and preferences reverse between them.

In practice

A savings plan described as "keep 92% of your income" is chosen over one described as "save 8%".
A surgical consent form quoting a 5% complication rate is heard differently from one quoting a 95% success rate.

Attributed to: Tversky & Kahneman
Further reading: Wikipedia

Page for this entry

Dates that mark a boundary, such as a birthday or the start of a week, increase the rate at which people begin working toward goals.

Dai and colleagues found searches for diets, gym attendance and commitment-contract creation spiking after temporal landmarks. The proposed mechanism is that a landmark separates the imperfect past self from a present one with a clean record, which makes aspirational behaviour feel available. The effect is real and the durability of the behaviour it starts is another matter.

In practice

A change programme launched on the first of the month gets more participation than the same programme on the fourteenth.
A resolution made on a birthday outlasts one made on an ordinary Tuesday, briefly.

Attributed to: Hengchen Dai, Katherine Milkman & Jason Riis
Also called: Temporal landmarks

Page for this entry

Independent events are expected to self-correct, so a run of one outcome is taken to make the other due.

After four reds, black is not more likely. The Monte Carlo casino in 1913 saw black come up 26 times running, and the house took a fortune from players betting on the correction.

In practice

An investor who has had four losing quarters increases position size because a good one is due.
A hiring manager who has rejected six candidates in a row starts believing the next must be better, and it is the same pool.

Attributed to: Probability and judgement research
Also called: Monte Carlo fallacy
Further reading: Wikipedia

Page for this entry

Overwhelming an opponent with many weak claims, since refuting each costs more time than making it.

Named by Eugenie Scott for the creationist Duane Gish. It is Brandolini's law weaponised as a format: the asymmetry is exploited deliberately rather than merely suffered.

In practice

A consultation response listing forty objections forces a fifty-page reply, and thirty-eight of the objections were never serious.
A witness answers a question with eleven claims, and the cross-examination has time for two.

Attributed to: Named for Duane Gish by Eugenie Scott
Standing: The tactical application of Brandolini's law.
Further reading: Wikipedia

Page for this entry

Where the best option can only be identified by combining information distributed across members, groups usually fail to find it and choose the option the shared information favours.

The paradigm distributes clues so that each member, reading only their own, prefers an inferior candidate, while the full set favours another. Groups reliably pick the inferior one. The failure is not of reasoning but of pooling: the information required was present in the room and never assembled.

In practice

A committee with all the facts between them selects the option every individual preferred on partial information.
An investigation finds afterwards that four people each held a piece and no meeting ever put them together.

Attributed to: Garold Stasser & William Titus
Also called: Hidden profile paradigm
Standing: The experimental complement to shared information bias: the profile is hidden because no individual holds enough of it.

Page for this entry

Once an outcome is known it seems to have been predictable all along.

Fischhoff's 1975 studies: told an outcome, subjects misremembered their own prior probability estimates as closer to it. It is why post-mortems overstate what was knowable and why decision logs are worth keeping.

In practice

An inquiry concludes the warning signs were obvious, and the report reads the 40 signals that mattered without the 4,000 that did not.
A manager reviewing a failed hire remembers the interview doubts and not the six candidates they had the same doubts about who worked out.

Attributed to: Baruch Fischhoff
Also called: Knew-it-all-along effect
Further reading: Wikipedia

Page for this entry

Streaks in independent trials are read as evidence of a temporary skill state.

Gilovich, Vallone and Tversky's 1985 paper found no streak effect in basketball shooting. Miller and Sanjurjo showed in 2018 that their method contained a subtle bias, and a small real hot hand survives — the original claim is now qualified rather than overturned.

In practice

A fund manager with four good quarters is given more capital, which is a bet on a pattern the data barely supports.
Note the correction: Miller and Sanjurjo showed the original method was biased, so a small genuine hot hand survives and the confident dismissal was also wrong.

Attributed to: Gilovich, Vallone & Tversky
Possible explanations: One explanation is that streaks of the length people notice occur often in independent trials, so a real pattern is inferred from none. Another is that a genuine small effect exists and is masked in game data because defences tighten on a player who is scoring.
In its favour: People do over-read short streaks, which is well replicated, and the original paper reshaped how sports statistics are read.
Against it: Miller and Sanjurjo showed in 2018 that the original method contained a bias against finding streaks, and a small genuine hot hand survives correction, so the strong claim that the hot hand is purely illusory no longer holds.
Standing: Later work finds a small genuine hot-hand effect in some sports, so the original claim is qualified.
Further reading: Wikipedia

Page for this entry

People often give somewhat more to a single identifiable victim than to an equivalent statistical one, though the effect is modest and condition-dependent.

Schelling's 1968 essay contrasted the resources mobilised for a named trapped child against those for statistical lives saved by road safety, and Small, Loewenstein and Slovic supplied experimental evidence in 2007. The average effect is modest and condition-dependent, and charity fundraising has operated on it regardless.

In practice

An appeal for one named child raises more than the identical appeal for four thousand.
A safety case for a named worker's exposure moves a board that the aggregate statistic did not.

Attributed to: Thomas Schelling; tested by Small, Loewenstein & Slovic
Also called: Identified life effect
Standing: The researched mechanism behind the tragedy-and-statistic adage.
Further reading: Wikipedia

Page for this entry

People can value objects they successfully built above comparable objects they did not build.

Norton, Mochon and Ariely's studies with IKEA boxes, origami and Lego. Builders valued their own output near the price of expert work, and the effect disappeared when participants failed to complete the task or built and then destroyed the object.

In practice

A team that spent six weeks building an internal tool rates it above the commercial product it replaced and cannot say on what dimension.
A householder who tiled the bathroom themselves prices the house accordingly and no surveyor agrees.

Attributed to: Norton, Mochon & Ariely
Also called: Effort justification
Standing: The effect depends on successful completion; it disappeared where the build failed or was undone.
Further reading: Wikipedia

Page for this entry

People overestimate their influence over outcomes that are largely or wholly chance.

Langer's 1975 studies found people paid more for lottery tickets they had chosen themselves, and threw dice harder for high numbers. Both are chance, and both feel controllable.

In practice

A trader who picks their own positions is more confident than one given the same portfolio to manage.
A driver feels safer than a passenger on the identical journey, and the road does not know who is holding the wheel.

Attributed to: Ellen Langer
Further reading: Wikipedia

Page for this entry

Two salient but unrelated things are remembered as co-occurring, creating a relationship that is not there.

Chapman and Chapman's 1967 work on projective tests: clinicians reported diagnostic signs that were not in the data, because the pairing was memorable rather than frequent.

In practice

A manager remembers that the two worst incidents both happened after a particular engineer's deploys, and has forgotten the eleven that did not.
A parent is certain a food causes their child's bad nights, having noticed the coincidences and not the counter-instances.

Attributed to: Chapman & Chapman
Further reading: Wikipedia

Page for this entry

Specifying in advance when, where and how one will act closes much of the gap between intending something and doing it.

Gollwitzer's 1999 synthesis distinguished the goal intention, which states what one wants, from the implementation intention, which states the situational trigger and the response. Naming the cue delegates initiation to the environment rather than to willpower at the moment of choice. Effects across health, study and voting behaviour have been substantial relative to the cost of the intervention.

In practice

A team that agrees who raises a blocker and at which meeting raises blockers; one that agrees blockers should be raised does not.
A person who decides to exercise after work does less than one who decides to change at the desk at half past five.

Attributed to: Peter Gollwitzer
Also called: If-then plans

Page for this entry

People confidently report the reasons for their own judgements while having no access to the processes that produced them.

Nisbett and Wilson's 1977 paper, "Telling more than we can know." Subjects choosing among identical stockings showed a strong position effect and confidently reported reasons of quality and texture.

In practice

An interviewer explains they chose the candidate for their communication skills, having decided in the first minute for reasons they cannot access.
A shopper explains their preference for one brand in terms of quality, having been influenced by shelf position.

Attributed to: Nisbett & Wilson
Also called: Telling more than we can know
Further reading: Wikipedia

Page for this entry

Over budget, over time, under benefit, over and over.

Flyvbjerg's data across 258 projects in 20 countries: nine in ten exceed budget, with rail averaging 45% over. The consistency is what makes it a law rather than a complaint.

In practice

A tram extension is announced at £400m and eight years, and opens at £900m in year fourteen carrying two-thirds the forecast passengers.
A hospital IT programme costed at £2bn is abandoned at £10bn, and the business case was optimistic on all three dimensions simultaneously.

Attributed to: Bent Flyvbjerg

Page for this entry

Framing denial of an accusation as confirmation of it, so no response can exonerate.

Named for Kafka's Trial. Its structure is unfalsifiable by construction, which is what distinguishes it from a merely unfair accusation — no available response counts as evidence against.

In practice

An accusation of bias where denial is read as defensiveness and agreement as confession leaves no reply that counts as evidence.
A review process in which any objection to the process is logged as resistance cannot be objected to.

Attributed to: Named for Franz Kafka

Page for this entry

After n successes in n exchangeable trials, under a uniform prior the next is estimated to succeed with probability (n+1)/(n+2); never treat a finite run as certainty.

For exchangeable Bernoulli trials with a fixed unknown success probability and a uniform prior, n successes and no failures give a posterior predictive probability of (n+1)/(n+2). It is what makes "it has never failed" quantifiable: after 100 successes the estimated failure rate is about 1 in 102 rather than zero, provided those assumptions hold. Often mislabelled as a rule of Bayes.

In practice

A supplier with forty consecutive on-time deliveries is not certain to deliver on time, and the honest estimate is about one failure in forty-two.
A component with a four-year unblemished record has a quantifiable failure probability, and "it has never happened" is not the number.

Attributed to: Pierre-Simon Laplace
Standing: Often mislabelled as a rule of Bayes.
Further reading: Wikipedia

Page for this entry

Excessive data hinders decision-making rather than improves it.

Gross's 1964 term, later popularised by Toffler. The finding that matters: past a point, more information reduces decision accuracy while increasing decision confidence.

In practice

A board given forty pages of appendix and one page of recommendation reads the recommendation with more confidence and no more information.
A shopper given fourteen specifications for two phones chooses on brand, having spent an hour reading.

Attributed to: Bertram Gross; Alvin Toffler

Page for this entry

People expect small samples to resemble the parent population and over-read patterns in them.

Tversky and Kahneman's 1971 paper, aimed at researchers rather than laypeople: psychologists were over-reading small-sample results and expecting replication that the statistics did not support.

In practice

A team concludes a change worked from eleven users, and eleven users would show that pattern by chance about a fifth of the time.
A manager judges a hiring source from three hires, and three is not a sample.

Attributed to: Tversky & Kahneman

Page for this entry

If all you have is a hammer, everything looks like a nail.

Kaplan stated it in 1964 and Maslow in 1966. The organisational version is more costly: a firm with a consulting arm finds every problem needs consulting, and believes it.

In practice

A firm with a strong data science team frames a cultural problem as a measurement problem and builds a dashboard.
A litigator asked to resolve a family dispute produces a settlement agreement, which is what they have.

Attributed to: Abraham Kaplan; Abraham Maslow
Also called: Maslow's Hammer; golden hammer

Page for this entry

Treating real-world uncertainty as if it had the clean, enumerable probabilities of a game.

Taleb's term. Casino odds are known and enumerable; real risk is neither, so a risk model that presents itself as a game is smuggling in the thing that most needs questioning.

In practice

A risk register that assigns probabilities to five named scenarios has quietly asserted that the list is complete.
A trading model that performs beautifully in backtest is playing a game whose rules were written by the past.

Attributed to: Nassim Nicholas Taleb
Further reading: Wikipedia

Page for this entry

Making decisions based solely on quantitative observations, ignoring qualitative factors.

Yankelovich's 1971 critique of metric fixation, named for McNamara's use of body counts in Vietnam rather than authored by him. The sequence is instructive: measure what can be measured, disregard what cannot, presume what cannot be measured is unimportant, then that it does not exist.

In practice

A programme measured on units delivered reports success while the thing it was built to change is unchanged.
A school judged solely on examination results has no way to record the two years of behaviour work that made the results possible.

Attributed to: Daniel Yankelovich; named for Robert McNamara
Also called: Quantitative fallacy
Further reading: Wikipedia

Page for this entry

People choose a task carrying a deadline over a more valuable task without one, even when the urgent task's reward is smaller.

Zhu and colleagues found in 2018 that the mere presence of a time constraint pulled choice toward the urgent option regardless of payoff, and that the effect was strongest among people who believed themselves busy. Urgency is a visible property and importance is a judgement, so the cheaper cue wins.

In practice

A day spent entirely on things with deadlines has advanced nothing that matters and felt productive.
A team whose queue is sorted by due date works on whatever is nearest its date, not on what is worth most.

Attributed to: Meng Zhu, Yang Yang & Christopher Hsee
Also called: Urgency over importance
Standing: Related to the Eisenhower Matrix, which prescribes the corrective.

Page for this entry

Defending a modest claim under challenge while advancing an ambitious one when unchallenged.

Shackel's 2005 term from medieval fortification: the bailey is the desirable indefensible ground, the motte the defensible one nobody wants. The equivocation is between the two rather than within a word.

In practice

A claim defended as "we simply mean people should be considerate" was advanced as something far stronger before it was challenged.
A framework sold as a rigorous methodology retreats under scrutiny to "it's just about talking to each other".

Attributed to: Nicholas Shackel
Further reading: Wikipedia

Page for this entry

A coherent story imposed on past events creates false confidence in understanding their causes.

Taleb's term. The test is whether the story would have been just as satisfying had the outcome reversed. Most business and biographical explanation fails it.

In practice

A firm's decade of growth is explained by its culture, and the identical culture would have explained its collapse.
A retrospective on a successful launch names the reason; the same team's failed launch would have named the same factor with the sign reversed.

Attributed to: Nassim Nicholas Taleb

Page for this entry

Rejecting a workable institution by comparing it to an ideal rather than to the available alternatives.

Demsetz's 1969 term. Its symmetry is worth keeping: the market version compares real government to ideal markets, and Munger's unicorn fallacy is the same error pointed the other way.

In practice

A council's imperfect recycling scheme is attacked by comparison with an ideal one that no council operates.
A working vaccination programme is judged against zero risk rather than against the disease, which is the comparison that matters.

Attributed to: Harold Demsetz
Also called: Perfect solution fallacy
Further reading: Wikipedia

Page for this entry

Rescuing a generalisation by redefining the category to exclude every counterexample.

Flew's 1975 example. The tell is that the redefinition happens after the counterexample appears and is tailored to exclude it, which makes the generalisation unfalsifiable rather than true.

In practice

A claim that a methodology never fails survives every failed project as "they weren't really doing it".
A political movement disowns each embarrassing adherent as not a true representative, and the generalisation becomes unfalsifiable.

Attributed to: Antony Flew
Further reading: Wikipedia

Page for this entry

Harm caused by inaction is judged less culpable than equivalent harm caused by acting.

Ritov and Baron's studies on vaccination: parents rejected a vaccine with a small fatality risk over a disease with a larger one, because the vaccine death would be caused by their action.

In practice

A regulator who declines to approve a treatment is not blamed for the deaths that follow, and would be blamed for the deaths from approving it.
A manager who lets a project drift is judged less harshly than one who cancels it, and the drift costs more.

Attributed to: Ritov & Baron
Further reading: Wikipedia

Page for this entry

People judge themselves less likely than others to suffer negative events.

Weinstein's 1980 study: students rated themselves less likely than peers to divorce, develop cancer or be fired, and more likely to own a home. The bias is largest for events people believe they control.

In practice

A project manager rates their own project's chance of on-time delivery well above the base rate for projects like it.
A driver rates themselves above average and is, by definition, usually wrong about half the time.

Attributed to: Neil Weinstein
Also called: Unrealistic optimism
Further reading: Wikipedia

Page for this entry

People avoid information they expect to be unwelcome, monitoring investments and diagnoses less often when the news is likely bad.

Galai and Sade coined the term in a paper dated 2003 for an asset-pricing and liquidity anomaly. The specific claim that investors check portfolios less often in falling markets was tested by Karlsson, Seppi and Loewenstein, and later investor data have produced the opposite meerkat pattern.

In practice

An investor stops opening the monthly statement in a falling market and resumes when it recovers.
A manager stops running the report that shows the churn figure, and would say they had simply been busy.

Attributed to: Galai & Sade; tested by Karlsson, Seppi & Loewenstein
Also called: Information avoidance
Possible explanations: The explanation is anticipated regret: checking turns a suspected loss into a certain one, so avoidance protects the feeling rather than the portfolio.
In its favour: Information avoidance is well documented in other domains, including medical testing, where people decline results they expect to be unwelcome.
Against it: Later investor data show the opposite pattern, with attention rising in falling markets, so the effect may be specific to particular markets, periods or investor types.
Standing: Later investor data have produced the opposite meerkat pattern, with attention rising in falling markets.
Further reading: Wikipedia

Page for this entry

Reading genuine distress as evidence of deception, because the innocent under suspicion display the same signs as the guilty.

Ekman's term, from Othello reading Desdemona's fear as proof of guilt. It is the central problem for all demeanour-based credibility assessment, including in asylum and criminal proceedings.

In practice

An innocent employee in a disciplinary meeting shows exactly the distress an evasive one would, and the panel reads it as guilt.
An asylum applicant recounting real trauma inconsistently is scored as inconsistent, and trauma produces inconsistency.

Attributed to: Paul Ekman, from Shakespeare
Standing: A central problem for all demeanour-based credibility assessment.
Further reading: Wikipedia

Page for this entry

Decisions are judged by how they turned out rather than by what was knowable when they were made.

Baron and Hershey's 1988 studies: the same surgical decision was rated as better when the patient survived. It is why organisations that punish outcomes rather than process teach people to avoid documented risk-taking.

In practice

A driver who took a risk on a blind bend and arrived safely is called confident; the same manoeuvre ending in a scrape is called reckless.
An investment judged reckless after a loss and shrewd after a gain was the same decision made on the same information.

Attributed to: Baron & Hershey
Also called: Resulting
Further reading: Wikipedia

Page for this entry

Stated confidence is often more extreme than accuracy, particularly on difficult items.

Lichtenstein and Fischhoff's calibration studies illustrate it: people stating 90% confidence were right about 75% of the time, with the gap widening on harder questions. Calibration can reverse towards underconfidence on easy tasks, and improves greatly in expert domains with rapid feedback.

In practice

A project team asked for a date they are 90% sure of misses it more often than one time in ten.
A forecaster's 90% range should contain the outcome nine times in ten, and in practice contains it about seven, including among professionals.

Attributed to: Lichtenstein & Fischhoff
Further reading: Wikipedia

Page for this entry

Assuming in the premises what the argument is meant to establish.

Aristotle's petitio principii. "Freedom of speech is good because a society benefits when people can speak freely" is the form — the conclusion is doing service as its own premise.

In practice

"The policy is fair because it treats everyone equitably" has assumed the conclusion in the premise.
A business case that justifies a project by the strategic importance of the project has not made an argument.

Attributed to: Aristotle
Also called: Begging the question; circular reasoning

Page for this entry

People systematically underestimate how long tasks will take to complete.

Kahneman and Tversky's 1979 term. The Sydney Opera House was estimated at seven million dollars and four years; it cost 102 million and took fourteen. Reference class forecasting is the remedy.

In practice

A team late on its last six projects estimates the seventh from how long the work should take, rather than from how long the last six took.
A householder who allows a weekend for painting a room is not being foolish about paint; they are omitting the moving, masking and drying that the estimate never contained.

Attributed to: Kahneman & Tversky
Further reading: Wikipedia

Page for this entry

Treating sequence as causation; the earlier event is taken to have produced the later one.

Classical logic. The rooster crows and the sun rises. Most medical folk remedies rest on it, since most conditions resolve regardless of what was taken.

In practice

A cold clears four days after the remedy is taken, which is also how long a cold takes.
A team changes its process, the next quarter is better, and the market also turned — and the process change is credited.

Attributed to: Classical logic
Also called: False cause
Further reading: Wikipedia

Page for this entry

People will expend extra physical or cognitive effort to complete a task sooner, even when waiting would be easier.

Rosenbaum's subjects, asked to carry one of two buckets to a far end, repeatedly chose the nearer bucket and carried it further, against the obvious economy. The proposed explanation is that holding an unfinished subgoal in working memory is itself costly, so people pay in effort to discharge it. It is the opposite failure to procrastination and much less discussed.

In practice

An employee answers the trivial email immediately and defers the report, having optimised for closing items.
A team ships the easy sub-feature first, in an order nobody would have chosen deliberately.

Attributed to: David Rosenbaum and colleagues
Also called: Hastening subgoal completion
Further reading: Wikipedia

Page for this entry

Value falls steeply for the near future and shallowly thereafter, producing preferences that reverse as the moment arrives.

Ainslie's work. The reversal is the finding: asked in advance, people choose the larger later reward; asked at the moment, the same people choose the smaller sooner one.

In practice

A saver signs up to a scheme that raises contributions out of every future pay rise, and would not have accepted the same rise from today's salary.
Everyone signs up for the 7am gym slot in January and stops attending in February, and both decisions were sincere.

Attributed to: George Ainslie
Also called: Hyperbolic discounting

Page for this entry

People choose the path of least resistance, and this shapes language, search, and habit formation.

Zipf's 1949 principle. It explains word-frequency distributions, why people accept the first adequate search result, and why the default option in any system carries most of the traffic.

In practice

A library's most-borrowed shelf is the one by the door, and the catalogue is consulted by the few who did not find what they wanted there.
A form with an optional field has that field left blank by nearly everyone, whatever the guidance says.

Attributed to: George Kingsley Zipf

Page for this entry

A hypothesis raised for no evidential reason receives the attention due to a candidate, and the effort spent examining it is mistaken for grounds to consider it.

The illustration is a detective who names one resident of a city as a murder suspect with no evidence, and then investigates. Even honest investigation now spends its effort on a hypothesis selected at random from millions. The general lesson is that the expensive step is promotion to consideration, and it usually happens without anyone noticing it was a step.

In practice

A board that spends an hour on the one option somebody named has let the naming do the work of a shortlist.
A diagnosis considered because it was mentioned receives tests that the base rates do not justify.

Attributed to: Eliezer Yudkowsky
Also called: Undeserved promotion to attention
Standing: Related to anchoring and to the availability heuristic, and distinct from both: the fault is in what got onto the list, not in how it was weighed.

Page for this entry

Concern felt per victim falls as the number of victims rises, so response to mass suffering is weaker than to individual suffering.

Slovic's 2007 finding: donations for one named child exceeded donations for two, and fell further for eight. Concern per victim falls as victim count rises, so the aggregate response is worse than the individual one.

In practice

An appeal naming one family raises more than the same appeal naming forty, and the forty include the one.
A safety case built on one worker's exposure moves a board that the annual fatality statistic did not.

Attributed to: Paul Slovic
Also called: Compassion fade; collapse of compassion
Further reading: Wikipedia

Page for this entry

A concession is valued less as soon as it is known to come from an adversary.

Ross and Stillinger's finding: the identical arms-reduction proposal was rated worse when attributed to the Soviets than to the Americans. It is why third-party mediation and blind evaluation exist.

In practice

A proposal rejected in March is adopted in September, when a new director tables the same document under a different cover.
A concession offered by the union is read by management as a trap, and the same terms proposed by a mediator are accepted.

Attributed to: Ross & Stillinger
Further reading: Wikipedia

Page for this entry

Likelihood is judged by resemblance to a stereotype rather than by base rates.

Tversky and Kahneman's finding, and the Linda problem is the sharpest case. The heuristic is usually serviceable — resemblance often does track likelihood — and fails exactly where base rates are extreme.

In practice

Told a man is shy and tidy, people judge him more likely a librarian than a farmer, ignoring that farmers vastly outnumber librarians.
A recruiter shown a candidate who "looks like an engineer" weights the resemblance and not the base rate of who applies.

Attributed to: Tversky & Kahneman
Further reading: Wikipedia

Page for this entry

Stated valuation barely responds to order-of-magnitude changes in the quantity at stake.

Desvousges's study found willingness to pay to save 2,000, 20,000 and 200,000 birds was statistically indistinguishable. It is why contingent valuation is treated with suspicion in policy work.

In practice

A consultation finds residents will pay the same to clean one river as to clean four, and the four include the one.
A charity appeal raising the same sum whether it names ten beneficiaries or ten thousand is meeting the same insensitivity.

Attributed to: Desvousges et al.; Kahneman
Also called: Scope neglect; quantity insensitivity

Page for this entry

Groups spend their discussion on information everybody already holds and neglect information held by only one member, which is the information the group needed.

Stasser and Titus showed in 1985 that shared items are more likely to be mentioned, more likely to be repeated, and more likely to be treated as valid, simply because more people can introduce and confirm them. The result is that discussion tends to pool what was already common and leave the decisive fragment with the one person who had it.

In practice

A hiring panel spends forty minutes on the CV everyone read and two on the reference only one member saw.
A review meeting reaches confident agreement on the things that needed no meeting.

Attributed to: Garold Stasser & William Titus
Also called: Common knowledge effect
Standing: The hidden-profile paradigm is the experimental setup that demonstrates it.
Further reading: Wikipedia

Page for this entry

Belief that a single solution can solve complex problems.

Named from Brooks's 1986 essay No Silver Bullet, which argued no single development improvement would yield an order-of-magnitude gain. Nearly forty years of tooling have largely borne that out.

In practice

A team that adopts a new framework every year is as fast as it was in 2015, and each adoption was justified on its own.
An organisation that buys a new project tool every two years has made forty small improvements and no order-of-magnitude one.

Attributed to: After Brooks's essay No Silver Bullet
Also called: No silver bullet

Page for this entry

Asserting that a first step must lead to an extreme end without establishing the mechanism.

Not fallacious where the causal chain is actually demonstrated. The test is whether a mechanism is supplied linking each step, or whether the argument relies on the emotional weight of the endpoint.

In practice

A proposal to record meeting attendance is opposed as the first step towards surveillance, and nobody names the step in between.
A lawyer arguing that a precedent will be extended sets out the three cases in which courts have already extended it, which is the same shape of argument and not a fallacy.

Attributed to: Classical rhetoric
Standing: Not fallacious where the causal chain is actually demonstrated.
Further reading: Wikipedia

Page for this entry

Invoking an exemption for one's own case without a principled ground for the exception.

The test is whether the exemption would be granted to anyone else in the same position. If the principle applies only to the person invoking it, it is not a principle.

In practice

A manager who missed the deadline explains the circumstances, having dismissed the identical explanation from a supplier in March.
A department that argues its work is too unusual for the standard reporting has made a claim every department could make.

Attributed to: Classical rhetoric
Further reading: Wikipedia

Page for this entry

Refuting a weakened version of the opponent's position rather than the position itself.

The inverse of the principle of charity. The reliable tell is that the version being refuted is one no advocate has stated, and that quoting an actual advocate would weaken the refutation.

In practice

A consultation response answers the objection that the scheme would abolish parking, which no objector made, and never reaches the one about the delivery bay.
A rebuttal quotes an anonymous commenter rather than the report it claims to refute, because the report is more careful.

Attributed to: Classical rhetoric
Standing: The inverse of the principle of charity.
Further reading: Wikipedia

Page for this entry

Inquiry concentrates where measurement is easy rather than where the answer is likely to be.

Named from the drunkard searching for keys under the lamppost. It is why so much social science measures what surveys capture, and why administrative data reshapes fields when it becomes available.

In practice

A firm optimises the checkout because it has checkout data, and never measures the customers who left before the catalogue finished loading.
A ministry evaluates the one programme that had a survey attached, and the three that did not are written up as lacking evidence.

Attributed to: David Freedman
Also called: Drunkard's search
Further reading: Wikipedia

Page for this entry

Continuing investment based on past costs rather than future value.

Arkes and Blumer's 1985 theatre-season study is the empirical case. The label Concorde fallacy invokes that project as a symbol of escalation after large prior investment, though archival histories show its continuation also reflected electoral politics, industrial policy, Anglo-French commitments, cancellation costs and strategic concerns.

In practice

Also called the Concorde fallacy, though that project's continuation had several causes besides what had already been spent.
A household that has spent £30,000 on a renovation approves the next £10,000 because of the £30,000, which is gone either way.

Attributed to: Arkes & Blumer
Also called: Concorde fallacy
Further reading: Wikipedia

Page for this entry

Alternating between tasks is slower and more error-prone than completing them in blocks, and part of the cost survives any amount of preparation.

Rogers and Monsell showed in 1995 that switch costs shrink with warning time but do not vanish, which points to a residual component incurred at the moment of the switch itself rather than to incomplete preparation. Jersild had measured the effect in 1927. The cost is per switch, so it scales with fragmentation rather than with the number of tasks.

In practice

A developer interrupted four times an hour loses more than four interruptions' worth of time.
A day of thirty-minute meetings on six subjects produces less than a day of three two-hour blocks.

Attributed to: Arthur Jersild; revived by Robert Rogers & Stephen Monsell
Also called: Switch cost; the cost of interleaving

Page for this entry

A pattern defined after the data are seen will always be found, and proves nothing.

Named from the marksman firing at a barn and painting the target around the holes. It is why pre-registration matters — a hypothesis specified after the data is not a hypothesis at all.

In practice

An analyst runs eleven segment breakdowns, finds one significant result, and writes the report as though that segment had been the question.
A hypothesis specified after the data is not a hypothesis, which is why pre-registration changes the false-positive rate rather than the honesty of researchers.

Attributed to: Epidemiology folklore
Further reading: Wikipedia

Page for this entry

Judging a class by only its detectable members: all toupees look terrible, because the convincing ones are not counted as toupees.

The sample available to an observer is the failures, because success in this class consists of not being noticed. The error is general and shows up wherever competence is invisible by definition: good editing, good security, good administration and good acting are all judged from the instances that failed.

In practice

An IT department is judged on the outages, since the years without one produce no evidence of anything.
A reader who concludes that translations are clumsy has read the clumsy ones and not noticed the rest.

Attributed to: Unattributed
Also called: Selection on the detectable
Standing: A folk statement of selection bias on an unobserved population. Compare survivorship bias, which selects on outcome rather than on detectability.

Page for this entry

Deflecting a charge by alleging the accuser is guilty of the same thing.

The deflection is usually true and always irrelevant. Its legitimate use is narrow: it bears on the accuser's credibility as a witness, never on whether the charge is accurate.

In practice

A minister challenged on missed targets replies that the previous government missed its own, and the targets are still missed.
A witness whose own expenses were irregular is cross-examined on that record, which bears on whether to believe him and not on whether the defendant claimed.

Attributed to: Classical rhetoric
Also called: Appeal to hypocrisy; whataboutism
Further reading: Wikipedia

Page for this entry

Aggregate judgments of groups can exceed individual expert judgments.

Galton's 1907 ox-weight competition: 787 guesses averaged 1,197 pounds against an actual 1,198. The conditions are strict — independence, diversity, and an aggregation mechanism — and correlated errors destroy it.

In practice

A team asked to estimate a project independently and then averaged lands closer than any member; the same team asked in a meeting converges on whoever spoke first.
A village fete sweepstake on the weight of a cake collects sixty guesses, and the mean beats the baker's.

Attributed to: Francis Galton; popularised by James Surowiecki
Further reading: Wikipedia

Page for this entry

Model

Explanatory accounts and formal models.

Essay: Why reasoning works the way it does

People switch decision strategies according to the effort and accuracy a task demands, so there is no single process to model.

Choosing lunch, people satisfice on the first acceptable option; choosing a mortgage, they build a comparison table. Payne's finding is that the switch is sensible, not lazy — the strategy is fitted to the stakes.

In practice

Buying coffee, you take the first acceptable option; buying a house, you build a spreadsheet with fourteen columns and still visit twice.
A hiring panel screens 200 CVs on two criteria, then spends four hours on the final three candidates.

Attributed to: Payne, Bettman & Johnson
Also called: Contingent processing

Page for this entry

Success requires every one of several conditions to hold, while failure requires only one to be absent, so successes resemble each other and failures do not.

Tolstoy opened Anna Karenina with the observation that happy families are all alike and each unhappy family is unhappy in its own way. Diamond generalised it in 1997 to explain why so few large mammals were domesticated: a candidate species had to satisfy diet, growth rate, breeding, temperament, panic response and social structure, and failing any one was sufficient. The structure is conjunctive, which is why the survivors look uniform.

In practice

A functioning project looks like every other functioning project, and each failed one has its own story.
A checklist where every item must pass will be failed by a different item each time.

Attributed to: Leo Tolstoy; generalised by Jared Diamond
Also called: Many ways to fail, one way to succeed
Further reading: Wikipedia

Page for this entry

Most of what feels like thinking is retrieval of a conclusion reached earlier or by somebody else, and it arrives with the same sensation as reasoning.

The point is phenomenological: a retrieved conclusion and a derived one feel identical from the inside, so there is no internal signal warning that a thought was taken off the shelf. Yudkowsky's remedy is to notice when an opinion arrives too fast for the question's difficulty, which is the only available tell.

In practice

A view held confidently and never examined was acquired at nineteen from someone whose name has gone.
A team's instant agreement on an unfamiliar question is evidence that the question was not unfamiliar to the cache.

Attributed to: Eliezer Yudkowsky
Also called: Thoughts retrieved rather than computed
Standing: Consistent with research on fluency and on the illusion of explanatory depth; the specific cache metaphor is the coinage.

Page for this entry

Sorting people by whether an action benefits them and benefits others yields four types, and the one who harms others while gaining nothing is the most dangerous because he is unpredictable.

Cipolla, an Italian economic historian at Berkeley, circulated the essay privately in 1976. The matrix crosses gain to self against gain to others and gives the intelligent, the bandit, the helpless and the stupid. His definition is the contribution: a stupid person causes losses to others while deriving no gain and often incurring losses himself, which makes him harder to model than a bandit, whose incentives can at least be predicted. He noted the M in his own name was an invention.

In practice

A colleague pursuing his own advantage can be negotiated with; one damaging the project and himself at once cannot.
A committee that assumes bad actors are self-interested is unprepared for the member with no stake in the outcome.

Attributed to: Carlo M. Cipolla
Also called: The basic laws of human stupidity; the stupidity matrix
Possible explanations: The useful part is the matrix: crossing benefit to oneself with benefit to others separates the merely selfish, who gain at others' expense, from the destructive, who lose while inflicting loss, which ordinary usage runs together.
In its favour: The distinction is genuinely clarifying, and it names a category, harm without gain, that most accounts of behaviour leave out.
Against it: It was written as satire and offered as laws, no measurement supports the claims about prevalence, and the categories are assigned after the outcome, which makes them descriptive rather than predictive.
Standing: Written as satire and offered as laws; no measurement supports the claims about prevalence. The matrix is the durable part, because it separates malice from stupidity on a criterion that can actually be applied.

Page for this entry

Categories form at the level that maximises information gained per unit of cognitive effort.

Rosch's finding: people say "chair" rather than "furniture" or "Windsor chair" because the middle level carries the most information for the least effort. Basic-level categories are where naming settles across languages.

In practice

A new starter is told "use the standard template", not "the standard corporate quarterly business review template, version 4".
A doctor says "take it with food" rather than naming a meal, because the middle level is what the patient will actually remember.

Attributed to: Eleanor Rosch

Page for this entry

Reasoning runs on a fast automatic system and a slow deliberate one, and most documented biases are the first operating unchecked.

Popularised by Kahneman in 2011. The strict two-system division is now contested in favour of a continuum of automaticity, but the distinction survives as a framework for why deliberate checking works at all.

In practice

A driver brakes before consciously seeing the child, and then spends the evening reasoning about whether the road was safe.
An interviewer forms a view in ninety seconds and produces a page of justification afterwards, sincerely.

Attributed to: Stanovich & West
Also called: System 1 and System 2
Possible explanations: The explanation is capacity: fast associative processing is cheap and always running, and deliberate checking is expensive, so it is deployed selectively.
In its favour: It predicts what is observed: time pressure and cognitive load increase biased answers, and prompting deliberate checking reduces some of them.
Against it: The strict division into two systems is contested in favour of a continuum of automaticity, the labels describe rather than explain, and several biases persist under careful deliberation.
Standing: Popularised by Kahneman. The strict two-system division is now contested in favour of a continuum of automaticity.

Page for this entry

Simple rules that deliberately ignore information can beat complex models where uncertainty is high and data are thin.

Gigerenzer's recognition heuristic: asked which German city is larger, American students outperformed German students by recognising fewer names. The ignorance was doing the work.

In practice

A triage nurse using three questions sorts patients as accurately as a scoring model with fourteen inputs.
A recruiter using one strong signal and ignoring the rest matches the panel's outcome and takes a tenth of the time.

Attributed to: Gerd Gigerenzer
Also called: Ecological rationality; less-is-more effect
Standing: The principal rival to the heuristics-and-biases programme, treating heuristics as adaptive rather than defective.

Page for this entry

Decisions must be taken on information that is incomplete, late, and partly wrong, and this is the normal condition rather than a failure.

Clausewitz's term. His conclusion is the operationally useful part: since the fog cannot be lifted, command must be organised to function within it rather than to wait for clarity.

In practice

An incident commander acts on a picture that is twenty minutes old and partly wrong, and would still be acting on one if they waited.
A board takes a decision on a market that has already moved, because the reporting cycle is monthly and the market is not.

Attributed to: Carl von Clausewitz
Further reading: Wikipedia

Page for this entry

Ignorance of difficulties can paradoxically enable ambitious projects to start.

Hirschman's 1967 finding from development projects: had the difficulties been known, nothing would have been started, and the difficulties were solved once people were committed. He warned it is not a policy.

In practice

A team that fully understood the integration would never have started, and having started, solved it.
A first-time renovator who had seen the survey properly would not have bought, and the house is now finished.

Attributed to: Albert O. Hirschman

Page for this entry

Judgments are made by pairwise comparison; perceived magnitude is recovered from relative rather than absolute assessments.

Thurstone's 1927 method underlies modern rating scales and paired-comparison testing. Its practical value is that people compare far more reliably than they rate on absolute scales.

In practice

A wine panel asked to rank eleven bottles disagrees; asked to compare pairs, the same panel produces a stable ordering.
A recruitment shortlist scored absolutely is inconsistent between panellists, and the same panellists agree on head-to-head comparisons.

Attributed to: Louis Thurstone

Page for this entry

Decision-makers satisfice under limits of attention and information.

Simon developed the underlying ideas in Administrative Behavior in 1947; the term itself first appeared in print in Models of Man in 1957, and March and Simon applied related ideas to organisations in Organizations in 1958. His point is that satisficing is rational given the cost of search rather than a departure from rationality. He received the 1978 Nobel for it.

In practice

A family choosing a school visits three, likes the second, and stops, having spent exactly as long on it as the decision was worth.
A buyer who stops at the first acceptable supplier has correctly priced their own time, and a buyer who evaluates forty has not.

Attributed to: Herbert Simon; March & Simon

Page for this entry

Reasoning is directed towards a preferred conclusion while remaining subjectively rigorous.

Kunda's 1990 review. The mechanism is that people search harder for flaws in unwelcome evidence and stop searching when welcome evidence appears — the standards are real, and applied asymmetrically.

In practice

An analyst who wants a project approved scrutinises the risk assumptions hard and the benefit assumptions lightly, and applies real rigour to both halves.
A board reviewing a decision it already made asks tougher questions of the dissent than of the paper.

Attributed to: Ziva Kunda
Further reading: Wikipedia

Page for this entry

People assume they see reality objectively and that dissenters must be uninformed or biased.

Ross's term. The three-step consequence is the interesting part: the dissenter is first assumed uninformed, then assumed irrational, then assumed malicious — in that order.

In practice

A manager concludes the dissenting team member has not understood the strategy, then that they are being difficult, then that they are undermining it.
Two departments each believe the other is briefing against them, and both are reading the same neutral email.

Attributed to: Lee Ross
Further reading: Wikipedia

Page for this entry

Distance in time, space or probability shifts thinking toward the abstract and idealistic, while nearness shifts it toward the concrete and feasible.

Distant matters are represented by their central features and desirability, near ones by their details and feasibility, which is why people hold high principles about the distant future and act on convenience in the present without experiencing a contradiction. The two modes are not in conflict because they are rarely active at once.

In practice

A board that adopts an ambitious five-year commitment declines the first concrete step toward it in the same meeting.
A person certain about what they would do in a crisis is reasoning in a mode the crisis will not permit.

Attributed to: Robin Hanson, from construal level theory
Also called: Construal level; abstract and concrete modes
Standing: The underlying construal level theory is an established research programme; the blog formulation is a popularisation that draws stronger conclusions about hypocrisy than the source supports.

Page for this entry

Small changes to how options are presented can shift behaviour without restricting choice.

Thaler and Sunstein's 2008 framework. Average effects are heterogeneous across intervention types and domains, and meta-analytic estimates are affected by publication bias. The other contested part is the libertarian half: whether an architecture designed to produce a particular choice really leaves the choice free.

In practice

A canteen that puts fruit at eye level and chips at the end sells more fruit without removing anything.
A form that pre-selects the monthly donation raises more than one that leaves the field blank, and both leave the choice open.

Attributed to: Thaler & Sunstein
Also called: Choice architecture; libertarian paternalism
Possible explanations: The explanation is that choices are always presented somehow, and features that should be irrelevant, such as order, framing and the default, reliably shift what people pick.
In its favour: Some interventions work well and cheaply, particularly defaults and simplified forms, and the framework changed how governments design services.
Against it: Average effects are heterogeneous, meta-analytic estimates are affected by publication bias, and several headline results have not replicated. Critics also question the ethics of designing choices people do not notice.
Standing: Average effects are heterogeneous across intervention types and domains, and meta-analytic estimates are affected by publication bias.

Page for this entry

Advantage goes to whoever cycles through observe, orient, decide, and act faster than the opponent can reorient.

Boyd developed it from air combat, and orientation is the step that carries the weight: it is where prior experience and framing enter, and where an opponent can be induced to misread the situation.

In practice

An incident team that reframes the problem faster than the outage evolves regains control; one still working from the first hypothesis does not.
A competitor who reads a market shift and repositions before you have finished the analysis has beaten you with the same information.

Attributed to: John Boyd
Also called: Boyd cycle
Further reading: Wikipedia

Page for this entry

A forced choice between acting to cause a lesser harm and refraining while a greater harm occurs.

Foot's 1967 formulation, extended by Thomson. Widely used and widely criticised as a guide to actual moral decisions, since real choices rarely present certain outcomes and no third option.

In practice

A committee allocating one ventilator between two patients is told the trolley case clarifies the principle, and finds neither outcome certain and a third bed possibly freeing up.
A regulator writing an automated braking rule must settle in advance what the vehicle does, which is the rare case where the certainty the problem assumes is real.

Attributed to: Philippa Foot
Possible explanations: One explanation for the pattern of answers is the distinction between doing and allowing. Another is the doctrine of double effect, under which harm as a side effect is judged differently from harm as a means. A third is proximity, since pushing a person is more aversive than pulling a lever.
In its favour: It isolates moral intuitions under controlled conditions, and the broad pattern of responses recurs across many countries, which is why it remains a standard tool in moral psychology.
Against it: Critics argue that sacrificial dilemmas measure willingness to accept an artificial premise rather than moral judgement, and that responses correlate with traits such as callousness. Real choices rarely offer certain outcomes or only two options.
Standing: Widely used and widely criticised as a guide to actual moral decisions.
Further reading: Wikipedia

Page for this entry

Series of small decisions can cumulatively lead to unwanted outcomes.

Kahn's 1966 term. Nobody chose to eliminate passenger rail or corner shops; each individual decision was small and reasonable, and the aggregate outcome was chosen by no one.

In practice

A high street loses its butcher, then its baker, then its greengrocer, and every household that drove to the supermarket did something reasonable.
A team that takes one urgent request a week has, within a year, no capacity for anything planned, and no meeting contained that decision.

Attributed to: Alfred E. Kahn
Further reading: Wikipedia

Page for this entry

Acquiring some explicit reasoning skill can make a person worse than they were, because it displaces working intuitions before it replaces them.

The claim is that the intuitions being overridden were doing real work, and the explicit substitute is initially worse at the job. It is offered as a warning against the confidence that arrives with a new framework, and it is unusual in being a self-criticism from a community devoted to the skill in question.

In practice

A manager who has read one book on decision-making makes worse decisions for six months than before reading it.
A driver taught to think about the steering drives worse until the thinking stops.

Attributed to: LessWrong community
Also called: Partial rationality as a hazard
Possible explanations: The proposed explanation is displacement: a new explicit method overrides intuitions that were doing real work, and the substitute is initially worse at the job, so performance falls before it recovers.
In its favour: The shape is familiar from skill acquisition, where explicit instruction temporarily disrupts a fluent performance, and the warning against confidence arriving with a new framework is sound.
Against it: It is argued from anecdote within one community and not measured, and it is unfalsifiable in practice, since any failure can be attributed to being in the valley and any success to having crossed it.
Standing: A coinage of the rationalist community, argued from examples rather than tested. Argued from anecdote within the community and not measured. The related finding that instruction in a heuristic can impair performance before it improves it does appear in the expertise literature.

Page for this entry

Precept

Statements of how one ought to act or design.

Essay: Razors, rules and the tools of careful thought

Inferring from the absence of a record that the thing did not occur, which holds only where a record would be expected.

No mention of a common practice in surviving records is weak evidence when the records are administrative rather than descriptive. Sherlock Holmes's dog that did not bark is the valid form: a guard dog's silence is evidence because barking was expected.

In practice

No minutes record the decision, but the meeting had no secretary and never produced minutes for anything.
A candidate's CV omits a two-year gap in a format that lists only years, so the omission carries no information at all.

Attributed to: Historiography
Also called: Argumentum ex silentio; the dog that didn't bark
Further reading: Wikipedia

Page for this entry

Revise belief in proportion to how much more likely the evidence is under one hypothesis than another.

The practical form: ask how surprising the evidence would be if you were wrong. Evidence that both hypotheses predict equally well should move nothing, however dramatic it feels.

In practice

A test comes back positive; the useful question is how often it comes back positive for people who do not have the condition.
An outage coincides with a deployment, but deployments happen daily and outages are rare, so the coincidence is weaker evidence than it feels.

Attributed to: Thomas Bayes
Also called: Bayes's theorem

Page for this entry

Act only on a rule you could will everyone to follow, and treat people as ends rather than merely as means.

Kant's 1785 test. The second formulation — treating people as ends — does more practical work than the universalisability test, which most maxims can be written to pass.

In practice

A manager considering a small deception asks whether a workplace where everyone did this could function, and finds it could not.
A landlord who treats a tenant purely as a payment stream, and not as a person with a claim on being told about the boiler, fails the second formulation without breaking any rule.

Attributed to: Immanuel Kant
Also called: Universalisability test
Further reading: Wikipedia

Page for this entry

Don't remove a fence until you understand why it was put there.

Chesterton's 1929 formulation: the reformer who cannot say why the fence is there is not yet qualified to remove it. It is an argument about the burden of inquiry, not a prohibition on removal.

In practice

A new head of operations removes the two-signature rule on refunds and discovers in month four why it was introduced in the first place.
A team deletes a scheduled job nobody could explain, and the month-end reconciliation fails for the first time in six years.

Attributed to: G. K. Chesterton

Page for this entry

Effort spent on what one can affect expands that range, and effort spent on what one merely worries about shrinks it.

Covey's distinction between a circle of concern and a smaller circle of influence inside it is a sorting device rather than a discovery. Its value is diagnostic: attention is finite, most people's concern extends far past their influence, and the exercise of locating the boundary is the intervention.

In practice

A team that stops discussing the market and starts on its own onboarding changes something within a quarter.
A meeting spent on decisions taken elsewhere ends with nothing anyone present can do.

Attributed to: Stephen Covey
Also called: Concern versus influence

Page for this entry

Avoid assigning prior probability 0 or 1 unless logically certain.

Lindley named it for Cromwell's 1650 plea to consider that you may be mistaken. Formally, a prior of zero cannot be updated by any evidence whatever — certainty is unrevisable by construction.

In practice

An investigator who has ruled out one suspect entirely cannot revise, whatever the new evidence shows.
A forecaster who says a project has "zero chance" of slipping has left themselves nowhere to go when the first milestone moves.

Attributed to: Dennis Lindley, after Oliver Cromwell
Further reading: Wikipedia

Page for this entry

When a cause is unclear, ask who benefits; motive narrows the field of explanation.

Cicero used it as a forensic principle. It narrows a field of explanation and proves nothing on its own; benefit is evidence of motive, not of agency.

In practice

A department reorganisation that leaves one director with three more teams is not proof of anything, and it is the first question to ask.
A sudden change in a supplier's recommended specification is worth checking against who supplies the newly recommended part.

Attributed to: Cicero
Standing: A heuristic, not a proof; benefit is evidence of motive and not of agency.
Further reading: Wikipedia

Page for this entry

Locate the fact on which both parties would change their minds, and argue about that rather than about the conclusion.

The procedure asks each side what would change their view, and looks for an item that appears on both lists. Where one exists, the disagreement has a testable centre and the discussion can be about it. Where none exists, the parties have learned something more useful than any further exchange would have produced: they are not disagreeing about the same thing.

In practice

Two managers arguing about a hire find the crux is whether the role is permanent, which neither had raised.
A pair who can name nothing that would change their minds have discovered the shape of the disagreement.

Attributed to: Center for Applied Rationality
Also called: Finding the shared crux

Page for this entry

Where direct verification is unavailable, identify a thing by whether it exhibits the characteristics of the category.

The test is what most practical identification actually uses, since the defining property is usually unobservable and the correlates are not. It is sound in proportion to how costly mimicry is. Where imitating the characteristics is cheap, the test selects for imitators, which is why it works for birds and fails for confidence tricks.

In practice

A regulator who identifies a bank by what it does rather than by its licence has applied the test correctly.
A hiring process that screens for the visible markers of competence selects for people who acquired the markers.

Attributed to: Attributed to James Whitcomb Riley and others
Also called: If it walks like a duck
Standing: A statement of abductive inference. Its weakness is the same as abduction's: mimicry defeats it, and the cases where identification matters most are the cases someone had reason to disguise.
Further reading: Wikipedia

Page for this entry

Sort tasks by urgency against importance; the trap is that urgency captures attention that importance deserves.

Attributed to Eisenhower via a 1954 speech quoting an unnamed college president. The urgent-unimportant quadrant is where most organisational time goes and where almost none of the value is.

In practice

A department spends every morning on inbound requests and never on the process redesign that would end most of them.
A householder deals with post, calls and errands all week and never books the boiler service that would prevent the emergency.

Attributed to: Attributed to Dwight D. Eisenhower
Also called: Urgent-important matrix

Page for this entry

Weigh actions by summing the intensity, duration, certainty, and extent of the pleasure and pain they produce.

Bentham's 1789 attempt to make morality calculable. Its failure is instructive — the dimensions cannot be commensurated in practice — and cost-benefit analysis inherits both the ambition and the problem.

In practice

A regulator scores a policy on lives saved, years of life, and distress avoided, and cannot state the exchange rate between them.
A hospital allocating one theatre between a painful condition and a shortening one is doing this arithmetic and cannot show its working.

Attributed to: Jeremy Bentham
Also called: Hedonic calculus
Further reading: Wikipedia

Page for this entry

Decompose an unknown quantity into factors you can estimate or bound, and combine them into a useful order-of-magnitude answer.

Fermi's students estimated the number of piano tuners in Chicago from population, households, pianos per household and tunings per year. Combining several rough estimates can land within an order of magnitude, particularly where the errors are not all biased in the same direction.

In practice

A team asked how many support tickets a new market will generate estimates from population, adoption rate and contact rate, and lands within a factor of two.
A planner sanity-checks a £40m forecast by asking how many customers that implies per day, and finds it exceeds the town's population.

Attributed to: Enrico Fermi
Also called: Back-of-the-envelope estimation; order-of-magnitude reasoning

Page for this entry

Make haste slowly; speed achieved by skipping preparation is repaid with interest.

Augustus's motto, adopted by the Medici with a tortoise and sail. The operational form: the time spent on preparation is not lost against the schedule, it is the only thing that protects it.

In practice

A team that spends the first fortnight on tooling delivers the six-month project faster than one that started coding on day two.
A cook who preps everything before turning on the heat serves faster than one who chops between stirs.

Attributed to: Augustan motto; adopted by the Medici
Further reading: Wikipedia

Page for this entry

Where two problems resist separate solution, consider that they may be one problem under two names; establishing the identity solves both at once.

A detective works two apparently unrelated cases, a missing animal and a missing sportsman, and the investigation collapses when the officer leading it and the suspect being sought prove to be one person. The formal content is the identification step: two problem statements are shown to share a solution set, so any solution to one is a solution to the other. Mathematicians call this a reduction and spend careers hunting for one. The detective arrives at it by other means.

In practice

A team chasing a slow report and an unrelated billing discrepancy finds one clock set wrong, and closes two tickets with one commit.
A council with a parking problem and a school-run problem has one problem, and had been funding two teams to not solve it.

Attributed to: Ace Ventura; formalised for this ledger
Also called: Einhorn is Finkle; problem unification; the two-case collapse
Standing: Entered in jest. The attribution is a 1994 comedy and the name is this ledger's. The move underneath is real and carries serious names elsewhere: reduction in computational complexity, unification in physics, and root-cause analysis in engineering.

Page for this entry

Virtue lies between deficiency and excess, and the right point differs by person and circumstance.

Aristotle's doctrine, and the frequently missed part: the mean is relative to the person and situation, not a midpoint on a scale. Courage for a soldier and for a civilian sit at different points.

In practice

The mean is relative: courage in a firefighter and in an accountant sit at different points, and neither is a midpoint on a scale.
A manager's right level of oversight for a new starter is intrusive for a veteran, and the same figure is both deficiency and excess.

Attributed to: Aristotle
Also called: Doctrine of the mean

Page for this entry

Never attribute to malice what can be adequately explained by stupidity or error.

Hanlon's razor works because incompetence is more common than malice, and because assuming malice provokes it. It fails where the actor has a demonstrated pattern, at which point the razor is a way of not looking.

In practice

A payment that never arrived is far more often a wrong sort code than a decision to withhold.
A colleague who ignored your message probably has 340 unread ones, and treating it as a snub creates the conflict it assumed.

Attributed to: Robert J. Hanlon
Further reading: Wikipedia

Page for this entry

Never attribute to malice that which is adequately explained by stupidity—but don't rule out malice.

The addition matters. Hanlon's razor can become a way of never looking at motive, and Heinlein's version keeps the possibility open without making it the default.

In practice

A supplier whose invoices are always wrong may be incompetent, and the pattern of errors always favouring them is worth noticing.
A colleague who repeatedly forgets to copy you in may be disorganised, and the meetings they forget are consistently the ones where your view would be inconvenient.

Attributed to: After Robert A. Heinlein

Page for this entry

What can be asserted without evidence can be dismissed without evidence.

From Hitchens's 2007 book God Is Not Great, restating the burden of proof as a conversational rule. The underlying principle is considerably older, and the label was attached around the same period. It settles who must move first and nothing about who is right.

In practice

A claim that a product improves outcomes, offered with no trial, needs no trial to set aside.
A meeting assertion that "customers hate the new flow", offered without data, can be answered by asking for the data rather than by disproving it.

Attributed to: Christopher Hitchens
Also called: Onus probandi
Further reading: Wikipedia

Page for this entry

You can't derive an 'ought' from an 'is'.

Hume's 1739 observation that writers slide from statements about what is to statements about what ought to be without noticing. No factual finding alone settles a policy question, which is why "the science says" is always incomplete.

In practice

A study showing a policy reduces harm does not establish that the policy should be adopted, because that requires a view about what else is worth giving up.
A finding that a group performs differently on a measure carries no conclusion about what anyone is owed, and treating it as one skips the step.

Attributed to: David Hume
Also called: Is-ought problem

Page for this entry

You understand a position only if you can state it well enough to pass as a sincere adherent.

Caplan's 2011 proposal. It is a stronger standard than steelmanning because it is falsifiable: hand your statement to actual adherents and see whether they can tell.

In practice

A policy analyst who can write the opposing brief well enough that its advocates would sign it has understood the objection.
A manager who can state a resigning employee's reasons in terms the employee would accept has learned something the exit interview did not produce.

Attributed to: Bryan Caplan

Page for this entry

Start from the desired end state, or from how a plan would fail, and reason back to the actions required.

Attributed to Jacobi and popularised by Munger: invert, always invert. The premortem is its practical form, and it works because failure modes are easier to enumerate than success paths.

In practice

A team designs a launch by asking what would have to be true for it to fail, then removing those conditions.
A retirement plan built by starting from the desired income and working back is more tractable than one built forward from savings.

Attributed to: Carl Jacobi; popularised by Charlie Munger
Also called: Invert, always invert

Page for this entry

Discussion becomes impossible on topics people have made part of their identity, so holding fewer beliefs as identity preserves the ability to change your mind.

Graham's observation is that the subjects on which productive argument is impossible are not the difficult ones but the identity-laden ones, and that the labelling is what does it. Once a position is who you are, evidence against it is an attack rather than information. The prescription is unusual in being about what to call yourself rather than what to believe.

In practice

An engineer who identifies as a partisan of one language cannot evaluate its successor.
A team that has named itself after its methodology will not be the team that replaces it.

Attributed to: Paul Graham
Also called: Identity and unwinnable arguments
Standing: Consistent with research on motivated reasoning and identity-protective cognition, which finds reasoning ability recruited to defend group-relevant conclusions.

Page for this entry

Reliable success requires accurate assessment of both sides; most failure comes from misjudging one's own capacity.

Sun Tzu's fuller statement: know both and you need not fear a hundred battles; know only yourself and you will win and lose in turn. Misjudging one's own capacity is the more common failure.

In practice

A negotiator who has read everything about the other side and never established what his own board will accept concedes a term he cannot deliver.
A firm that has studied its competitor exhaustively and never audited its own delivery capacity loses the contract it won.

Attributed to: Sun Tzu

Page for this entry

Answer the version of an argument in which your easiest objection is unavailable, since dodging into a convenient world leaves the argument untouched.

The move being blocked is escaping a dilemma by disputing its setup. Told that a machine could prove your beliefs false, the convenient response is that no such machine exists. Alexander's rule is to grant the setup and answer what remains, because the objection about the machine was never the reason you disagreed.

In practice

A risk scenario dismissed as unrealistic has been dismissed on the one feature that was not the point.
A trainee who answers every case study by querying the premise has learned to avoid the exercise.

Attributed to: Scott Alexander
Also called: LCPW; steelmanning the hypothetical
Standing: The complement to the principle of charity: charity strengthens the opponent's case, this strengthens the hypothetical it rests on.

Page for this entry

Prefer the simplest adequate explanation over more complex ones.

Ockham never wrote the phrase attributed to him. The razor is a tie-breaker between explanations of equal adequacy, not a preference for simple explanations over adequate ones.

In practice

Two accounts of the outage fit every log entry; the one requiring a single failure is preferred to the one requiring three coincidences.
The razor breaks ties between explanations of equal adequacy — it does not prefer a simple account that fails to explain the data over a complex one that succeeds.

Attributed to: William of Ockham
Also called: Law of parsimony; lex parsimoniae
Further reading: Wikipedia

Page for this entry

Listen to the person in front of you; they are usually telling you the answer.

Osler's teaching from around 1904. It survives because it is empirically sound: history-taking still yields more diagnostic information than any single test.

In practice

A doctor who asks when the pain started and what makes it worse has the diagnosis before the blood results arrive.
A support engineer who lets the customer describe what they were doing solves faster than one who goes straight to the logs.

Attributed to: William Osler

Page for this entry

Before committing, assume the plan has already failed and reason backwards to how.

Klein's 2007 technique. The mechanism is prospective hindsight, from Mitchell, Russo and Pennington's 1989 research, which found that assuming an outcome has already occurred raises the number of causes people generate by about 30% compared with asking what might go wrong.

In practice

A team asked what might go wrong lists four things; asked to write the post-mortem for a failure that has already happened, it lists eleven.
A launch review that opens with "it is six months from now and this failed — why?" surfaces the dependency that "any concerns?" never does.

Attributed to: Gary Klein; mechanism from Mitchell, Russo & Pennington
Also called: Prospective hindsight
Further reading: Wikipedia

Page for this entry

Inherited practice embodies accumulated judgement that no individual reasoner can reconstruct, so the burden falls on the reformer.

Burke's 1790 argument. It is the reasoning Chesterton later compressed into the image of the fence, and it places the burden of inquiry on the reformer rather than forbidding reform.

In practice

A court that follows a rule nobody can now justify from first principles is relying on the judgement of everyone who applied it before.
A profession's apprenticeship requirement encodes judgement nobody can articulate, which is an argument for asking before removing it rather than for never removing it.

Attributed to: Edmund Burke
Standing: The argument Chesterton's fence later compressed into an image.

Page for this entry

Interpret an argument in its strongest plausible form before answering it.

Also called steelmanning. Its practical benefit is selfish: an argument built against the strongest version of the opposition survives contact, and one built against the weakest does not.

In practice

A barrister who prepares against the weakest version of the other side's case is surprised in court by the strongest.
A team that can state the objection to their own design better than its critics can has usually found the real flaw first.

Attributed to: Analytic philosophy; Neil Wilson
Also called: Steelmanning
Standing: Also known as steelmanning.
Further reading: Wikipedia

Page for this entry

Refuting a claim by deriving a plainly false consequence from it.

A valid form, distinct from the fallacy of caricature. Euclid's proof that there is no largest prime is the classic: assume one exists, construct a larger, and the assumption dies.

In practice

A claim that any delay is unacceptable is tested by asking about a delay of one second, and the claimant supplies a threshold they had denied having.
A policy proposal is tested by asking what it implies at scale, and it implies something its author rejects.

Attributed to: Classical logic
Standing: A valid form, distinct from the fallacy of caricature.
Further reading: Wikipedia

Page for this entry

Forecast a project by the distribution of outcomes for similar past projects rather than by reasoning about this one.

Flyvbjerg's method, adopted by the UK Treasury Green Book. It works because it bypasses the inside view entirely: the question becomes what happened to projects like this, not what will happen to ours.

In practice

A couple budgeting a wedding from what other people's weddings cost lands closer than one pricing every item themselves.
A team that asks how long similar migrations took at other firms gets a better estimate than one that reasons carefully about its own.

Attributed to: Kahneman & Tversky; Flyvbjerg
Also called: The outside view
Further reading: Wikipedia

Page for this entry

If reversing a proposed change also seems bad, the objection is probably status quo bias rather than reasoning.

Bostrom and Ord's 2006 test. Applied to cognitive enhancement: if raising intelligence is bad and lowering it is also bad, the current level must be optimal, and that needs an argument nobody has.

In practice

A council that will neither raise nor lower the limit on a road it has never surveyed is asserting that the present figure is right by accident.
A team that opposes both lengthening and shortening the sprint has revealed that the objection is to changing it.

Attributed to: Bostrom & Ord
Further reading: Wikipedia

Page for this entry

Divide 72 by the growth rate to get the doubling time; compounding is fast enough to be counterintuitive without it.

Pacioli published it in 1494. At 6% growth, doubling takes twelve years; at 10%, seven. It is the reason compound interest and exponential spread both defeat intuition without arithmetic.

In practice

A saver told the fund returns 6% hears a number; told it doubles every twelve years, they open the statement.
A fee of 1.5% a year against 0.2% costs roughly a third of the final pot over forty years, which the arithmetic shows and intuition does not.

Attributed to: Luca Pacioli
Further reading: Wikipedia

Page for this entry

Ask what follows from the consequences, not just from the action.

Its practical form is asking what happens next and then what happens after that. Price controls, rent caps and pesticide programmes all look sound at the first order and fail at the second.

In practice

A rent cap holds rents for sitting tenants and thins the supply the next cohort will need, which shows up only in the second year.
A firm that cuts its training budget improves this year's margin and raises next year's recruitment cost by more.

Attributed to: Systems and strategy literature

Page for this entry

Those who make decisions should bear a share of the downside, or their judgement cannot be trusted.

Taleb's 2018 argument. Its sharp form is that advice is only credible from someone exposed to being wrong, which is why forecaster track records matter more than forecaster credentials.

In practice

A builder who guarantees his own roof for ten years quotes differently from one who subcontracts and moves on.
A consultant who recommends a restructure and leaves before it is implemented has given advice at no risk, which is information about the advice.

Attributed to: Nassim Nicholas Taleb
Further reading: Wikipedia

Page for this entry

Go where the money is - focus effort on the most likely or high-value targets.

Named for Willie Sutton, who denied ever saying it. As a clinical heuristic it means test for the likely thing first, and it is Occam's razor applied to diagnostic sequencing.

In practice

A patient arriving breathless in their sixties gets the cardiac workup before the rare metabolic screen, because that is where the answers are.
An engineer debugging a failed deployment checks the change made an hour ago before auditing the framework, and is right four times in five.

Attributed to: Named for Willie Sutton
Further reading: Wikipedia

Page for this entry

The best way to improve often is by removing negatives rather than adding positives.

Taleb's argument. Removing what harms is more robust than adding what helps, because the harms are more reliably identified — which is why medicine's biggest gains came from sanitation, not treatment.

In practice

A team that cancels three meetings and adds nothing recovers more capacity than the productivity tool bought the previous quarter.
A patient advised to stop two of their six medications improves, and no new prescription would have done it.

Attributed to: Nassim Nicholas Taleb
Also called: Subtractive knowledge

Page for this entry

If you use a ruler to measure a table, you also measure the ruler by the table.

Taleb's formulation. If a test's error rate exceeds the base rate of what it detects, a positive result tells you more about the test than about the subject.

In practice

A tape measure of unknown accuracy laid against a table tells you as much about the tape as about the table.
A screening tool that flags 3% of staff as risks in a population where 0.1% are risks is measuring the tool.

Attributed to: Nassim Nicholas Taleb, after Wittgenstein

Page for this entry

The most effective intervention is often the one that stops forcing and lets a situation resolve itself.

The Daodejing recommends governing a large state as one cooks a small fish, which is to say without constant turning. The difficulty is that the doctrine supplies no test: the ruler who does nothing because intervention would make things worse and the ruler who does nothing from indifference are indistinguishable at the time.

In practice

A manager who stops reorganising a functioning team finds output rises, having attributed the previous dip to the team.
A manager who applies the same restraint to a failing team is described afterwards as absent.

Attributed to: Laozi, Daodejing
Also called: Non-action; effortless action; governing by not interfering
Possible explanations: One explanation is informational: an official who intervenes constantly destroys the local knowledge that would have solved the problem. Another is psychological, since forcing an action recruits resistance that leaving it alone does not.
In its favour: The pattern it names is recognisable in medicine's watchful waiting, in rewilding, and in management research on autonomy, where restraint outperforms constant adjustment.
Against it: It gives no criterion for telling restraint from neglect, and the outcome is used afterwards to decide which one it was.
Standing: Dating and authorship of the Daodejing are disputed. The precept gives no criterion for distinguishing restraint from neglect, which is the standing objection to it.
Further reading: Wikipedia

Page for this entry

When you hear hoofbeats, think horses; prefer the common explanation until the common one is excluded.

Woodward's teaching from around 1940, and the bedside form of Occam's razor. Hickam's dictum is its necessary check, since the elderly and chronically ill routinely have several conditions at once.

In practice

A traveller home from Manchester with a fever has flu; the same fever after a month in the tropics earns the wider screen.
A support engineer who checks the power supply before the firmware is following it, and one who does that with an elderly complex system is not.

Attributed to: Theodore Woodward
Also called: Don't look for zebras
Standing: The bedside form of Occam's razor, with Hickam's dictum as its check.

Page for this entry

Figure

Sayings, images and puzzles.

Essay: Paradoxes that broke theories, and sayings that outlived them

People's choices between gambles reverse when a common outcome is added to both, which expected utility theory says cannot happen.

Allais put his pair of choices to an audience of economists in 1953, several of whom had helped build the theory, and most chose inconsistently. The pattern is the certainty effect: a sure thing is weighted more heavily than its probability warrants, so removing certainty from both options changes the ranking. Kahneman and Tversky built prospect theory on this result.

In practice

A buyer pays far more for the warranty that removes the last one per cent of risk than for one that removes ten points in the middle.
A gambler who refuses an even chance on his stake buys a lottery ticket with far worse odds.

Attributed to: Maurice Allais
Also called: The certainty effect
Standing: Robust and repeatedly replicated. The dispute was never about the finding but about whether to call it irrationality or the wrong model of preference. Compare the Ellsberg paradox.
Further reading: Wikipedia

Page for this entry

As a model grows realistic enough to explain a system, it becomes as hard to understand as the system itself.

Bonini's point applies to any simulation detailed enough to be trusted: at that point nobody can say why it produced its output. Large forecasting models and machine-learning systems hit the same wall.

In practice

A risk model with 400 parameters passes validation and cannot be explained to the board, so the board approves it on the strength of the validation.
A supply chain simulation reproduces last year exactly and nobody can say which of its assumptions is carrying the result.

Attributed to: Charles Bonini
Further reading: Wikipedia

Page for this entry

Two exactly equal options give no basis for choosing, so a purely rational chooser stalls.

Buridan never wrote it; the example is attributed to him by critics. Real choosers break ties arbitrarily, which is why the paradox is mainly useful for showing that indifference is not the same as inability.

In practice

A committee with two equally good candidates and no tiebreak rule meets four times and appoints neither.
Two identical products at the same price sit unbought while a third, slightly worse and clearly different, sells.

Attributed to: Ancient antecedent; later attributed to Jean Buridan
Possible explanations: The explanation is structural: a decision rule that selects the highest-valued option returns nothing when two options are tied, so something outside the rule, noise or an arbitrary convention, has to break it.
In its favour: It isolates a real feature of decision theory, that indifference is not the same as preference, and it is why practical systems build in tie-breakers.
Against it: No actual chooser stalls, so the paradox describes a property of a model rather than of behaviour, and the example does not appear in Buridan's writings.
Standing: The example does not appear in Buridan's writings; it is an ancient antecedent attributed to him by critics.
Further reading: Wikipedia

Page for this entry

Discrete moments of decision are claimed to determine life outcomes more than circumstances do.

The claim is attractive because it is actionable and misleading in proportion to how attractive it is. The evidence on cumulative advantage, opportunity hoarding and social closure describes outcomes shaped by position and by compounding, and the decisions available to a person are themselves largely set by that position.

In practice

Two people making the identical decision from different starting positions get different outcomes, reliably.
A doctrine that locates outcomes in decisions has explained success to the successful.

Attributed to: Tony Robbins
Also called: It is in your moments of decision that your destiny is shaped
Possible explanations: The explanation offered is that a decision commits a person to a path whose later steps follow, so the moment of choosing carries the weight. The researched version is narrower: implementation intentions convert a decision into behaviour by specifying when and where.
In its favour: Commitment devices and specific plans do improve follow-through, which is the defensible core.
Against it: Research on cumulative advantage, opportunity hoarding and social class finds circumstances explaining much of what the principle assigns to moments of decision, and the evidence offered is testimonial and selected on success.
Standing: Testimonial evidence selected on success, with no comparison group. Contradicted in emphasis by research on situational influence, on cumulative advantage and on the structural determinants of life outcomes, all of which find circumstance doing a great deal of the work.

Page for this entry

People prefer known risks over unknown risks, even with equivalent expected values.

Ellsberg's 1961 urns: people prefer a bet on a 50-50 urn to a bet on an urn with unknown composition, even after being offered the choice of colour. Uncertainty about the odds is treated as worse than bad odds.

In practice

A board approves a project with a 60% modelled success rate over one with an unmodelled but plausibly better chance, because the number feels safer.
A hiring panel prefers a candidate with a mediocre but legible track record to one whose record they cannot interpret.

Attributed to: Daniel Ellsberg
Standing: Compare the Allais paradox, which breaks the same theory through certainty rather than ambiguity.
Further reading: Wikipedia

Page for this entry

When it is not necessary to make a decision, it is necessary not to make a decision.

Falkland's rule from around 1640, and the discipline it demands is unusual: it forbids acting when action would be premature, which is harder than it sounds for anyone under pressure to appear decisive.

In practice

A team defers naming a delivery date until the discovery work is finished, and is criticised for indecision by people who would have criticised the wrong date.
A doctor who declines to diagnose before the second test looks less confident than one who commits, and is more often right.

Attributed to: Lucius Cary, Viscount Falkland

Page for this entry

Some problems are best not solved on their own terms; changing the terms is available and usually unconsidered.

Alexander cut it in 333 BC rather than untying it. The move's value is that changing the terms is nearly always available and nearly never considered, because the puzzle as posed is what everyone is looking at.

In practice

A firm facing an impossible integration deadline renegotiates the contract's scope instead of the timeline.
A dispute over who owns a shared driveway is settled by both parties selling to a developer, which nobody had proposed.

Attributed to: Alexander the Great
Further reading: Wikipedia

Page for this entry

The more urgent the need for decision, the less apparent become the consequences of that decision.

From the Murphy's law canon. It is the practical companion to the fog of war: urgency compresses the time available to think exactly when the consequences are largest.

In practice

A director asked for a decision by close of business chooses the option that can be explained fastest, not the one that survives a year.
A team forced to pick a vendor in a week picks the one with the best-prepared answers, which is a different quality from the best vendor.

Attributed to: Murphy's law canon

Page for this entry

Successful people are claimed to reach decisions promptly and revise them slowly, and unsuccessful people to do the reverse.

Hill drew the rule from his interviews. Read as advice it is dangerous in exactly the cases that matter, since slow revision is indistinguishable from refusal to update, and the successful people who exemplify it are those whose first decisions happened to be right. The defensible fragment is that reopening settled questions has a cost.

In practice

An executive who never revisits a decision is praised for resolve until the decision is wrong.
A team that reverses course on new evidence is accused of lacking direction by people using this rule.

Attributed to: Napoleon Hill
Also called: Decide quickly, change slowly
Possible explanations: The explanation offered is that promptness signals and builds confidence, while frequent revision signals its absence. The researched version separates the two: speed helps where delay is costly, and revision helps where new information arrives.
In its favour: Decisiveness does have value where information is not improving and delay carries cost, which is a real class of decisions.
Against it: It is directly at odds with evidence on belief updating, which rewards revision in proportion to new evidence, and the rule cannot distinguish slow revision from refusal to update. The evidence offered is testimonial and selected on success.
Standing: Testimonial evidence selected on success, with no comparison group. Directly at odds with evidence on belief updating, which rewards revising promptly on new information. The rule confuses decisiveness with accuracy.

Page for this entry

It always takes longer than you expect, even when you take Hofstadter's Law into account.

Hofstadter's 1979 law is self-referential by design, which is the point — building in a buffer does not help, because the estimate of the buffer is subject to the same law.

In practice

A team that added a 30% buffer to a six-month estimate delivers in eleven months.
A move scheduled for a weekend with a day spare takes the following week off as well.

Attributed to: Douglas Hofstadter
Further reading: Wikipedia

Page for this entry

If you can write the problem down clearly, the problem is half solved.

No identifiable source despite very wide repetition. It survives because the underlying observation is sound: most of the difficulty in a problem is in stating it precisely.

In practice

A team stuck three weeks on "the reports are slow" settles it in an afternoon once someone writes down which report, for which user, and at what hour.
The defensible version is that most of the difficulty is in the statement: a support ticket that says what was expected and what happened is usually resolvable, and one that says "it's broken" is not.

Attributed to: Unattributed, internet origin
Standing: No identifiable source despite wide repetition.

Page for this entry

The folk belief that outcomes must even out in the short run; large samples converge, individual trials do not.

Not a law. Only the law of large numbers holds, and only in the limit — a coin that has landed heads ten times has no memory, and the folk belief is the gambler's fallacy wearing respectable clothes.

In practice

A gambler who has lost eight hands believes a win is due, and the deck has no memory.
A hiring manager whose last three hires failed expects the next to succeed on the same reasoning that predicted the last three would.

Attributed to: Folk belief
Also called: Folk form of the gambler's fallacy
Possible explanations: The explanation for the belief is that people expect short sequences to look like the long-run average, so a run of one outcome feels overdue for correction.
In its favour: The underlying mathematics is real in the limit: with enough trials, proportions do converge on their expected values.
Against it: Independent trials have no memory, so nothing evens out in the short run. As used in everyday talk it is the gambler's fallacy under a respectable name.
Standing: Not a law; only the law of large numbers holds, and only in the limit.
Further reading: Wikipedia

Page for this entry

Most favorable predictions about the future fall within the speaker's expected lifespan.

Named by Garreau for Pattie Maes, who noticed that AI researchers predicted human-level machines shortly before their own retirement. It applies to immortality, singularity and fusion predictions alike.

In practice

An AI researcher of 50 predicts human-level machines in 25 years; the same researcher at 35 predicted 40.
Fusion, longevity and general intelligence forecasts cluster suspiciously near the forecaster's retirement.

Attributed to: Pattie Maes, named by Joel Garreau
Further reading: Wikipedia

Page for this entry

For every complex problem there is a solution that is neat, plausible and wrong.

Mencken's line is a warning about the relationship between a solution's appeal and its adequacy. The simple answer is easier to state, easier to remember and easier to campaign on, so it wins the competition for attention regardless of whether it fits the problem. The test it implies is practical: if an answer to a genuinely tangled question can be given in one sentence and provokes no objections, that is information about the answer.

In practice

A policy that fits on a poster is not thereby wrong, and the fact that it fits on a poster is not evidence that it is right.
A post-mortem that identifies one cause for a failure with eleven contributing conditions has produced something neat.

Attributed to: H. L. Mencken
Also called: The clear, simple and wrong answer

Page for this entry

A choice between two equally unpleasant alternatives that lead to the same result.

Named for John Morton, Henry VII's tax collector: the lavish can afford to pay and the frugal have savings to pay from. The structure is that the argument's branches converge on the same conclusion.

In practice

An employee told to justify a low expense claim looks profligate for having made one and disorganised for having no receipt.
A candidate asked why they left a job quickly looks unstable, and one who stayed long looks unambitious.

Attributed to: Named for John Morton
Further reading: Wikipedia

Page for this entry

When a reliable predictor has already acted on a prediction of your choice, taking one box and taking both are each defensible, and decision theory divides on which is right.

Newcomb devised it and Nozick published it in 1969. A predictor who has been right almost every time has put either a fortune or nothing into the opaque box, according to whether it predicted you would take that box alone. Causal reasoning says the contents are already fixed, so take both. Evidential reasoning says that one-boxers end up rich, so take one. Nozick reported that people divide sharply and each side thinks the other is simply failing to see it.

In practice

An employer who reads candidates well has priced the offer before the negotiation starts.
A friend who knew you would cancel has already made other plans, which is what you discover when you do not cancel.

Attributed to: William Newcomb; published by Robert Nozick
Also called: Newcomb's paradox
Standing: A live problem in decision theory rather than a claim about behaviour, and the standard test case between causal and evidential decision theory.
Further reading: Wikipedia

Page for this entry

Buyers choose the option that is easiest to justify if it fails rather than the one most likely to succeed.

The decision is rational for the buyer and poor for the organisation. Choosing the incumbent and failing is a misfortune; choosing the challenger and failing is a mistake with a name attached. As long as the personal cost of being blamed exceeds the personal benefit of being right, the safe option wins whatever the comparison says.

In practice

A procurement team with a better and cheaper bid in front of it selects the vendor everyone has heard of.
A doctor ordering the additional test is protecting against the account of events, not only against the illness.

Attributed to: Corporate procurement folklore
Also called: The defensible purchase
Standing: Compare the sucker effect and organisational justice; the mechanism is asymmetric personal risk rather than ignorance.

Page for this entry

Make explanations as simple as possible, but not simpler.

The as-simple-as-possible phrasing is a later paraphrase, not a verified Einstein quotation. The qualification is what earns it a place — Occam's razor without it licenses oversimplification.

In practice

A model cut to three variables predicts the pilot and fails the rollout, because the fourth variable was the one that moved.
An outage explained as human error is simple, fits the facts, and removes from view the four conditions that let one keystroke matter.

Attributed to: Attributed to Albert Einstein
Also called: Einstein's razor
Standing: The as-simple-as-possible phrasing is a later paraphrase, not a verified quotation.

Page for this entry

Unanimous agreement can reduce confidence in a conclusion, because beyond a point complete agreement is better explained by a fault in the process than by the truth.

Gunn and colleagues formalised it in 2016 using witness identifications. As confirming witnesses accumulate, the probability that the suspect is the right one rises and then falls, because a long run of identical answers is more likely to come from a biased line-up than from a series of independent observations. Ancient law had the intuition: the Talmud records that a unanimous conviction by the Sanhedrin acquitted the accused.

In practice

A committee whose twelve reviewers all score a proposal identically has learned less about the proposal than one with eleven agreements and a dissent.
A test suite that has never once failed is evidence about the test suite as much as about the code.

Attributed to: Gunn and colleagues
Also called: Too good to be true
Standing: Holds where systematic error is plausible. Where it is not, agreement is straightforwardly good evidence. Compare the Abilene paradox, where the agreement is real and nobody wants it.

Page for this entry

Diagnosis skips the trivial cause because it is beneath the diagnostician's dignity to check it.

The failure is social rather than logical. An expert asked to investigate a problem begins at the level of their expertise, and checking whether the device is switched on implies the questioner might be a fool. The cost is borne by the hours spent on sophisticated hypotheses before someone junior enough to be unembarrassed looks at the cable.

In practice

An escalated fault resolved by a restart was escalated past three people who would have felt foolish suggesting it.
A consultant's report identifies a strategic misalignment in a company whose invoices go to the wrong address.

Attributed to: Unattributed
Also called: It works better if you plug it in
Standing: Compare Sutton's law and Occam's razor, which prescribe checking the likely and the simple first.

Page for this entry

A man with one watch knows the time; a man with two is never sure.

From the Murphy's law canon. Its serious form is a real problem in instrumentation and forecasting: disagreeing sources cannot be averaged without knowing which is biased.

In practice

A ship carrying two compasses three degrees apart must choose which to steer by, and averaging them is a decision rather than a solution.
A company with two dashboards reporting different revenue figures has less usable information than one with a single imperfect figure.

Attributed to: Murphy's law canon
Further reading: Wikipedia

Page for this entry

Small changes that seem trivial can accumulate to produce significant changes.

Eubulides's paradox: one grain is not a heap, adding one never makes a heap, yet a heap exists. It is the standing problem for every threshold in law and policy, from age limits to blood alcohol.

In practice

A manager asked when the backlog became a crisis can name no week in which it did, and the crisis is on the agenda.
Every threshold in law and policy inherits the problem: no single day makes someone an adult, and a line must be drawn anyway.

Attributed to: Eubulides of Miletus
Also called: Paradox of the heap
Further reading: Wikipedia

Page for this entry

A gamble with infinite expected value is one that almost nobody will pay much to play.

Nicolas Bernoulli posed it in 1713 and Daniel Bernoulli answered it in 1738. A coin is tossed until it falls heads, and the prize doubles with every tail, so the expected payout is unbounded. Almost nobody will stake more than a modest sum. Bernoulli's resolution was the diminishing marginal utility of money, the first statement of a principle that now runs through economics; bounded utility and the finite wealth of the house account for the rest.

In practice

A pitch promising an enormous payoff at a tiny probability raises less money than a modest one with a plausible return, which is not irrational.
A lottery sells much the same number of tickets whether the rolling jackpot is large or enormous.

Attributed to: Nicolas and Daniel Bernoulli
Standing: Resolved rather than open, and historically important as the origin of utility theory. Compare the Allais and Ellsberg paradoxes.
Further reading: Wikipedia

Page for this entry

Perfect is the enemy of good; don't let perfectionism prevent completion.

Voltaire's 1770 line, borrowed from an Italian proverb. Its operational form is that the marginal value of further refinement usually falls below the cost of delay long before anyone stops refining.

In practice

A report circulated at four-fifths finished in March changes three decisions; the same report perfected and issued in September changes none.
A kitchen refit is held up eleven weeks over a tile finish nobody has mentioned since.

Attributed to: Voltaire
Also called: Perfect is the enemy of good

Page for this entry