The field guides glossary

Before deciding, ask how you will feel about each option in ten minutes, in ten months and in ten years.

In most online communities about 90 per cent only read, 9 per cent contribute occasionally, and 1 per cent produce most of the content.

A group can take a course of action that every member privately opposes, because each mistakes the silence of the others for agreement.

Lapses that come from divided attention at the moment of encoding or of acting, not from lost memories.

Construct an argument of the same form whose conclusion is plainly false.

The arguer is compared to something contemptible, and the comparison is offered as though it were an argument.

Subordinates perceive a sustained pattern of hostile verbal and non-verbal conduct from their supervisor, excluding physical contact.

Meaning is changed by where the stress falls, or by which part of a passage is reproduced, while every word attributed to the speaker is one he said.

A sound general rule is applied to a case that lacks the feature the rule was made for.

Two parties who gain from cooperating each take a local action that solves their own problem by obstructing the other, and the partnership decays without anyone intending it.

When a decision passes through many hands and a model, no single person feels they made it.

An automated decision passes through so many hands that no one of them can be held answerable for its outcome, and the system itself cannot be.

A structure that breaks the link between a decision and anyone who can be asked about it, so complaints arrive nowhere.

Acting by default rather than waiting, because visible effort feels safer than doing nothing.

We explain our own behaviour by the situation and the behaviour of others by character.

A refuted claim is preserved by adding an assumption whose only support is that it saves the claim.

The arguer is attacked in place of the argument, on grounds that would not bear on the claim even if every word of the attack were true.

The arguer is said to be so committed to a side that nothing he says on the subject can count.

The arguer’s position, interest or circumstances are offered as the reason his argument fails.

When one side knows more about quality or risk than the other, the better-than-average side withdraws and the market can unravel to nothing.

We read our feeling about a thing as information about its risks and benefits.

One member of an either-or is found to hold and the other is denied on that ground alone.

The consequent of a conditional is found to hold and the antecedent is inferred from it.

A safety argument is true of the system as it was when the argument was written, and quietly becomes false as the system, its staff and its demands change around it.

Media coverage transfers salience rather than opinion: the issues given most attention become the issues the public names as most important.

Confidence in using AI tools runs ahead of the understanding needed to judge what they produce.

What cannot be settled by experiment or observation is not worth debating.

Lay judges often weight advice more heavily when it is labelled as coming from an algorithm than from a person, but the preference reverses by task and fades among experts.

People abandon a statistical forecast after seeing it err, even when it is still more accurate than the human judgement they return to.

Systematic, repeatable unfairness produced by a system and its data rather than by a mind.

A matter of degree is argued as though it admitted only the two end states.

We tend to overestimate the effect of a technology in the short run and underestimate it in the long run.

Offered a known gamble and an unknown one, we take the known, even when the unknown may well be the better bet.

Speeding up one part of a process improves the whole only in proportion to that part’s share; the untouched remainder sets a hard ceiling.

The ambiguity sits in the grammar rather than in any word, so one sentence yields two readings and the argument quietly takes the one it needs.

An analogy drawn to make one point is carried on and made to yield conclusions it was never built to support.

The first number we are given pulls every later judgement towards it, even when it is irrelevant.

A single vivid case is treated as evidence of a general effect.

A single vivid case is offered where a rate is required.

We attribute minds, intentions and feelings to things that have none.

Authority is invoked without being named, so that it cannot be checked.

A practice is defended on the ground that it is what everybody does.

A proposition is affirmed or denied on the strength of what would follow from accepting it.

A frightening outcome the speaker does not control is described, and the description is offered in place of the case.

The hearer is offered a compliment, or an enemy’s discomfort, as his reason for agreeing.

A penalty is attached to disagreement, and the penalty is offered in the place a reason should occupy.

An expert opinion is cited where the expert is not an authority on the question, or where the question is not one authority can settle.

An argument is dismissed by supplying a motive for making it.

Something is held to be good because it is natural, or bad because it is not.

Treating an idea as better because it is new.

A thing is recommended by its newness.

Distress is presented where a ground was owed, so that refusing the claim becomes a refusal of the sufferer.

Treating something likely, or merely possible, as though it were certain.

A benefit to the hearer is supplied in place of a reason for the claim.

A claim is supported by the number of people who accept it.

A practice is defended by its age.

A claim is supported by the prosperity of the person making it, or by their lack of it.

A charge is conveyed without being asserted, so that it can be understood and cannot be answered.

A claim is repeated until familiarity does the work that evidence did not.

Concluding a claim is false because the argument offered for it was bad.

A conclusion is treated as false because the argument offered for it was bad.

The absence of evidence for a claim is offered as evidence that it is false, or the absence of evidence against it as evidence that it is true.

The arguer’s inability to imagine how something could be so is offered as a reason for thinking it is not so.

An observed indicator is taken to establish the condition it usually accompanies.

Each side arms because the other has, so both spend heavily and neither ends up safer than it began.

People will state an answer they can see is wrong when everyone ahead of them has stated it.

A regulator can control a system only if it can take at least as many distinct responses as there are distinct disturbances to counter.

A word never presented is recalled as vividly as the studied words that all point to it.

A campaign funded and directed by an interested party is presented as spontaneous public opinion, so that a sponsor’s position appears to be a constituency’s.

Our state and our preoccupations decide which cues reach us, and the rest go unseen.

An account of how checking spreads until organisations reshape their work to produce the traces an auditor can verify, and verification of the traces replaces knowledge of performance.

A claim gains weight from the rank of whoever made it, whatever the claim itself is worth.

The steeper the perceived power difference in a team, the less likely the junior member is to say what they have seen.

The machine’s answer wins over our own checks, and we stop looking at the evidence it was meant to summarise.

Operators monitoring a highly reliable automated system check it less and less, and so miss the rare occasion when it is wrong.

The machine does something the operator did not expect, and the operator cannot work out what it is doing, or why, in the time available.

A risk gains public credibility through repetition rather than measurement, because each retelling makes it easier to recall and makes doubt more costly to voice.

We judge how likely something is by how easily an instance of it comes to mind.

The claim that correcting a false belief strengthens it, which has largely failed to replicate.

The more people we see doing something, the readier we are to do it too, whatever its merits.

A vague profile that would suit nearly anybody reads to us like a sharp likeness of ourselves.

A vivid detail about one case drowns out the plain figure for how common such cases are.

A test result or a match is turned into a probability about the case without the prevalence of the condition being used.

Services whose output is people’s time get steadily more expensive relative to goods, because their productivity barely rises while wages rise everywhere.

The conclusion is assumed among the premises, so the argument is valid and supplies no reason to accept it.

We judge an argument valid when we agree with its conclusion.

A belief outlives the evidence that created it, even after that evidence is withdrawn.

Copying the practices of measured high performers makes organisations more alike without making them better, because the practices were never what produced the performance.

The outputs of a project are delivered while the benefits promised in its business case are never measured, never realised, or quietly redefined.

Passion is inversely proportional to the amount of real information available.

Helping someone raises our opinion of them, even when we disliked them before.

Copying the visible practices of a successful organisation imports the form without the conditions that made it work, and the conditions are usually what mattered.

Any headline that ends in a question mark can be answered with the word “no”.

We recognise bias in other people and not in ourselves, because we judge ourselves from the inside.

Asked what follows logically, people answer from content, word order and familiarity instead.

Replacing a working system wholesale rather than piece by piece trades a known set of defects for an unknown one, and buys an interval in which nothing ships.

Discussion time runs inverse to importance: groups debate the trivial and wave the consequential through.

A committee will spend more time on the item it understands than on the item that matters.

In AI, general methods that use more computation eventually beat methods built on human knowledge of the domain.

Odd or incongruous material is recalled better than ordinary material, but only under narrow conditions.

Technical language and apparatus are used to make a claim unchallengeable rather than clearer.

The two alternatives are exhaustive, and at least one of the consequences attached to them does not follow.

Telling people how they compare with the average moves those already better than average in the wrong direction, unless the message also says which direction is approved.

Durable regulation usually needs two coalitions: one that argues for it on moral grounds in public, and one that profits from it quietly.

Whatever we buy with the last of a budget gets blamed for the pinch, and is enjoyed less for it.

With limited time, knowledge and processing power, we take the first option that is good enough.

Judge whether an association is causal by weighing nine viewpoints — strength, consistency, specificity, temporality, gradient, plausibility, coherence, experiment, analogy — none decisive alone.

Adding a road to a network in which drivers choose their own routes can make every driver slower.

The energy needed to refute nonsense is an order of magnitude greater than the energy needed to produce it.

Visible small disorder, left unrepaired, is held to signal that nobody is watching, and so to invite larger disorder.

Adding people to a late software project makes it later.

An agency that is the sole supplier of its output to a sponsor who cannot see its costs can secure a budget well above what the service requires.

Safety time held inside individual task estimates is consumed rather than pooled, so a project can run out of protection long before it runs out of work.

Variation in end demand is amplified at each step back up a supply chain, so the stage furthest upstream sees swings many times larger than anything the customer did.

The arguer explains why you hold your mistaken view before establishing that it is mistaken.

Pay for things as a package and we use fewer of them, because no single item carries a price to justify.

The number of people who would have to leave before the work could not be carried on, and a warning that the number is usually smaller than anyone thinks.

A claimed priming effect in which the sound of bye calls up buy and nudges spending upward.

The more people who witness a situation that calls for help, the less likely any one of them is to act.

The more a quantitative indicator is used to make decisions, the more it will be corrupted and the more it will distort the process it was meant to monitor.

Copying the visible forms of a successful practice while missing the mechanism that made it work, and expecting the results to follow.

When the load carried by a failed element is redistributed to its neighbours, each with a finite margin, a single local failure can propagate until it takes down a system far larger than itself.

Card and phone payments hurt less than handing over cash, so we spend more.

An outcome feels likelier when it sits in a big group, though the size of the group says nothing about its odds.

One contributing cause is named as the cause of an outcome that several conditions jointly produced.

A first step is opposed because it is forecast to set off a chain of consequences ending somewhere unacceptable.

Organisations adopt the structures their environment rewards and then insulate the real work from them, so policy and practice run on separate tracks.

The five stages of grief, borrowed as a schedule for how staff will react to organisational change, despite failing as a description of grief itself.

If fewer things cannot account for what is known to be true, posit more — and keep adding until the explanation is adequate.

A face is rated slightly more attractive when seen in a group than when seen on its own.

The cases that support the conclusion are reported and the cases that do not are left out.

Do not remove a rule or a structure until you can say what it was put there to do.

A contest in which a threat works only if the other side believes you will not back down, so the advantage goes to whoever can visibly destroy their own ability to back down.

Too many options degrade the decision, or prevent it being made at all.

We remember the option we chose as better than it was, and the ones we rejected as worse.

A view is dismissed because of when it was held rather than because of what is wrong with it.

Always and inevitably, everyone underestimates the number of stupid individuals in circulation.

Act where you know enough to judge; know where that knowledge ends, and decline or defer beyond it.

The circle of competence says stay where you know the ground; explore–exploit says sample the unknown to find better options. The number of chances left to use what you learn decides.

The conclusion supports a premise which supports the conclusion, around a loop wide enough to be lost.

When a distinguished but elderly scientist says something is possible, he is almost certainly right; when he says it is impossible, he is very probably wrong.

Clarke warns experts against declaring things impossible; Sagan demands evidence in proportion to how improbable a claim is. The strength of the evidence against the prior decides.

Random data contains streaks and gaps, and we read them as structure.

A reward offered for reducing something creates a supply of the thing, so the scheme increases what it was meant to suppress.

When what we did clashes with what we believe, we ease the tension by bending the belief.

A group of people who all want the same outcome will not act to produce it, because each does better letting the others pay for it.

Early on, a technology is easy to change but its harms are unknown; once the harms are known, it has become hard to change.

Having backed a course, and been seen to back it, we defend it harder as the bad results arrive.

Coordinated action often requires not merely that everyone knows a fact, but that everyone knows that everyone knows it, and no exchange of fallible messages can establish that.

Redundant components that share a cause fail together, so a reliability figure obtained by multiplying independent failure probabilities can be wrong by orders of magnitude.

Measures that produce the appearance of safety or compliance for an audience while doing little to the risk they are said to address.

A statement about a class is made without saying how much of the class it covers, so that it can be defended as “some” and used as “all”.

Give up the part of your position that has been defeated, explicitly, and keep arguing for the rest.

A small group with a large stake each will organise; a large group with a small stake each will not, so policy tilts towards the few even when the many lose more in total.

The conclusion, if accepted, undermines the premises that were used to reach it.

A false account given sincerely, with no intention to deceive and usually no sense of doubt.

We search, read and remember evidence in ways that shield a hypothesis from rejection.

We test our favoured hypothesis directly and never run the test that would rule it out.

A conjunction is judged more probable than one of the things it is a conjunction of.

Everyone is conservative about the subject they know best.

Given new evidence we move our beliefs in the right direction, but by too little.

We recall past attitudes and behaviour as closer to our present ones than they were.

A belief spreads through a closed network by cascade and then seals itself, because the evidence that would refute it is issued by the sources the belief has already disqualified.

Material is easier to retrieve in the setting where it was learned than in a different one.

Retracted information keeps shaping inferences after the person has accepted the retraction.

A distinction is rejected because no sharp boundary can be drawn between its ends, or a boundary is rejected because any particular placing of it would be arbitrary.

The premises offered cannot all be true, so they establish the conclusion and its denial equally.

A thing looks better or worse depending on what we happened to see immediately before it.

A general rule is built out of a case that was exceptional in the respect that matters.

Any organisation that designs a system will produce a design whose structure copies the communication structure of that organisation.

Each person added to a group adds more links than hands, so the cost of keeping everyone aligned grows faster than the group does.

Organisations divide work competently and then fail to put it back together, because division is visible and integration is not.

The sheer number of unreliable reports of something is offered as a reason to think some of them must be reliable.

Where one person holds wide discretion, the organisation reorganises itself around access to that person, and proximity becomes the currency the formal structure no longer describes.

A criticism is refused on the ground that the critic has not read enough of the literature he is criticising.

No set of mutually inconsistent observations exists for which some human intellect cannot conceive a coherent explanation, however complicated.

When performance is judged on outcomes, providers raise the average by selecting the cases most likely to succeed and quietly shedding the rest.

As a qualification becomes more common, the level demanded for a given job rises, so more education buys the same position.

Every defeasible argument comes with a fixed set of questions it must be able to answer, and asking one is the whole counter-move.

Never assign a probability of exactly zero or one to anything that is not a matter of logic, or no evidence can ever move you.

Faces from our own racial group are recognised more accurately than faces from other groups.

The idea that a crowd dissolves individual judgement into a primitive collective mind comes from Le Bon and has no evidential support; systematic study finds crowd action patterned and bounded by shared identity.

An idea comes back as our own because we remember the idea but not where we got it.

Material we cannot recall is often still stored, and fails only because the right cue is missing.

To find out who did something, ask who benefits from it.

Two things found together are taken to stand in a causal relation on that ground alone.

The quickest way to get the right answer online is not to ask a question but to post a wrong answer.

Time and attention move to the tested subjects, so the curriculum shrinks to the shape of the accountability measure.

Once we know something we cannot model a mind that does not, so we explain it badly.

Companies selected as exemplars by a bestselling management book tend to fall back towards the average soon after publication, leaving the principles drawn from them unsupported.

A position is established by discrediting the other options on a list the arguer drew up.

Interfaces built so that the choice favouring the operator is easy and the choice favouring the user is effortful, hidden or embarrassing.

The claim that narcissism, Machiavellianism and psychopathy are concentrated at the top of organisations and account for much of the harm done there, which the measured literature supports far more weakly than the claim is made.

A person or body answering an accusation denies the conduct, attacks the credibility of whoever raised it, and reverses the roles so that the accused appears as the injured party and the accuser as the aggressor.

Each choice wears judgement down, and later ones drift towards the easy default.

The sense that things are sliding, drawn from a tidied past and a noisy present.

An option nobody would pick can still steer the choice between the two that matter.

We assign blame for a mishap in whatever way leaves us least likely to suffer one.

A contested term is defined into another term, so that a substantive question is settled by the definition rather than by evidence.

A term is redefined in the middle of an argument so that an objection made against the original meaning no longer applies.

Anonymity does not release restraint so much as hand control of behaviour to whatever the local norm happens to be.

A correcting loop that acts on delayed information will overshoot, and if it corrects hard enough relative to the delay the overshoot grows into a standing cycle.

Each successful handover to an automated system invites the next, and the human checks thin out along the way.

We set our clearest and most detailed goals for the parts of life that matter least.

The same sum of money is easier to spend as small units than as one large one.

The antecedent of a conditional is found not to hold and the consequent is denied on that ground.

What most people actually do and what most people approve of are two different norms, and a message that confuses them can push behaviour the wrong way.

Work is reorganised so that the knowledge needed to do it moves out of the worker and into procedures, machines or management, leaving a narrower job behind.

The reaction to a behaviour increases the behaviour, because the response isolates the people labelled and makes the label the most available identity they have.

Dissent that is assigned rather than felt reassures a group without changing its mind.

Poor performers are moved into management because that is where they can do least damage to the work.

We cash in winners early and hold on to losers, waiting to get back to even.

Differences that loom large side by side barely register once we live with the choice.

A request large enough to be refused makes a smaller follow-up more likely to be granted, because the retreat reads as a concession that ought to be reciprocated.

Where travellers can switch between driving and public transport, road speeds settle at the speed of the alternative, so widening the road can leave everyone slower.

Evidence that supports one conclusion is used to announce a different and usually stronger one.

If it looks like a duck, swims like a duck and quacks like a duck, it is probably a duck.

The duck test says fluent explanation means understanding; Morgan’s canon says credit the lesser capacity if it explains the behaviour. The base rate of fluency without understanding decides.

A cognitive limit of about 150 is said to cap the number of stable relationships a person can hold, and so the size of any group that runs on relationships alone.

The skill to do a task is the skill to judge it, so the least able overrate themselves most.

How long an experience lasted barely enters our later judgement of how good or bad it was.

An association measured between groups need not hold, and can reverse, for the individuals inside them.

People and organisations continually trade thoroughness for efficiency, and because the trade almost always works, the adjustment that normally produces success is the same one that occasionally produces the failure.

We rate an outcome more highly for having worked hard to get it.

The response to a perceived unfairness that neither complains nor leaves, but quietly reduces what is given.

We recall our own contribution more readily than anyone else, and claim a larger share of the result.

Everything should be made as simple as possible, but no simpler.

The method that worked last time comes to mind first and stops us looking for a better one.

From a calm state we badly misjudge how hunger, fear or craving will steer what we, or others, actually do.

When admitting the obvious would mark a person as incompetent or disloyal, a plainly false claim can be maintained in public by everyone at once.

The people a decision is about are absent when it is made, so the standing remedy is to seat them, or someone who is accountable to them and can lose the argument on their behalf.

Owning something raises what we demand to give it up above what we would have paid for it.

A term keeps its shape across an argument and changes its sense, so that a valid-looking inference joins two claims that are not about the same thing.

When performance falls short, the gap can be closed by lowering the goal, and a goal that is allowed to move will move.

Where reporting is safe, reported error rates rise, so a falling error count is as likely to mean that people have stopped telling you as that the work has become safer.

Two parties each act to restore their position relative to the other, so every move provokes the counter-move that makes it necessary again.

A regularity observed in a group is explained by an inner nature the group members are taken to share.

A word’s origin is offered as evidence of what it now means, or of what the thing it names ought to be.

The improbability of an event is calculated after the event has been singled out, and the improbability is offered as evidence.

A conclusion is drawn from two premises both of which are negative, from which nothing follows.

A conclusion asserting that something exists is drawn from premises that assert only universals.

A causal account of why something happened is answered as though it were a justification of it.

Balance trying new options against using the best one known: explore while there is long to benefit, exploit as time runs out.

We assume other people are in it for pay and security, while we are in it for the interest.

The unpleasantness attached to a memory fades faster than the pleasantness attached to one.

Fallacies as strictly defined are almost never committed; the moves that are common almost never meet the strict definition.

We take our own view for the common one and expect more agreement than we will find.

Members of a group over-estimate how many colleagues share their reading of a decision, and score silence as agreement.

A categorical statement is reversed, subject for predicate, where its form does not permit it.

Two options are presented as the whole field, and a third that was available is not mentioned.

Two things are placed on a par by a shared feature, where they differ by a wide margin in the respect that matters.

A whole event that never happened can be remembered, with detail and confidence, as if it had.

A figure is given to a sharpness its source cannot support, and the sharpness is what persuades.

Two cases are treated as alike in the respect at issue on the strength of likeness in some other respect.

An incumbent defends its position not by making a case but by raising doubts that make any change look risky, so that the safe course is to do nothing.

A penalty attached to a behaviour can license it, by converting a moral obligation into a service with a stated price.

A propaganda pattern defined by high volume across many channels, rapid repetition, and no commitment either to accuracy or to consistency between its own claims.

High-volume, multi-channel, rapid and inconsistent messaging is used to make the truth unlocatable rather than to establish a claim.

The time to reach a target grows with its distance and shrinks with its size: far, small targets are slow to hit; near, large ones are quick.

A remedy that relieves the symptom quickly sets in motion a slower consequence that returns the symptom, larger.

Whatever we are thinking about seems to matter far more to happiness than it does.

A delusion held by two people in a closed relationship, sustained less by transmission from one mind to the other than by the absence of anyone outside able to contradict it.

Agreeing to a small request raises the chance of agreeing to a much larger one later, because the first compliance changes how the person describes themselves.

Grading a workforce against a fixed distribution and cutting the bottom band, a practice whose claimed benefits rest largely on simulation rather than measurement.

The habits and informal authority that let a founder build an organisation become the constraint on it once the organisation outgrows what one person can hold.

A syllogism appears to have three terms and has four, because one word has been used in two senses.

Describe the same facts as a gain or as a loss and people choose differently, though nothing else has changed.

The more equally attractive two options seem, the harder they are to choose between — yet the closer they are, the less the choice matters.

The right to hold a view is offered as a reason to accept it, or as a defence against its being examined.

When a benefit cannot be withheld from those who do not pay for it, the rational move is to let someone else pay.

After noticing something for the first time, we see it everywhere and take that for a real increase.

An object’s usual job hides its other uses, so the tool we need sits unseen in front of us.

We read other people’s actions as proof of character and overlook the pressures they were acting under.

Rejection by the established view is offered as evidence that the view is right.

A complex system that works has almost always grown from a simple system that worked; one designed complex from scratch rarely works.

We expect chance to even itself out, so after a run of one outcome the other feels overdue.

A run of one outcome in an independent process is offered as a reason to expect the other.

Where goals are unclear and attention is scarce, decisions are made by the chance meeting of problems, solutions, participants and occasions rather than by anyone solving anything.

A sustained attempt to make a person distrust their own perception and memory, so that their account of events can be discounted — by others, and eventually by themselves.

We spot a newspaper’s errors on subjects we know, then trust the same paper completely on subjects we do not.

Material we produce ourselves is remembered better than the same material read.

A claim is judged by where it came from rather than by what supports it now.

So many claims are made so quickly that answering any of them costs more time than the audience will give.

Women and minority staff move normally through the lower and middle grades and then meet a barrier at senior level that their qualifications, tenure and performance do not explain.

Women and minority candidates are appointed more often to leadership posts that are already in trouble, so their tenure starts with a raised chance of visible failure.

A procedure adopted to achieve an end becomes the end, and is defended even where it no longer serves the purpose.

A gap in the available explanation is filled with a preferred explanation, on the ground that nothing else currently occupies it.

As an online discussion grows longer, the probability that someone compares something to Nazis or Hitler approaches one.

Low expectations held by a teacher or a supervisor are said to depress the performance of the people they are held about, the mirror image of the Pygmalion claim.

A measure that was informative while nobody was trying to move it stops being informative once people are required to move it.

When a search can find it, we keep the location and let the answer go.

When bad and good money must be accepted at the same value, the bad circulates and the good is hoarded: bad money drives out good.

Any sufficiently advanced incompetence is indistinguishable from malice.

Assume a speaker is being informative, truthful, relevant and clear; where they seem not to be, the gap is where their real meaning lies.

We read a group decision as the preference of every member, and one member as the whole group.

A group ends its discussion further out than its members started, and more confident about it.

After discussing an issue with like-minded others, people hold more extreme versions of the views they arrived with.

A group converges on one view and its members quietly abandon the doubts they came in with.

A cohesive group under pressure is said to suppress dissent and converge early on a course of action worse than any member would have chosen alone.

Growth loads capacity, service quality falls, demand falls with it, and the fall in demand is read as evidence that the capacity was never needed.

A claim or a person is discredited by naming somebody disreputable who shares the position.

An overall impression of a person colours every separate judgement made about them, so ratings of distinct qualities move together and carry far less information than their number suggests.

One good quality spills over, so whoever impresses in one way gets rated well in every other.

Knowing that a company performed well leads observers to rate its culture, leadership and strategy highly, so studies asking what made great companies great are measuring the result rather than the cause.

Never attribute to malice that which is adequately explained by stupidity.

Hanlon says assume error before malice; cui bono asks who gains; Cipolla says harm may serve no one. Whether the mistakes consistently benefit someone decides.

Confidence exceeds accuracy on hard question sets and falls short of it on easy ones.

A general claim is drawn from a sample too small to have shown anything else.

Behaviour is said to change because it is being observed — a claim named after a study whose own data do not support it.

The wording is chosen in advance to be loose enough that any outcome can afterwards be claimed as the one predicted.

Rules of thumb that make judgement fast and cheap, and sometimes systematically wrong.

The time to choose between equally likely options rises with the logarithm of how many there are, not in proportion to their number.

Patients can have as many diseases as they damn well please: several causes may be present at once.

When the information needed for the best decision is divided among members, groups discuss what they already share and miss the answer.

A few organisations operate hazardous technologies with far fewer accidents than their risk profile predicts, and the explanation offered is a set of organising practices rather than better technology.

Once we know the outcome, we misremember having expected it.

Once reviewers know how an episode ended, they cannot recover how uncertain it looked beforehand, and the outcome makes the warning signs seem obvious and the decision negligent.

Where evidence is thin or unwelcome, the question is settled by the most senior person in the room.

What is asserted without evidence may be dismissed without evidence.

It always takes longer than you expect, even when you allow for it taking longer.

A position is attributed to an opponent who cannot be shown to exist, and is then refuted.

Ties form between similar people far more often than chance allows, so a pipeline built on referral reproduces its own composition, and a criterion called fit turns that pattern into a rule.

After a run of successes we expect another, though the evidence for real streaks is mixed.

Calling an outcome human error names the last person in the chain and ends the inquiry, turning the thing that needs explaining into the explanation.

Humorous material is recalled better than neutral material, largely at the expense of the material around it.

Improvement means deterioration: an announced improvement usually disguises a cut in what you receive.

A short wait cuts a reward’s value steeply, so smaller-sooner wins until both options are far away.

With enough users, every observable behaviour of a system will be depended on by somebody, whatever the documentation promises.

One named person with a face draws more sympathy and money than thousands counted in a statistic.

You understand an opposing view only when you can state it so well that its own adherents cannot tell you from one of them.

Both sides of a question are answered with equal warmth, each under a different sense of the key word, so that a position appears to have been taken and none has.

A conclusion is established soundly, and it is not the conclusion that was owed.

A correlation produced by a third factor acting on both variables is reported as a link between them.

Things we assembled ourselves seem worth far more to us than they are to anyone else.

What a statement implies but does not assert is treated as though it had been asserted, so that a contrast the speaker never made is used as a premise.

A term is taken in its whole extent in the conclusion having been taken in part only in the premise.

We believe we know others better than they know us, and that our side sees theirs more clearly.

We credit ourselves with steering outcomes that are really down to chance or to other people.

We think we understand how things work in detail, until we try to explain one.

We credit whoever handed us an outcome for a satisfaction we produced ourselves.

Our nerves and intentions feel written on our faces, but other people read far less of them than we assume.

Evidence that hangs together feels predictive, so we trust forecasts that weak data cannot support.

Improvement that did not occur is remembered, because the past self is reconstructed to flatter the present one.

We see two things moving together when, counted properly, they have nothing to do with each other.

Most people place themselves above the median on most desirable traits, which cannot be true of most people.

Each repeat of a claim makes it more believable, whether the claim is accurate or not.

We expect good and bad news to hit harder and linger longer than it does once it arrives.

A reaction-time task whose scores are widely read as a measure of hidden prejudice.

Group-linked associations activate automatically, whether or not the person endorses them.

We favour people who share our label, even when the label was handed out at random.

Using rewards and penalties to steer behaviour, on the assumption that people will do what pays.

New road capacity is filled by traffic that did not exist before it was built, so congestion returns to close to where it started.

Disagreement among experts is taken to show that nothing in the field is known.

Seeking more information even when no possible answer would change what we do.

When people act in sequence and can see each other’s choices but not their reasons, it becomes rational to ignore your own information, and the group converges on a position resting on the first few signals.

Each person sensibly defers to those who spoke first, and the group ends up knowing less than its members.

Where one person or unit is the only route to something the organisation needs, withholding it is a rational way to hold position.

Initiatives are added faster than they are retired, so each receives too small a share of attention to change anything.

Activity that displays innovation — labs, hackathons, accelerators, sprints — while the budgets, procurement rules and decision rights that would let anything new be adopted stay exactly as they were.

How likely a sample result is gets judged without using how large the sample was.

Organisations in the same field come to resemble one another, not because the shared form works better, but because resemblance secures legitimacy and resources.

Since structure shapes architecture, move the team boundaries first and let the system you want follow.

To work out how to succeed, first work out how you would fail, and then avoid those things.

A claim is offered with no support beyond the fact that it has been asserted.

Large projects come in over budget, over time and under benefit, at a rate that has not improved across the decades for which records exist.

Any organisation large enough to need full-time officials tends to be run by them, though the evidence makes this a tendency rather than the iron law Michels claimed.

A legislative committee, an agency and an organised interest can close a policy area to outsiders, because each holds something the other two need.

Automating the easy parts of a job leaves the operator the hard parts, less practice at them, and the duty of taking over exactly when things are worst.

An evaluative conclusion is drawn from premises that are wholly descriptive.

Performance falls before it rises after a change, because the supports for the old practice go before the new practice is competent.

Users spend most of their time on other sites, so they expect yours to work the way those sites already do.

Making the use of a resource more efficient can increase the total amount of it consumed, because efficiency makes the service it provides cheaper.

No matter who you are, most of the smartest people work for someone else.

An assertion is made in the form of a question, so that it need not be defended.

A reporting system works only when staff know in advance where the line falls between an honest mistake, which will be treated as information, and a reckless act, which will not.

Assuming the world settles its accounts, so that a person’s misfortune is read as something they earned.

Denial of the charge is counted as evidence for it, so that no answer is available to the accused.

A system should remain secure even if everything about it except the key is public; secrecy of design is not a defence.

Debugging is twice as hard as writing code, so code written as cleverly as you can manage is, by definition, beyond your ability to debug.

Several defences are offered together, each adequate alone, which cannot all be true at once.

Because layering a new rule on the old ones is politically cheaper than replacing them, policy accumulates complexity, and the cost of that complexity is paid in citizens’ time.

Technology is neither good nor bad; nor is it neutral.

An official label applied to a person changes how others treat them and how they come to see themselves, and can produce more of the behaviour it names.

Within spaced study, the wider the gap between repetitions, the longer the material lasts, up to a point.

Every abstraction that hides real complexity will sometimes fail to hide it, and then you need to understand what lies underneath.

Having mastered a method, we start finding problems everywhere that seem to call for it.

Publishing ranked institutional results changes behaviour inside the institutions, improving the published measure and pulling effort towards whoever sits nearest the line.

A state must simplify what it administers into categories it can read, and when the simplification is then imposed as a design, it destroys what made the thing work.

Raters differ systematically in where they place people on a scale, some high, some low, some all in the middle, so a rating carries information about the rater as well as about the person rated.

Judged on its own, the lesser option can beat the option that would beat it in a direct comparison.

Each retelling of a memory trims the minor details and inflates the few that carry the point.

Working with what material means makes it stick far better than handling its surface or repeating it.

A reinforcing engine of growth eventually meets a constraint, and pushing the engine harder does nothing but raise the cost of hitting it.

For non-perishable things such as ideas, books and technologies, every year survived adds to the expected years remaining.

Lindy says the old outlasts the new; Amara says we underrate a technology’s long-run effect. What you are forecasting, and over what horizon, decides.

A long-disputed claim that the language we speak shapes what we notice and think.

Linus and Cunningham say open errors attract correction; Brandolini says refuting nonsense costs far more than producing it. Whether the crowd is cooperating decides.

Given enough eyeballs, all bugs are shallow: with a large enough base of testers and co-developers, problems are found quickly and the fix is obvious to someone.

The share of a list we recall falls as the list grows, even as the total recalled rises.

Given enough opportunities, one-in-a-million events happen to ordinary people about once a month; a striking coincidence is expected, not evidence.

Roughly four per cent of survey respondents will endorse almost anything, so small percentages in a poll are mostly noise.

A question is put whose presupposition cannot be rejected by any answer it allows.

People do what makes sense given their goals, their knowledge and their focus of attention at the moment, so an action that looks irrational afterwards was reasonable from inside.

A fallacy label is used to end an exchange rather than to improve it.

Search enough places and something will cross the line by chance, so a single hit means little on its own.

Losses sting more than equal gains please, by a disputed factor of about two.

A relationship estimated under one policy regime cannot be used to predict what a change of regime will do, because the behaviour that produced it will change too.

A capacity limit, not a bias: immediate memory holds a handful of chunks, and probably nearer four than seven.

Effort spent shaping how a manager sees the work is rewarded about as well as the work itself, so substituting the first for the second is a rational response to how appraisal is built.

Availability is restricted, or made to appear restricted, so that the thing is valued more highly and decided on more quickly.

A propaganda model in which five structural filters, rather than censorship, make a commercial press system produce coverage broadly aligned with the interests of those who fund and source it.

A representation is not the thing it represents; every model leaves things out, and what it leaves out decides where it fails.

Leave a buffer between what you expect and what you need, sized to how wrong you could be.

Technology changes exponentially; organisations change logarithmically — and the gap between them keeps widening.

Two terms for the same thing are swapped inside a claim about what somebody knows or believes, and the claim changes truth value.

Concluding that two things differ because someone knows them under different descriptions.

Physical symptoms spread through a group by observation and expectation rather than by any organic agent, most often in closed settings such as schools and factories.

Advantage compounds: an early lead in resources, credit or recognition attracts more of the same, so small initial differences widen into large ones.

Measure what is easy, disregard what is not, then treat what cannot be measured as unimportant and finally as non-existent.

Assume you are observing from a typical place and time, not a special one, unless you have evidence otherwise.

The cost of a meeting is the time of everyone in it, and no budget records that, so meetings are treated as free.

Retrieving some material actively suppresses related material that was not retrieved.

Money gets sorted into pots by source and purpose, and each pot is spent by its own rules, though a pound is a pound.

Liking grows with repeated contact, even when nothing about the thing has changed but its familiarity.

We decide whether to believe something by sizing up the person who says it.

The value of a communications network grows in proportion to the square of the number of connected users.

The practical knowledge built up by doing a particular job in a particular place cannot be written down as a procedure, and is lost when the procedure replaces it.

The belief that measurement can replace judgement, that published numbers secure accountability, and that people are best motivated by rewards tied to those numbers.

The position between two stated positions is taken to be the correct one because it lies between them.

Ordinary people continued to deliver what they believed were dangerous shocks because a man in a laboratory coat asked them to.

The claim that believers respond to a publicly failed prophecy by proselytising harder rests on one contaminated case study, and later surveys of failed prophecies find it is not the usual response.

People can hold only a handful of items in mind at once — Miller said about seven; later work puts it nearer four.

Information supplied after an event alters what a person remembers seeing.

People believe others are moved by money and security while believing themselves moved by interest and mastery.

A figure called the average is offered as the typical case, in a distribution where it is not.

One person is singled out for sustained hostile acts by a group of colleagues, so that the harm comes from the pattern and its duration rather than from any single act.

The last few items in a list are recalled better when heard than when read.

We read money at its face value and forget what it will buy.

A population of near-identical units is efficient and uniformly vulnerable: one pathogen, exploit or design flaw matched to the shared type takes all of it at once.

Our current mood makes material that matches it easier to recall than material that does not.

The number of transistors that can be placed economically on a chip doubles roughly every two years.

When a largely automated system fails, responsibility concentrates on the nearest human operator, who had neither the control nor the information to prevent it.

Once someone else carries the cost of an outcome, the party who controls it takes less care, and no bad faith is required.

Having done something good makes us freer to do something questionable next.

We praise and blame people for outcomes they did not control, which philosophy calls moral luck.

A group or condition is defined as a threat to shared values and the reaction runs ahead of any measured harm, leaving behind institutional changes that outlast the episode.

A descriptive conclusion is drawn from a premise about what ought to be the case.

Do not explain a behaviour by a higher mental faculty if a lower, simpler process can account for it.

A question containing a wrong but fitting term is answered rather than questioned.

Not knowing what we will win can make us try harder than knowing we will win more.

Paying people for something they were doing out of duty, reciprocity or identity can reduce how much of it they do.

An ambitious claim that could not survive challenge is advanced under cover of a modest one that can, with the arguer retreating to the modest claim when pressed and returning to the ambitious one when not.

Two claims share a name, one striking and wanted, one modest and defensible, and the arguer occupies whichever the moment requires.

Relieving the constraint in a flow system does not remove the constraint; it moves to the next resource, and effort spent anywhere else adds cost without adding throughput.

Evidence of the kind demanded is produced, and a further or different demand is substituted for it.

People withhold bad news from those it concerns, even when they had nothing to do with it and no reason to fear blame.

To explain or predict behaviour, look first at how people are rewarded; incentives usually outweigh instructions, exhortation and stated values.

Any text that criticises someone else’s editing or proofreading will itself contain an error, and the harsher the criticism, the worse the error.

The adage that anything that can go wrong will go wrong, sustained by selective memory.

The man-month is a unit that assumes people and time can be traded for one another, which holds only for work that requires no communication.

Given several options at once, we split evenly between them, so whoever draws the categories sets the shares.

We expect other people to be more self-serving than they actually turn out to be.

Believing we see things as they are, so anyone who disagrees must be misinformed or biased.

Settle who has to show what, before the evidence is discussed.

A natural property is offered as the meaning of “good”, so that an evaluative question is answered by a definition.

An event that could have caused harm but did not is filed as evidence that the system works, rather than as evidence that it nearly failed.

A condition without which the outcome cannot occur is treated as one that brings it about, or the reverse.

Bad things register louder and stay longer than good things of matching size.

Under uncertainty, the probability is not weighted badly — it is not used at all.

We remember little of what happened in the moments just before our own turn to perform.

The first 90 per cent of a project takes 90 per cent of the time, and the remaining 10 per cent takes the other 90 per cent.

A workable proposal is rejected by comparison with an ideal that is not among the options.

A counterexample is met by narrowing the definition of the subject term until the counterexample no longer falls under it.

A company’s good deeds make its products seem better, even on qualities the good deeds cannot touch.

Work that has to be done but earns nobody a promotion, distributed not by skill but by who is expected to say yes.

In systems that are both interactively complex and tightly coupled, serious accidents are a property of the design rather than a failure of the operators, and cannot be engineered away by adding more safety devices.

Because disaster has never happened to us, we treat it as something that will not, and react late when it does.

A group repeatedly accepts a departure from its own standard without harm, and the departure becomes the standard against which the next one is judged.

A warm memory of the past loosens our grip on money, making us readier to spend it and give it away.

Ideas, tools and evidence from outside are rejected for being from outside.

A group discounts knowledge, tools or solutions because they came from outside it, and pays to rebuild what it could have adopted.

What the researcher hopes to find leaks into how participants behave and how their behaviour gets recorded.

A rule for choosing between explanations that fit equally, not a claim that the simple one is true.

Among explanations that fit the evidence equally well, prefer the one that needs the fewest assumptions.

Occam says cut the extra cause; Chatton says add one when the lean account leaves facts unexplained. The unexplained residue decides.

Harm done by doing something feels worse to us than the same harm let happen by doing nothing.

A document that undertook to weigh a question presents only the considerations on one side of it.

We expect life to go better for us than for the people around us, and plan on that basis.

Add a mandated percentage to a project’s own cost estimate at approval, set from the outturn of past projects of the same type, and reduce it only as specific risks are actually retired.

Organisational silence is the collective withholding of concerns and dissent, produced by shared beliefs about what it is safe to say rather than by individual timidity.

Which way a familiar object faces is remembered wrongly, confidently, after thousands of exposures.

When the news is likely to be bad, we leave it unopened, including news about how we ourselves are doing.

Commons institutions that last for generations share eight design features, most of which work by making monitoring cheap and defection visible.

We perceive less variety among members of other groups than among members of our own.

An operator who has not been doing the task loses the picture of its current state, and so recovers slowly and badly when handed control.

We grade a choice by the result it happened to get, not by the case that could be made for it at the time.

Confidence in our judgements exceeds their accuracy, though how much depends heavily on how the questions were chosen.

A reward for something we already enjoy can replace the enjoyment as the reason for doing it.

Rewarding an activity someone already chose to do can reduce the chance they choose it again once the reward stops.

At any moment only a band of policy options can be advocated by someone who wants to stay in office, and what changes a policy is usually the movement of that band rather than an argument about the option.

A generalisation is stated with its force intact and qualified until almost nothing is left inside it.

A chatbot that talks like a friend gets trusted like one, though it is a product with someone else’s goals.

Seeing a definite thing, most often a face, in material that contains nothing of the kind.

In many systems a small share of causes produces most of the effect; find the vital few before treating the many.

Work expands to fill the time available for it, and administrative staff grows at a rate largely unrelated to the work there is to do.

Being shown some of the items you are trying to recall makes the remaining ones harder to retrieve.

Paying for measured output reliably raises the measured output, and reliably degrades everything the measure leaves out.

We store an experience as its sharpest moment plus its ending, and its length barely registers.

Safety regulation delivers less than promised because people take more risk once protected.

A theory of human behaviour can make itself true, because people use it to act, and the world then moves towards the model rather than the model towards the world.

A familiar word keeps its emotive force while its descriptive content is quietly changed, so that approval or disapproval is transferred to something new.

We expect the bad version of events, for ourselves or for the world, more often than the record supports.

Promotion on performance in the current job moves people up until they reach one they cannot do, and there they stay.

A manager uses the authority of a small post to control and belittle subordinates in ways that serve the manager rather than the work.

Concepts studied as pictures are recalled and recognised better than the same concepts studied as words.

A pilot succeeds on conditions a rollout cannot reproduce, so the better it is run the less its result tells you.

Effects measured in small trials run by the people who designed a programme shrink, often to nothing, when it is evaluated independently at scale.

Improvement produced by the expectation of a treatment rather than by the treatment.

A new scientific idea wins not by converting its opponents but because they eventually die and a generation familiar with it replaces them.

Plans built from the inside run late and over budget, because the forecast ignores how similar projects went.

Each person doubts the rule in private and stays quiet, sure that everyone else believes in it.

A group sustains a norm that most of its members privately reject, because each believes the others accept it.

Without a clear signal of intent, a parody of an extreme view cannot be reliably told apart from a sincere statement of it.

Adverse information about the arguer is supplied before he speaks, so that whatever he says arrives already discounted.

A system pushed towards one actor’s goal pushes back, because everyone else acting on it works harder to restore the position they wanted.

Evidence is commissioned, selected and timed to support a decision already taken, so the appearance of an evidence base survives while its function does not.

Pleasant material is processed, judged and recalled more readily than unpleasant material of the same intensity.

An affirmative conclusion is drawn where one of the premises denies rather than affirms.

Older adults attend to and remember positive material more than negative, relative to younger adults.

One event is taken to have caused another because it came first.

After choosing, we upgrade what we bought and downgrade what we turned down.

Be conservative in what you send and liberal in what you accept: emit strictly correct output, tolerate imperfect input where its meaning is clear.

Postel says accept what others send liberally; Hyrum says every behaviour you tolerate becomes a dependency. How long the interface must live, and with how many users, decides.

In any bureaucracy, those devoted to the organisation itself come to control it, and those devoted to its stated purpose lose influence.

Groups quietly adapt written procedure to the conditions they usually face, and the adaptations only collide on the day the groups are coupled tightly together.

Before committing, assume the plan has already failed, have everyone write down why, and then change the plan.

When an activity threatens serious harm, lack of full scientific certainty is no reason to delay measures to prevent it.

Precaution says do not act on uncertain harm; the two-way door says act fast when the move can be undone. The cost of undoing, and the part that cannot be undone, decide.

A decision is opposed not for its own consequences but for what it would commit the decider to next.

People state a public position they do not hold in order to avoid social cost, so public opinion can diverge sharply and invisibly from private opinion.

An evaluation of a person that follows from the category they are assigned to.

Do not tune for speed before measurement shows where the time goes; most code is not on the critical path, and optimised code is harder to change.

Knuth says do not optimise before measuring; Wirth says software slows faster than hardware speeds up. The number of calls, users and layers decides.

Anything we get now counts for extra, simply because it is now, against anything that comes later.

Attending work while ill, produced by the structure of cover and absence rules rather than by dedication.

What arrives first gets the firmest hold, and frames how everything after it is read.

What we have just met shapes how we read what comes next, and we never notice the hand-off.

When one party acts for another, cannot be fully observed and does not share its objectives, no contract removes the resulting loss; it can only be traded against risk.

Argue against the strongest version of what the other party said, including the parts they left out.

A system should behave the way its users expect; when a feature would surprise them, redesign the feature rather than retrain the users.

A two-player game in which defection pays better than cooperation whatever the other side does, leaving both worse off than if both had cooperated.

An innovation is assumed to benefit everyone, and non-adoption is treated as a failure to be overcome.

Material that costs more effort to process is sometimes better retained, but only certain kinds of effort.

We assume the tastes and wants we have today will still be ours later, and choose for a future self who does not exist.

The argument is made too long and too dense to be examined, and the difficulty of examining it is taken for strength.

An argument is shown to be defective by showing that its form would equally establish something the arguer rejects.

An explanation at one causal level is offered as refuting an explanation at another.

An outcome that is certain only if an earlier stage goes well is treated as certain outright.

We treat a category as having a hidden inner nature that explains what its members are like.

A threatened freedom is reasserted, and the blocked option becomes more attractive for being blocked.

Psychological safety is a belief shared within a team that speaking up with a question, a doubt or an admitted error will not be held against you.

Studies with positive results reach print; studies that found nothing stay in the drawer.

Where status comes from demonstrating commitment and nobody gains by arguing for less, the standard ratchets upward until ordinary members are disqualified by a standard they helped to set.