Everything the adaptive question bank can teach and test in Critical Thinking, from foundations through advanced practice. Work through it in order, or start practicing and let the questions find your level.
The tell is that removing the person leaves the argument untouched — if the claim still stands or falls on its own evidence, the attack was never doing work. Note the limit: where the claim IS the speaker's testimony, their reliability is legitimately on the table.
The damage is that the audience now believes the opponent said the weak thing. Ask what the opponent would recognize as their own position; if they would not sign the summary, the refutation lands on nobody, and the real claim has quietly escaped scrutiny.
It works because the new topic is usually more interesting than the old one, so the audience follows willingly and nobody notices the original question went unanswered. The check is mechanical: write down the question, then ask whether it has been answered.
A hypocrite can be right. Pointing out that the doctor smokes says nothing about whether smoking harms you, and treating it as a rebuttal hands an argument's fate to the biography of whoever happened to make it.
It converts a specific charge into a contest of comparative sins, which nobody can win and everybody can join. The other party's conduct may deserve examination, but it belongs in its own conversation and cannot answer this one.
Where an idea came from is evidence about its likelihood, not about its truth — a claim with a disreputable history can still be correct, and one from a trusted origin can still be wrong. Origin is a reason to check, never a reason to conclude.
The move happens BEFORE the argument, which is what makes it hard to answer: by the time the case is made, the audience is hearing it through a frame that was installed unopposed. Unlike ad hominem it pre-empts rather than responds.
Emotion is often the right response to a fact, so the defect is not feeling but substitution — the feeling arrives instead of the evidence rather than because of it. Test by asking what would remain if the story were true but the tone were flat.
The credential is usually real, which is what makes it persuasive; it just belongs to a different subject. A physicist on nutrition is a layperson in that room, and the fix is asking which field the question actually lives in.
Widespread belief is weak evidence at best and none at all where everyone shares the same source or the same incentive. Ask what the many are relying on: if they are all repeating one claim, the count is one, not millions.
The category does no work — arsenic and botulinum are natural, insulin and vaccines are not. Whether a substance helps or harms is a question about the substance, and the word smuggles a conclusion in as a description.
Longevity is weak evidence that something is not catastrophic, and no evidence that it is good or best. The useful version asks WHY the practice survived; an answer like habit, or a beneficiary who liked it, is the tell.
The mirror of tradition, and it fails for the same reason: age is not a property that makes a claim true. Newer can mean better-informed or merely untested, and only looking at what changed tells you which one you have.
It confuses what is the case with what we would like to be the case, so it reliably protects comfortable beliefs from evidence. A consequence is a reason to act, and never a reason to think a claim false.
The work is done by what is left out, so the repair is to name a third option aloud — once it exists in the room, the forced choice dissolves. Note that some dilemmas really are exhaustive; the question is always whether this one is.
The defect is the missing middle, not the shape: each link needs its own evidence, and an argument that asserts the chain without supplying them is prediction dressed as inference. Some slopes are real and have measured steps.
Splitting the difference is a negotiation tactic, not a truth-finding one. When one side is simply wrong the midpoint is also wrong, and the habit rewards whoever states the more extreme position first.
The argument is valid and useless at the same time, which is what makes it hard to spot: nothing false is ever said. Trace which claim is supporting which, and the conclusion turns up doing double duty as its own support.
Both answers concede the contested claim, so the only honest move is to refuse the frame and address the presupposition directly. Naming it — "that assumes something I have not agreed to" — is what breaks the trap.
One word carries two senses and the argument needs both, so each step looks fine while the whole does not follow. The test is substitution: write out the definition at each use, and the switch becomes visible immediately.
The two things really do share some feature, which is what makes the comparison feel fair; it is the feature that MATTERS on which they differ, usually in magnitude, evidence or intent. Name the respect being compared and the gap appears.
Every analogy breaks somewhere, so the question is never whether it breaks but whether it breaks at the joint carrying the argument. Ask which shared property is doing the inferential work, then check that both cases really have it.
It rescues a generalization by making it unfalsifiable: each counterexample is expelled from the category rather than counted against the claim. The tell is a definition that moves only when it is threatened.
A rule is applied everywhere but here, with no reason given that would not also excuse the other cases. The repair is to state the exception as a general principle and see whether anyone would accept its other consequences.
The evidence demanded arrives and a new demand replaces it, so no possible answer ever settles anything. Pin the standard down in advance — asking what evidence WOULD change the view turns an endless exchange into a testable one.
Two claims share a name: one bold and interesting, one modest and defensible. Challenge the bold one and the modest one is produced as though it were what was meant. Ask which version is being asserted, and hold the answer.
It exploits the asymmetry that asserting takes seconds and refuting takes minutes, so the responder always looks evasive. The counter is to refuse the pace: take one claim, settle it properly, and treat the rest as unaddressed.
The requests are individually reasonable and collectively a denial-of-service attack, which is what makes the politeness load-bearing — it makes any refusal look like the rudeness. Judge the pattern and the persistence, never one question.
Sequence is necessary for causation and nowhere near sufficient, and the mind supplies the connection for free. Before accepting it, ask what else changed at the same time and how often the effect occurs without the supposed cause.
Most outcomes have several contributing causes, so naming one is a choice about where to look. The tell is that the chosen cause usually matches what the speaker already wanted to change, and the others go unmentioned.
Two things can be wrong — the sample is too small, or it is the wrong sample — and they need different fixes. More of a biased sample does not help, which is why sample size alone is a poor reassurance.
A story is one data point with unusually good lighting: memorable, specific, and no more representative for it. It is legitimate for showing that something is POSSIBLE, and worthless for showing how often it happens.
The examples are real, which is the defense offered, and their selection is the whole problem: any large group contains extremes, so quoting them says nothing about the group. Ask what the typical member does, not the worst.
Properties travel from parts to whole only sometimes, and never for the ones that depend on arrangement or on totals. Every player being excellent does not make the team excellent, because the team has a property players do not.
The mirror of composition, and it fails the same way from the other end: an average or a total says nothing about any individual. A wealthy country contains poor people, and a fast team contains a slow member.
The pattern is found after the data are in, so its apparent improbability is fake — a search over enough boundaries will always locate a cluster. It is only evidence if the region was specified before you looked.
The real comparison is always against the alternative actually available, not against an imagined ideal. Framing the choice against perfection lets any improvement be dismissed while doing nothing quietly wins by default.
Vague boundaries are everywhere and destroy nothing: there is no exact grain where a heap begins, and heaps exist. The fix is to test at the clear cases, where the distinction obviously holds.
A description of how things are cannot by itself yield a conclusion about what should be done; some evaluative premise is required, and in these arguments it is present but unstated. Surfacing it is usually enough to start the real disagreement.
The reverse direction, and it corrupts inquiry rather than ethics: because a finding would be unwelcome, the world is assumed not to contain it. It is how uncomfortable evidence goes unexamined by people acting in good faith.
Desire is doing the work that evidence should, usually without the arguer noticing. A useful check is to ask what you would need to see to give up the belief — if nothing comes to mind, the belief is being held for its comfort.
A comparable wrong elsewhere neither justifies this one nor answers for it; at most it raises a separate question about consistency. It is persuasive because it feels like fairness, and it leaves the original act entirely unaddressed.
The limits of one person's imagination are a fact about that person. Difficulty picturing a process is evidence about familiarity and background, and it is offered as though it were evidence about the world.
Not knowing is a state of the inquiry, not a finding. Both directions fail — unproven is not false, and unrefuted is not true — and the honest position when nothing is established is to say the question is open.
Whoever asserts owes the support, because demanding disproof of an unsupported claim sets an impossible task and makes any claim survivable. The question that restores order is what evidence was offered FOR the claim.
Being dismissed is something correct and incorrect ideas both experience, and the incorrect ones far more often. The ridicule is not the evidence; whatever Galileo had that his critics lacked was, and this move never supplies it.
The claim cannot be false, so it rules nothing out and predicts nothing — its air of profundity comes from sounding substantive. Ask what observation it forbids; if none, it is a definition wearing the costume of a discovery.
A bad argument leaves the claim exactly where it was: unproven, not disproven. Treating a spotted fallacy as a refutation is how people talk themselves out of true conclusions that happened to arrive badly defended.
Fluency, confidence and polish are easy to perceive and largely independent of accuracy, so they dominate judgment in live debate. The countermeasure is to read the transcript, where delivery disappears and the content is all that is left.
An interest gives a reason to scrutinize a claim and never a reason to reject it — people with something to gain are sometimes right, and motive is available to invent for anyone. Check the evidence, then decide.
B. Cognitive biases
It operates at three points — what you look at, what you count as good evidence, and what you remember — so noticing it at one does not fix the others. The practical counter is to specify in advance what would count against you.
The pull survives knowing the number is arbitrary and survives being warned, which is what makes it a bias rather than a mistake. The defense is procedural: form your estimate before the number arrives, or generate a counter-anchor deliberately.
Ease of recall tracks vividness, recency and coverage rather than frequency, so the events that feel common are the ones that were reported. Anything newsworthy is by definition rare, which is precisely why it was reported.
The judgment runs on similarity and ignores how many of each kind exist, so a description matching a rare type beats the arithmetic. The repair is to ask how common each category is before asking which the description fits.
Once you know the outcome, the path to it looks inevitable and the alternatives become hard to reconstruct. The effect is that past uncertainty disappears from memory, so decisions get graded against knowledge nobody had at the time.
It punishes good decisions that ended badly and rewards reckless ones that happened to work, which teaches exactly the wrong lesson. Grade the process: what was known, what alternatives existed, what the odds looked like then.
The mechanism is that the skill needed to do a thing well is largely the same skill needed to tell good work from bad, so the least able have the least equipment to notice. Confidence is therefore weakest evidence exactly where it feels strongest.
Ask people for a range they are ninety percent sure contains the answer and the truth falls outside it far more often than one time in ten. Widening ranges until they feel uncomfortably large is the calibration exercise.
Most people rate themselves less likely than average to suffer common misfortunes, which cannot be true of most people. It is why warnings feel addressed to someone else, and why protective measures get postponed rather than rejected.
One vivid trait sets the tone and the rest of the assessment falls into line with it, so attractiveness colors judgments of competence and a single error colors everything after. Rate each dimension separately, ideally at separate times.
You see your own situation and other people's behavior, so the explanations are built from different materials. The asymmetry vanishes when you know the context, which is why it weakens with strangers only.
It protects self-image at the cost of learning: if the wins were skill and the losses were luck, no feedback about skill ever arrives. The tell is an explanation whose direction always flatters, whatever the outcome.
The need for the world to be fair converts misfortune into evidence of fault, which is why victims get scrutinized for what they did wrong. It is comforting precisely because it implies that behaving well keeps you safe.
You are the constant of your own experience and a passing detail of everyone else's, so the felt visibility of a stumble or an outfit runs far ahead of the observed one. Others are attending to their own spotlight.
Your internal state is loud from the inside and nearly silent from the outside, so nerves that feel obvious go unnoticed and hints that feel broad land as nothing at all. Say the thing rather than assuming it was read.
Once something is understood, the state of not understanding it becomes unreachable, so explanations skip the steps that were hardest to climb. It is why experts write bad introductions and why testing on a real novice is irreplaceable.
Your sample of other people is drawn from people like you, so the estimate is built from a biased frame that feels like the world. It makes disagreement read as unreasonableness rather than as ordinary variation.
Negative information is weighted more heavily, processed more thoroughly, and remembered longer than positive information of the same size. The consequence is that a balanced account reads as an attack, and a single flaw outweighs several merits.
Ninety percent survival and ten percent mortality are the same fact and reliably produce different choices, so whoever chooses the wording has influenced the decision. Restate the option the other way before deciding.
Position in a list changes recall independently of content, so the middle of an agenda, a shortlist or a presentation is structurally disadvantaged. If order is arbitrary, vary it; if it is not, know which items the order is helping.
Real randomness is streakier than intuition expects, so genuinely random data looks patterned and evenly-spread data looks random. The test is whether the pattern was predicted in advance or noticed afterwards.
In a process where trials are independent, a run carries no information about the next trial. Note the live exception: where skill, confidence or fatigue are involved the trials may NOT be independent, and then a streak can be real.
The mirror of the hot hand, from the same false premise: the coin has no memory and nothing is owed. Believing a correction is due is what makes a losing run feel like a reason to increase the stake.
A test with excellent accuracy applied to a rare condition still produces mostly false positives, because the healthy majority is so much larger. The intuition fails hard here, so the arithmetic has to be done rather than felt.
The cases that failed are absent from the data by construction, so the surviving sample supports conclusions the full population would not. Ask deliberately where the missing cases went, and whether their absence is related to what you are measuring.
The current state becomes the reference against which everything else is a loss, so change needs to clear a bar that staying does not. Testing the reversal — would you switch back? — exposes whether the preference is real.
The asymmetry means an even-odds bet with equal stakes feels bad, which is rational for large stakes and costly for small repeated ones. Whether it is an error depends entirely on whether the bet is one-off or repeated.
The price to sell exceeds the price to buy for the identical object, so ownership itself moves the valuation. It explains why trial periods convert so well and why clearing possessions is harder than acquiring them.
Effort invested raises the value you place on the result, independently of the result's quality. It is why self-built plans get defended past their merits, and why involving people in making something makes them keep it.
Money and time already spent are unrecoverable whatever you do next, so they are irrelevant to the choice and reliably dominate it anyway. The clarifying question is what you would choose arriving fresh, with the same options and none of the history.
People predict their own projects from the specific plan and other people's from the track record, and the track record is the accurate one. The fix is to look up how long similar projects actually took, not to add a margin.
The discount between now and soon is far steeper than between two equally distant future points, so preferences reverse as the moment arrives. That reversal is the signature, and the reason commitment devices work.
Stated value barely rises between saving a few and saving many, because the response is driven by the mental image rather than the number. It makes intuition useless for comparing the SIZE of two goods.
Attention goes to what stands out, and what stands out is a property of presentation rather than importance. The consequence is that the loudest variable gets managed while the quiet one that actually drives the outcome is left alone.
Statements that are easy to read, rhyme, or have been heard before feel truer, and repetition alone raises belief. That is why a rebuttal that repeats the false claim can strengthen it — lead with the correction, not the myth.
The descriptions work because they are true of nearly everyone and phrased to feel specific, usually by pairing a trait with its opposite. The test is whether someone else's reading would fit you equally well.
Judgment is made against whatever preceded it, so the same option looks good after a poor one and mediocre after a strong one. Sequence therefore becomes a lever, which is why a decoy option is placed in the set at all.
A correction removes the claim without removing its effect on inference, because the original has already been woven into a causal story. Supplying a replacement explanation works where a bare retraction does not.
Beliefs acquire supporting reasons after adoption, so removing the one that started it leaves the structure standing. It explains why refuting the founding claim so rarely settles anything.
Holding an action and a belief that do not fit is uncomfortable, and the cheaper repair is usually to move the belief. That is why a difficult initiation raises commitment rather than lowering it.
Bias is inferred in others from behavior and in oneself from introspection, and introspection cannot see it — biased reasoning feels like reasoning. Assume it applies to you at the usual rate and build procedures rather than vigilance.
The standard of proof moves with the conclusion: welcome findings get "can I believe this", unwelcome ones get "must I believe this". Catch it by asking whether you would accept the same quality of evidence for the opposite result.
After a costly irreversible choice, the reasons for it multiply and the drawbacks fade, which protects the decision from honest review. It is why testimonials from buyers are weak evidence about the thing bought.
The fast system answers everything and the slow one only engages when something signals difficulty, so errors survive wherever a hard question feels easy. Deliberately slowing down is what the model actually prescribes.
C. Evidence, statistics, and scientific reasoning
Four live possibilities remain when two things move together: the first causes the second, the second causes the first, something causes both, or the pairing is chance. Ruling out three is the entire work, and the slogan only names the problem.
It produces a real association with no causal link between the two variables measured, which is why the correlation is genuine and the conclusion is still wrong. Adjusting for it requires having measured it, so the dangerous ones are those nobody recorded.
Plausibility is no guide, because both directions usually sound plausible once stated. Timing is the discriminator, and it is often unavailable in the data — which is exactly when the more comfortable direction gets assumed.
Compare enough series and strong correlations appear with no relationship whatever, because the search space is vast. A correlation found by looking needs a prediction to earn credence; one predicted in advance already has some.
Without one you cannot tell the treatment's effect from what would have happened anyway — recovery, seasonality, or the tendency of extremes to moderate. Before-and-after alone measures the passage of time as much as the intervention.
The response is to the ritual of being treated, and it is real enough to register on subjective outcomes, so any uncontrolled trial will find something. Its size varies by outcome type, which is why some endpoints need it controlled more than others.
Subjects blinded alone leaves the assessor able to nudge the measurement, especially on judgment-based outcomes. Both arms of the blinding exist because expectation moves what is felt and what is recorded, and they are separate leaks.
Randomizing balances the causes nobody thought to measure; statistical adjustment can only balance the ones that were. That is the whole difference, and it is why observational studies can be large, careful and still confounded.
Something outside the system — a policy boundary, a lottery, a weather event — assigns the exposure in a way unrelated to the outcome. The strength of the inference rests entirely on that assignment really being unrelated.
It balances known and unknown factors in expectation, not in any particular trial, so a small randomized study can still be unbalanced by luck. Larger samples make the balance tighter; randomizing does not guarantee it.
Small samples produce more extreme results in BOTH directions, so the best and worst performers on any small-sample measure tend to be small. Noticing a record proportion in a tiny group is often noticing the group is tiny.
The frame decides the population the result describes, and it is usually narrower than the conclusion drawn. Ask who could not have been sampled at all, and whether their absence is related to the outcome being measured.
The distortion enters at the door, before any analysis, so a flawless method on a selected sample produces a confident wrong answer. It is distinct from a small sample: more data collected the same way makes it worse, not better.
Whoever answers differs from whoever does not in ways usually connected to the topic, so a low response rate is a warning about direction rather than about precision. A large sample with heavy non-response is not reassurance.
Anything measured with error is partly luck, so the extremes contain more luck and the next measurement drifts back. It manufactures the appearance that punishing poor performance works and praising good performance backfires.
In a skewed distribution the mean is dragged toward the tail while the median stays with the bulk, so the choice of summary is a rhetorical decision. Income and wait times are the standard cases; asking for both is the defense.
Two options with identical averages can differ entirely in what a bad case looks like, and the bad case is often what the decision turns on. A summary reporting only a center has discarded the information you need.
A doubling means a lot when the baseline is common and almost nothing when it is rare. Relative figures are the headline because they are larger; ask what the underlying rate was and convert to people affected.
How common the condition is determines what a positive result means, and no property of the test can substitute. Two people with the same result and different prior likelihoods are in genuinely different situations.
The chance of the evidence given the hypothesis and the chance of the hypothesis given the evidence are different quantities, and they diverge sharply when the hypothesis is rare. Swapping them is the single most common probability error in public argument.
How often the evidence occurs by chance in the population is not how likely it is the accused is uninvolved; the size of the population searched is what converts one into the other. A one-in-a-million match in a city of millions is several people.
Pooling groups of different sizes and different baseline rates can reverse the direction of an effect, so both the aggregate and the subgroups can be computed correctly and disagree. Which one to act on depends on the causal story, not on the arithmetic.
A correlation between group averages need not hold, or even point the same way, inside the groups. It is why an area-level association is weak evidence about the individuals living there.
Significance answers whether an effect is distinguishable from zero, which depends heavily on sample size; a large enough study makes trivial effects detectable. The decision needs the magnitude, and that is a separate number.
Choices made after seeing the data — which outcome, which subgroup, which exclusions — multiply the chances of a result without any being obviously dishonest. The signature is a finding whose specification looks unusually particular.
Run enough independent tests at a fixed threshold and some pass by chance alone, so a single striking result from a large battery means little on its own. Report how many comparisons were made, or the reader cannot judge.
The width carries the information the point estimate hides: an interval spanning trivial and enormous effects is compatible with both. An interval that includes zero is not a finding of no effect, only of insufficient evidence.
A theory compatible with every possible observation has told you nothing about which world you are in. Ask what result its proponent would accept as refuting it; if the answer is none, its explanatory power is illusory.
A first result is a candidate finding, because the ways a single study can mislead are numerous and often invisible from the paper. Independent replication by a different team is what converts it, and many published effects have failed that step.
Studies that found nothing are less likely to be submitted, accepted or noticed, so the visible literature overstates effects. It means a body of positive findings is consistent with there being nothing there, if the null studies stayed unpublished.
It is a tie-breaker between explanations that fit the evidence equally well, not a reason to prefer the simplest story in general. Reality contains complicated things, and the rule is about unsupported additions, not complexity itself.
Once a proxy carries consequences, effort flows to the proxy rather than to the thing it stood for, and the correlation that justified it decays. Expect it, and watch what the measure has stopped tracking rather than whether it improved.
Starting a scale above zero magnifies whatever variation exists, so a difference of a few percent can be made to look like a collapse. Check where the axis begins before reading the shape of the line.
Encoding a single number as an area squares the visual difference; two axes on one chart can manufacture any apparent relationship; a chosen window can show the opposite trend from the full series. Read the axes and the date range first.
Digits beyond what the measurement can resolve are decoration that reads as rigor, and they hide the real uncertainty. A figure quoted to three decimal places from a survey of a few hundred people is claiming precision nothing supplied.
A claim conflicting with a large body of established results starts improbable, so the evidence has to overcome that starting point as well as support the claim. It is not a demand for a different standard, just the ordinary one applied to a low prior.
It assigns the burden rather than settling the question: an unsupported claim need not be refuted, and dismissing it is not the same as showing it false. Useful for declining an endless task, useless as a conclusion.
Incompetence, haste and misaligned incentives are far more common than coordinated intent, so they are the better default. It is a prior and not a rule — evidence of intent overrides it, and it should not be used to dismiss a pattern.
The inference is only as strong as the search: failing to find something you would certainly have found is informative, and failing to find something nobody looked for is not. Ask what the detection probability was.
If more exposure produces more effect, chance and many confounders become harder to sustain as explanations. Its absence is weaker news than its presence, since genuine relationships can have thresholds and plateaus.
A cause cannot follow its effect, so the timeline can refute a causal claim outright — which makes it the cheapest check available. Be careful that the date used is when the condition began, not when it was recorded.
A story of how something could work makes a hypothesis worth testing and settles nothing by itself, since plausible mechanisms exist for effects that turn out to be absent. Treat it as a reason to look, not as a finding.
Where the outcome requires judgment — reading a scan, scoring a behavior — the assessor's expectation leaks into the number. Blinding the measurement is a separate safeguard from blinding the participant and is skipped more often.
Decisions consume magnitudes: how much longer, how many fewer, at what cost. Significance says only that the effect is probably not zero, which is compatible with its being far too small to justify the intervention.
Fixing the hypothesis, the outcome and the analysis before seeing data removes the freedom that turns noise into findings. It also makes the null result publishable, which is the part that fixes the literature rather than one study.
D. Argument structure and logic
Indicator words help — since, because, given that mark support; therefore, so, hence mark what is supported — but plenty of arguments use none. The reliable test is to ask which claim the speaker would defend if the other were challenged.
Most real arguments leave out the premise that would be most contentious, often because it seemed too obvious to state. Find it by asking what would have to be true for the stated premises to actually deliver the conclusion.
Validity is about structure alone and is compatible with every premise being false; soundness adds that the premises are true. Calling an argument valid is therefore not a compliment about its content, and the two words are not interchangeable.
Correct reasoning transmits truth but does not create it, so garbage premises give garbage conclusions through an impeccable process. It is why checking the form is only half the job and the premises need their own audit.
If the form is valid and the premises hold, the conclusion cannot fail — which is also the limit: it extracts what the premises already contained rather than adding knowledge about the world.
The conclusion goes beyond the evidence, so it can be well supported and still false, and no number of confirming cases closes that gap. Strength comes from sample size, diversity and the absence of a plausible alternative, never from certainty.
It picks the hypothesis that would best explain what is observed, which is how diagnosis and detection work. The weakness is that best-of-the-ones-considered is not best overall, so the candidate list is where the reasoning usually fails.
Required but possibly not enough. Oxygen is necessary for fire and supplying it starts nothing, so demonstrating a necessary condition holds establishes very little about the outcome.
Enough but possibly not required, so its absence proves nothing — there may be other routes to the same result. The two are routinely swapped in everyday speech, and each swap reverses what can be inferred.
A single genuine case defeats a universal claim outright, which is why "all" and "never" are expensive words. The counterexample has to satisfy the claim's own terms, which is where most attempted refutations actually fail.
A universal claim is refuted by one case and confirmed by none; an existential claim is confirmed by one case and refuted only by exhausting the field. Knowing which is being made tells you what work each side owes.
Chaining categories works when the middle term is used in the same sense and covers the whole group both times. The usual break is a middle term that is undistributed — "some" on both sides — which licenses nothing.
The conditional plus its antecedent yields the consequent, and the form is valid regardless of subject matter. The soundness question moves entirely to whether the conditional itself is true, which is where these arguments actually stand or fall.
Denying the consequent denies the antecedent, which is the engine of falsification: a prediction that fails takes the hypothesis with it. It is harder to see intuitively than its affirming counterpart and is valid all the same.
The consequent can be produced by other causes, so finding it tells you the antecedent is possible rather than actual. Wet ground does not establish rain; it is consistent with rain, which feels like the same thing and is not.
The conditional promised what happens if P holds and said nothing about what happens otherwise, so its failure leaves the consequent open. It is the error of reading an if as an only-if, which ordinary speech encourages.
Flip and negate both parts and the result is equivalent, always — so anything proved from one is proved from the other. It is the legitimate transformation next to the two that are not, and the reason modus tollens works.
Swapping the parts, or negating both without swapping, produces a different claim that may be true or false independently. Both errors are invited by everyday phrasing, where an if is often meant as an if-and-only-if.
Valid when the disjunction is genuine and exhaustive, which is the premise worth attacking: a false dilemma feeds this form correctly-shaped garbage. Check whether the two options really are the only ones.
Adding detail can only narrow the set of ways to be true, so a fuller story is always less likely than any part of it. Vivid specifics make the opposite feel true, which is why the fallacy is so easy to elicit.
"All that glitters is not gold" can deny that everything glittering is gold, or assert that nothing glittering is — two different claims from one sentence. Rewrite with the quantifier or negation in a fixed position before disputing it.
Improve the opposing argument until its holder would endorse the improvement, then answer that. It costs nothing if you are right and saves a great deal if you are not, since a refuted weak version leaves the real disagreement standing.
Where a claim admits several readings, take the strongest one the words allow. It is not generosity but self-interest: attacking the weakest reading produces a win that persuades nobody and teaches you nothing.
An argument gives reasons to believe something in dispute; an explanation gives an account of something already accepted. Confusing them misdirects the criticism — demanding proof of a premise in an explanation attacks the wrong thing.
E. Decision-making and rationality
Multiply each outcome by its probability and add, which is the right summary for decisions you will face many times. It is silent about spread, so a positive expectation can hide a distribution you could not survive.
Once a loss removes your ability to keep playing, the sequence of outcomes matters and averages stop applying. A bet with excellent expectation and a small chance of ruin is still a bad bet if repeated.
The real cost of a choice is the best thing it displaced, which never appears on any invoice. That invisibility is why cheap-looking commitments accumulate, each individually justified against zero rather than against the alternative.
First-order effects are visible and intended; second-order effects arrive later, often through how people adapt, and often reverse the first. Asking "and then what?" twice is the whole technique.
Stating the failure as already having happened frees people to name risks that sound disloyal as predictions. It reliably surfaces concerns a risk register does not, because the framing removes the burden of arguing against the plan.
Assigning the role makes dissent structural rather than personal, which is what makes it sayable. The weakness is that a role played half-heartedly produces token objections, so the case has to be genuinely prepared.
The size of the revision depends on how much more likely the evidence is under one hypothesis than the other, not on how striking it feels. Evidence equally likely either way is no evidence, however dramatic.
Holding a number rather than a verdict lets you be updated by degrees instead of switching sides, and makes you accountable in a way "I think so" is not. It also makes calibration measurable over time.
A decision you can undo cheaply should be made fast and tested; one you cannot deserves the delay. The common error is spending the same deliberation on both, which is slow where speed is free and hasty where it is not.
You are choosing now on behalf of a future self with different preferences, so the mechanism has to survive that self's attempt to undo it. Removing the option beats resolving to resist it.
Writing the reasoning, the alternatives and the expected outcome at the time creates a record hindsight cannot rewrite. Without it you are grading past decisions with a memory that has quietly updated itself.
Ask how similar cases turned out before reasoning about why this one differs, because the specifics that feel decisive are available to everyone who was also wrong. Start from the base rate and adjust, rather than building up from detail.
Spreading across outcomes that are not perfectly correlated cuts variance while keeping the average. The correlation is the whole point — things that fail together provide no diversification however many of them you hold.
Set a threshold and take the first option clearing it, or search for the best available. Maximizing costs more time and more regret, and only pays where the differences between good options are large.
Past a point, more deliberation costs more than the improvement it buys, and the delay itself has a price. Setting the stopping rule in advance — a deadline, a threshold — is the practical defense.
Project forward and ask what you will wish you had done, which reweights toward omissions and away from short-term discomfort. It is a useful corrective for a specific distortion rather than a general decision rule.
Multiplying a tiny probability by an enormous payoff produces a large number, so naive expected value can be captured by anyone willing to name a big enough prize. The defense is skepticism about unverifiable probabilities, not about the arithmetic.
Averaging cancels errors only when they point in different directions, so independence is the load-bearing condition. Let a crowd see each other's answers and it converges on a shared mistake with extra confidence.
Shortcuts are correct answers to the question of what to do with limited time, and they fail in known places. Knowing which accuracy you traded away tells you when the shortcut stops being appropriate.
When the downside is catastrophic and the upside merely good, the right policy optimizes the tail rather than the average. Symmetric reasoning applied to asymmetric stakes is how survivable positions become unsurvivable ones.
Outcomes are decisions plus luck, so a single result is a poor readout of the reasoning that produced it. Separating them is what makes learning from a small number of high-stakes decisions possible at all.
F. Media and source literacy
Judging a source from its own pages rewards design and self-description, both of which are cheap. Opening other tabs to see who runs it and what others say takes a minute and is what professional fact-checkers actually do.
Each retelling drops caveats, converts tentative findings into firm ones and loses the sample's limits. Going to the original is the only way to see what was actually claimed, and it is usually one link away.
A funder's interest does not make a finding false, and it tells you where to look hardest — at outcome choice, comparator and what went unreported. Disclosure is a prompt for scrutiny, not a verdict.
Reviewers check that the work is competent and plausible, usually without the data and never by repeating the experiment. It is a filter against obvious error rather than a certificate of truth, and plenty of reviewed work fails to replicate.
Presenting two positions as evenly matched when the evidence is lopsided misinforms through structure rather than through any false statement. The fix is to convey where the weight of evidence sits, not to silence a side.
Every word can be accurate while the meaning is inverted, usually by removing a condition, a hypothetical, or the sentence that followed. Finding the surrounding passage is the only check, and it is why sources link to originals.
A claim's first appearance often reveals a satirical origin, a misread study or an account with a history. Tracing back is usually quick, and it settles questions that arguing about the claim itself never will.
Three separate questions: were they positioned to know, are they able to judge it, and what would they gain. A witness can be entirely honest and still wrong on the first two, which is the case sincerity cannot cover.
Corroboration counts only when the confirmations have separate origins, and most do not — syndication, aggregation and quoting the same press release produce many headlines from one claim. Trace each back before counting it.
Genuinely independent confirmation is powerful because the chance of separate sources erring identically is small. That power rests entirely on the independence, which is the thing to verify before treating agreement as weight.
Reading primary literature outside your field without the background to judge method mostly amounts to selecting which experts to agree with. Done well it means finding where experts converge and where they genuinely disagree, not replacing them.
A finding can be correctly reported, widely repeated and superseded years ago, and nothing marks the old version as stale. Check the date, then check whether later work revised it — the second step is the one usually skipped.
Headlines are frequently written by someone other than the reporter and optimized for attention, so the article often contains the caveat the headline dropped. Most people share on the headline alone, which is how the distortion spreads.
Expertise is specific and narrow, and eminence travels further than competence does. A distinguished scientist speaking outside their field carries the authority of the credential and none of the knowledge it certifies.
Anonymity protects a source and removes the reader's ability to weigh them, so the outlet's own reliability becomes the whole guarantee. Useful reporting says why the source would know and why they were granted anonymity.
Most deception needs no synthesis: a genuine image from another event, a clip cut before the qualifying sentence, a crop that removes the context. Ask when and where it was taken before asking whether it was generated.
Keentune is not affiliated with or endorsed by the organizations whose documentation informs these maps.
All exam, test, and product names and trademarks are the property of their respective owners and are used here for identification and reference only. Keentune is independent study practice — not affiliated with, authorized, or endorsed by any of these organizations.