Critical thinking is usually taught as a list of things to avoid, which is why most people finish the list and reason exactly as before. It is better understood as three separable questions asked in order. Is this argument shaped correctly — do the premises, if true, actually support the conclusion? Are the premises true — and how would I find out? And am I the thing distorting the answer? Chapters D and A are the first question, C and F the second, B and E the third. The order matters: a beautifully evidenced argument with a broken shape proves nothing, and a well-shaped argument built on a rumor proves nothing either. Read for the three questions; the named terms are just labels for where each one fails.
Each chapter opens with the short version. Tap one to read the detail.
Fallacies are relevance failures, not rudeness
~2 min
A fallacy is not a rude argument or a wrong conclusion — it is a move where the support offered does not bear on the claim. Naming one is only useful if you can say which link is missing, and the same sentence can be a fallacy in one context and a fair point in another.
The single most common mistake with fallacies is treating them as a list of forbidden phrases. They are not. Each one describes a specific broken connection between what was offered and what was claimed.
Take the attack on the arguer rather than the argument. "You would say that, you work for the industry" is fallacious as a refutation — the funding does not make the data wrong. But raised as a reason to check the data more carefully, it is not a fallacy at all; it is ordinary diligence about incentives. Same sentence, two jobs, and only one of them is broken.
That is why fallacy-naming so often makes an argument worse. Someone shouts "straw man" and the discussion becomes about the label instead of the claim. The useful move is to state the missing link: you answered a weaker version — here is the version I actually argued. Now the other person can respond to something.
Watch particularly for the ones that feel like arguments: changing the subject to an unrelated charge, discrediting a source before it has said anything, or substituting how a conclusion feels for whether it follows.
Rule: never name a fallacy without also saying, in one sentence, what the argument would need in order to work.
Biases are systematic, and knowing about them barely helps
~2 min
Cognitive biases are not random errors or stupidity — they are predictable, directional distortions produced by fast mental shortcuts that are usually efficient. Crucially, knowing a bias by name does almost nothing to stop it, which is why the remedies are procedural rather than mental.
The word "bias" suggests carelessness. The research points somewhere less comfortable: these are outputs of shortcuts that work well enough most of the time, running automatically, in people who know exactly what they are.
They are also directional, which is what makes them dangerous and useful. An arbitrary number you have just seen drags your later estimate toward it, and it drags it in a knowable direction. Ask someone whether a city's population is above or below 40,000, then ask them to estimate it, and their answer lands lower than if you had said four million first. They will deny the first number mattered.
The uncomfortable finding is that awareness is weak medicine. You cannot introspect your way out, because the distortion happens before the thought you can inspect arrives. What works is structure imposed in advance: write the estimate down before seeing anyone else's, decide the criteria before seeing the candidates, record what you predicted before you learn the outcome.
That last one matters most. Once you know how something turned out, your memory of how confident you were quietly revises itself upward, so you never accumulate evidence that your judgment needs work.
Rule: if a debiasing technique happens inside your head at the moment of judging, it will not work — put it on paper, earlier.
Every evidence claim is a hidden comparison
~2 min
Statistics in arguments almost never fail at the arithmetic. They fail at the comparison: what was this group compared against, and could that comparison group differ in some other way that explains the result?
When someone reports that a treatment, policy or habit works, they are always making a comparison — usually one they have not stated. Recovering it is most of the work.
Consider: "people who take this supplement report fewer colds." Compared with whom? If the comparison is people who do not take it, the two groups differ in more than the supplement. People who buy supplements also tend to sleep, exercise and visit doctors differently. Any of those could produce the result. That third factor driving both things is a confounder, and it is the default explanation, not an exotic one.
This is why deliberate assignment matters so much. If you randomly decide who gets the treatment, the groups are on average alike in every respect including ones nobody thought to measure — that is the whole point, and no amount of statistical adjustment on observational data reproduces it. Adjustment can only correct for variables you knew to record.
Two more comparisons hide routinely. A relative change with no baseline ("doubles your risk") is uninterpretable until you know the risk doubled from what. And a striking correlation found by searching many possible pairings is expected to appear by chance, which is why what was looked for should be declared before looking.
Rule: before evaluating any effect, say out loud what it was compared with — if you cannot, there is no finding yet.
Separate the shape of an argument from its contents
~2 min
An argument's form and its truth are independent, and confusing them causes most bad reasoning about reasoning. A perfectly valid argument can reach a false conclusion, and a true conclusion can arrive through a broken one.
Start by finding which sentence is being supported and which are doing the supporting. People state conclusions first, last, or by implication, so this is genuinely a step and not a formality.
Then check the shape independently of whether you believe anything. If the premises were true, would the conclusion have to follow? That is validity, and it is a property of arrangement alone. Soundness is validity plus premises that are actually true — two separate tests, and skipping the first is how a confident argument built on agreeable premises slips through.
The consequence people find counterintuitive: a valid argument can hand you a false conclusion, quite legitimately, when a premise is false. "All birds can fly; a penguin is a bird; therefore a penguin can fly" is impeccably arranged and wrong. Nothing is broken about the logic. The first premise is false.
Most real arguments also lean on a premise nobody said. "She has ten years of experience, so she should lead the project" needs the unstated claim that experience is what leading requires. Surfacing that hidden premise is usually more productive than disputing anything that was actually stated, because it is where the disagreement really lives.
Rule: name the conclusion, list the premises, then ask what else must be true for the jump to work — that missing sentence is the argument.
Judge decisions by what was knowable, not by how they turned out
~2 min
A good decision and a good outcome are different things, and conflating them destroys your ability to learn. Under uncertainty the only fair standard is the information and reasoning available at the time.
Someone drives home drunk and arrives safely. The outcome was fine; the decision was terrible. Everyone accepts this example and then judges their own past choices by results anyway — a habit that teaches the wrong lesson twice, punishing sound calls that lost and rewarding reckless ones that happened to land.
The repair is to evaluate the process: what did you know, what did you expect, and was the reasoning defensible then? This is only possible if you wrote it down beforehand, which is the practical link back to memory's quiet revisions.
Two further ideas do most of the remaining work. First, an average is not the whole story: a bet with a genuinely positive expected return can still be one you must not take, if a plausible outcome removes you from the game entirely. Survival is a constraint, not a term in the average. Second, most choices are not yes-or-no but instead-of. The real cost of a course of action is the best thing you gave up to take it, and that alternative is usually unnamed.
A cheap technique for both: before committing, assume the plan has already failed badly, and write the story of how. Stated as a certainty rather than a risk, objections arrive that politeness would otherwise suppress.
Rule: record the reasoning before the result arrives, and grade the reasoning.
Go sideways, not down
~2 min
Faced with an unfamiliar source, the instinct is to read it closely and judge it on how it looks. Professional fact-checkers do the opposite: they leave the page almost immediately and find out what independent sources say about it.
Reading a suspicious page carefully is a trap, because a page's appearance is entirely under its author's control. Design, citations, and a confident tone are cheap. What is not cheap is a reputation across sources that do not share an interest.
So the move is lateral: open other tabs and ask who runs this, who funds it, and what people with no stake say about it. This takes under a minute and settles most questions that close reading cannot.
Second, trace claims to their origin. A striking finding usually reaches you third-hand — a post about an article about a study — and each hop drops caveats and sharpens the number. The study itself often reports something more limited than the headline. A report about a study is not the study.
Third, calibrate what credentials buy. Independent review before publication screens for competence and method, not for truth; reviewed work is retracted regularly. And expertise is domain-bound — being genuinely eminent in one field carries no weight in another, though it is usually presented as if it does.
Finally, watch for equal airtime given to unequal evidence. Presenting two sides as balanced misrepresents the state of knowledge whenever one side is thinly supported.
Rule: before judging a source, spend sixty seconds finding out what other sources say about it.
Written by Keentune. We are not affiliated with or endorsed by the organizations whose documentation informs this guide, and any linked sources belong to their respective owners.
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.