Finding a fallacy in someone else's argument takes about four seconds. Finding one in your own can take years, and usually takes a second person.
This is not a character flaw. It's how the machinery is built. You experience your own reasoning from the inside, as a sequence of conclusions that felt correct as they arrived. You experience everyone else's from the outside, as claims that either hold up or don't. Their reasoning comes with visible seams. Yours arrives pre-assembled.
Which means the standard advice to "think critically" is aimed at the wrong target. Nobody needs help being critical of arguments they already disagree with. The skill worth building is the inward one, and it doesn't run on vigilance. It runs on a small set of questions you can ask on purpose, because the failure will never announce itself.
Seven of them are below. None take longer than a minute.
Name the evidence that would change your mind
Ask it before anything else: what specific finding would make me abandon this position?
A good answer is concrete. "If the three-year data showed no effect, I'd drop it." "If the cost came in above X, this doesn't work." A bad answer is a shrug, or a list of conditions that could never occur, or the discovery that you can't think of anything at all.
If nothing would change your mind, you are not holding a conclusion. You're holding an identity, and no amount of evidence in either direction is going to touch it. That's worth knowing before you spend an evening arguing about it.
The question also works in the other direction, and it's an unusually generous move in a live discussion: ask the other person what would change theirs. Then answer it yourself first.
Run the mirror test on your own sources
The single most useful check for confirmation bias takes one sentence: if this study had come out the other way, how hard would I be looking for problems with it?
Everyone knows the feeling of reading a paper that confirms their view and thinking "well, that settles it," then reading one that contradicts it and immediately noticing the sample size, the funding source, the survey wording, and the fact that it's a preprint. Both instincts are available at all times. Bias is not what you find. Bias is which one you switch on.
The asymmetry is the tell. When you catch yourself scrutinizing one side of a question with tools you didn't apply to the other, you're not evaluating evidence. You're shopping for it. The correction is not to stop scrutinizing, it's to run the same audit on the source you liked.
Ask what you'd do starting from today
Sunk cost hides better than any other pattern on this list, because in the moment it doesn't feel like bad reasoning. It feels like commitment, loyalty, or seriousness.
The test is a single question, and it's cleanest when asked plainly: if I were arriving at this situation fresh today, with no history and nothing already spent, would I choose this?
Two years into a degree, eight months into a rewrite, four years into a relationship, three days into a book you're not enjoying. The money is gone, the time is gone, and neither is retrievable by continuing. The only live question is what the next unit of investment buys, compared to what else it could buy.
Sometimes the fresh-start answer is still yes. That's a real result, and it feels much better than continuing by default. The point of the test isn't to make you quit things. It's to make continuing a decision rather than a drift.
Look at the list you didn't make
Cherry picking is mostly invisible from the inside, because the evidence you gathered is right there and the evidence you skipped is, by construction, not.
The check is to reconstruct the search. How did I find these examples? Did I go looking for cases that fit, or did I collect everything in a category and see how it broke down? If you can recall the moment you stopped searching, ask why you stopped there. Usually the answer is that you had enough to make the point, which is exactly when a real audit should begin rather than end.
A practical version: write down the three strongest cases against your position before you write the ones for it. If you can't produce three, you don't yet know the other side well enough to be confident, and the confidence should come down until you do.
Match your confidence to your depth
Dunning-Kruger gets cited as an insult, which buries the useful part. The useful part is that early competence in a subject feels almost identical to real competence, because you can't see the terrain you haven't reached yet. The map looks complete when it's small.
The calibration check is not "how sure am I?" It's "how much of this have I actually touched?" Count the primary sources you've read rather than summaries. Count the hours. Count whether you can state the strongest objection to your view in a form the other side would accept.
Then look at the gap. Confident and shallow is the combination that produces the most confident wrong opinions in any room. Uncertain and deep is uncomfortable and almost always more accurate. If your certainty about a topic outruns your engagement with it, the certainty is doing something other than tracking truth.
Find the first number you heard
Anchoring is the easiest of these to catch, because it leaves a timestamp.
Whatever you're estimating, salary, house price, project timeline, reasonable rent, ask what the first number in your head was and where it came from. A listing price. An opening offer. A figure a colleague mentioned once. Then notice how much of your current estimate is a small adjustment away from that number, in the direction of your interests.
The repair is to build the estimate from the other end. What does the work actually cost? What do comparable places rent for? Produce your own number before you look at theirs, and write it down, because the anchor rewrites memory efficiently and you will otherwise remember having thought of it yourself.
Ask where the failures went
Any time you're reasoning from examples of things that worked, survivorship bias is on the table.
The founders who dropped out. The methods that produced champions. The old buildings that were built to last. In each case the visible sample was filtered by an outcome, and the filtered-out group is silent by definition. The dropouts who failed didn't write memoirs. The badly built old buildings fell down, which is why the surviving ones look like proof of superior craftsmanship.
One question does most of the work: how many people did this same thing, and what happened to the rest? If nobody counted, the pattern you're seeing might be a pattern in the counting.
The part that makes it stick
Running these tests on demand is the easy half. The hard half is remembering to run them, and the trigger is not intellectual. It's physical. The feeling of being certain and slightly annoyed, of already knowing where the conversation is going, of wanting to reply before finishing the paragraph, that feeling is the signal. It shows up reliably right before the reasoning stops.
Nobody catches all of these in themselves. The realistic goal is to catch one, occasionally, before it hardens into a position you have to defend in public. That's a low bar and a genuine improvement, and it's more than most people in any given argument are attempting.
The next time you feel the click of certainty arriving early, that's the moment. Ask what would change your mind, and see if you have an answer.