YOUR BRAIN WAS DESIGNED TO LOSE. ONE EQUATION FIXES IT

In 1973, Daniel Kahneman and Amos Tversky handed participants a personality sketch of a man named Jack - 45 years old, married, four children, conservative, careful, enjoys carpentry, sailing, and mathematical puzzles. Then they told one group that Jack was drawn from a pool of 70 engineers and 30 lawyers. They told the other group the reverse: 30 engineers and 70 lawyers.

Both groups gave essentially the same answer. They estimated the probability that Jack was an engineer at roughly the same figure, regardless of whether the room was 70% engineers or 30%.

The base rate - the single most powerful number available - was ignored almost entirely. Participants heard a story, decided it sounded like an engineer, and stopped thinking. The actual composition of the room did not register.

Twenty-six years later, in 1999, Justin Kruger and David Dunning ran a different experiment. Participants took tests on logic, grammar, and humour, then estimated how well they had done. Those who scored in the bottom quartile - the 12th percentile on average - estimated themselves to be at the 62nd. Those at the top - above the 86th percentile - estimated themselves at roughly the 70th.

The worst performers overshot by fifty percentile points. The best performers undershot by sixteen. And neither group knew.

These are not curiosities. They are two halves of the same structural flaw in human reasoning, and together they explain more about why confident people are wrong, why prestigious advice is unreliable, and why the most important skill in any decision is not confidence but calibration - than any other pair of findings in cognitive science.

This article connects them, adds the mathematics underneath, and arrives at a solution that has been quietly working for two decades.

Save this one. It is the article the rest of the feed exists to illustrate.

The Problem, Stated Precisely

The human mind evaluates claims by story, not by frequency. It asks "does this sound right?" rather than "how often is this actually true?" That substitution has a name - the representativeness heuristic - and it is not a rare mistake. It is the default operating mode.

Every confident opinion you have ever heard works the same way Jack's description works. Someone tells you a story. The story sounds like a pattern you recognise. You assign a high probability. The base rate - how often this outcome actually occurs in the world, across all the cases you are not seeing - does not enter the picture, because it is not a story. It is a number. Numbers do not feel like anything, so the brain skips them.

The Dunning-Kruger effect is the version of this same failure applied to yourself. You cannot evaluate your own skill in an area you do not understand, because evaluation requires the very skill you are missing. The worst performers in Kruger and Dunning's study did not lack confidence. They lacked the metacognitive tools to see the gap between what they knew and what they thought they knew. That gap averaged fifty percentile points.

Two findings. One mechanism: a reasoning system that substitutes narrative fit for frequency, and has no internal alarm when it gets the substitution wrong.

The mathematical version of the solution has existed since 1763. It is one equation, and almost nobody uses it.

The Equation the Brain Refuses to Run

Thomas Bayes published his theorem posthumously in 1763. It is one line, and it answers one question: given new evidence, how should you update what you already believed?

$$P(H|E) = P(E|H) × P(H) / P(E)$$

In words: the probability of your hypothesis given the evidence equals the probability of seeing that evidence if your hypothesis were true, multiplied by the prior probability of the hypothesis, divided by the total probability of seeing the evidence at all.

The term that people skip - every time, across every study, in every formulation - is P(H), the prior. The base rate. The answer to the question "how likely was this before I heard the story?"

Here is what that looks like with real numbers.

A medical test is 99% accurate. The disease it tests for affects 1 in 10,000 people. You test positive. What is the probability that you actually have the disease?

The intuitive answer, the one most people give including a substantial fraction of physicians, is "very high, probably 99%."

The correct answer:

P(disease | positive test)
= (0.99 × 0.0001) / ((0.99 × 0.0001) + (0.01 × 0.9999))
= 0.000099 / 0.010098
≈ 0.98%

Less than one percent. A 99%-accurate test, on a disease affecting one in ten thousand, produces a positive result that is wrong more than 99 times out of 100.

The math is not complicated. It is fifth-grade arithmetic once the terms are translated. And yet the base rate - 1 in 10,000 - is ignored by essentially every intuitive response, because 1 in 10,000 does not feel like anything, and "99% accurate" feels very persuasive.

This is the Kahneman-Tversky finding, mathematised. Jack sounded like an engineer, so people said engineer, regardless of whether the room was 70% or 30% engineers. A positive test sounds like a disease, so people say disease, regardless of whether the condition affects 1% of the population or 0.01%.

The story always wins. The frequency always loses. And the gap between the story's answer and the frequency's answer is often not 5% or 10%. It is an order of magnitude or more.

Why Experts Are Not the Fix

The natural next thought is: experts must be better at this. They have training, they have data, they have experience.

Philip Tetlock tested that assumption for twenty years.

Beginning in 1984, he collected predictions from 284 experts - academics, policy analysts, journalists - on political and economic questions. He followed them for two decades. The results, published in 2005 as Expert Political Judgment, were severe: experts barely outperformed chance. Random assignment of probabilities would have done approximately as well.

In 2011, Tetlock received funding from IARPA - the intelligence community's advanced research agency - and launched the Good Judgment Project. Tens of thousands of volunteers, none with security clearances, made predictions on geopolitical questions. Their results were scored against those of career intelligence analysts with access to classified information.

The top 2% of volunteers - whom Tetlock labelled "superforecasters" - outperformed the intelligence community by roughly 30%. A hundred of them, working in teams of twelve, were so accurate that IARPA dropped the competing university-run teams from the tournament.

Three findings from Tetlock's work matter here, and they connect directly to both Kahneman and Dunning.

First: confidence and accuracy are uncorrelated among experts. The most confident predictions were not more likely to be correct. They were, if anything, slightly less likely. Confidence is a personality trait, not an accuracy signal.

Second: the best forecasters use base rates. They begin with the frequency - how often does this type of event occur? - and then adjust incrementally as evidence arrives. They run Bayes' theorem intuitively. The experts who lose start with a story and look for confirmation.

Third: the best forecasters update constantly. They do not commit to a position. They hold a probability, adjust it as new evidence arrives, and treat their current estimate as a draft rather than a statement. The experts who lose treat their initial judgment as their identity and defend it.

This is Bayes applied to people, and it produces the uncomfortable conclusion of this article:

The value of an opinion is not a function of the person's credentials or confidence. It is a function of their calibration - the degree to which their stated probability matches the actual frequency of being right.

A superforecaster with no security clearance beats an intelligence analyst with a classified briefing by 30%, because the superforecaster updates and the analyst anchors.

The Calibration Gap, in One Table

The table has a direction. As the method moves from narrative to frequency, the signal improves. As the source moves from credentials to calibration, the signal improves. Every row above the line is what you encounter in daily life. Every row below it is what actually works.

The Solution That Already Exists

The superforecasters are the proof of concept, and their method is neither secret nor difficult. It has three steps.

Step 1: Start with the base rate, not the story. Before evaluating any specific claim, ask: how often is this type of claim true, across all cases, not just the vivid ones?

Someone tells you about a startup that grew to $10 million ARR in its first year. Before judging whether the specific story is impressive or replicable, ask: what is the survival rate of startups in year one? What share ever reach $10 million? The answer to those questions - the base rate - is your prior, and everything else adjusts from there.

Step 2: Update in small increments. Each piece of new evidence shifts the probability. It does not replace it. "This founder was at Google" is a piece of evidence. It moves the probability slightly. It does not make it 90%.

The mistake - and it is the Kahneman-Tversky mistake, reproduced across fifty years of experiments - is treating a vivid new fact as though it overwhelms the prior entirely. Bayes says it does not. The prior sits there, heavy, immovable, and the evidence adjusts around it.

Step 3: Track your own accuracy. This is the step that defeats Dunning-Kruger directly. You cannot intuit your own calibration. You can measure it. Write down your predictions with a probability attached. After enough of them resolve, plot your stated confidence against your actual hit rate. The gap between the two lines is your calibration error, and seeing it is the thing that Dunning and Kruger found their bottom-quartile participants could not do: recognise the distance between where they are and where they think they are.

Superforecasters do this. Gamblers who survive do this. Insurance actuaries do this. Physicians who are good at diagnosis do this. The practice is identical across domains, because the underlying mathematics is the same, and it has been the same since 1763.

Why This Is the Article the Rest of the Feed References

Every interview on this account features a person who is confident about something. Every lecture features an expert making a claim. Every market analysis involves someone who thinks they know.

This article is the operating manual for evaluating all of them.

When a famous CEO says the market will go one direction: the confidence does not correlate with accuracy. Ask for the base rate. How often are CEO predictions about market direction correct?

When a renowned investor says a specific sector will dominate: the renown does not correlate with accuracy. Ask for the calibration. Does this person track and publish their hit rate?

When an audience full of people agrees that a speaker is right: the agreement does not correlate with accuracy. It correlates with how good the story was. Kahneman proved this in 1973 and the mechanism has not changed.

When you yourself feel certain: Dunning and Kruger proved that the feeling of certainty is strongest precisely where competence is weakest. The feeling is not information. It is noise shaped like a signal.

The single most useful mental habit available to a person navigating a world full of confident claims is this:

Before the story lands, find the base rate. Before the opinion registers, ask for the calibration. Before the certainty settles, check how often you have been this certain before, and how often it was justified.

The mathematics has been free since 1763. The psychological findings have been replicated for fifty years. The practical proof - that ordinary people using these methods outperform intelligence analysts with classified information - has been public since 2015.

The only thing missing is the habit.

The Close

People who scored in the 12th percentile rated themselves at the 62nd. That is a fifty-point gap between where they were and where they thought they were, and they had no way to see it from the inside.

The participants who heard that Jack enjoys carpentry and sailing said "engineer" regardless of whether the room was 70% engineers or 30%. The base rate was right there, stated in the question, and it did not register.

Intelligence analysts with classified briefings lost to volunteers using base rates and small updates by 30%.

These are three data points from three separate decades, tested by three separate teams, and they all describe the same failure: a mind that runs on stories, not frequencies, and has no alarm when the story and the frequency disagree.

The solution is not to stop trusting stories. Stories are how the mind works, and fighting that is unproductive. The solution is to run one additional step before the story settles into certainty: find the rate, check the rate, adjust the story to fit the rate.

One equation. Three steps. Published centuries ago, proven repeatedly, used by almost nobody.

The next time someone tells you they are certain - about a market, a technology, a person, a future - the question that matters most is not whether they are smart, experienced, or prestigious. It is whether they have ever measured how often their certainty was right.

If they have not, their opinion is a story. Stories are compelling. They are not evidence.

The base rate is evidence. It is free, it is old, and it is sitting there waiting, the same way it was sitting in 1973 when an entire room of participants looked past it to decide that Jack was an engineer.