Put a problem in front of someone that they have never seen, with no formula that fits and nothing in memory to look up, and watch what happens. Some people find the structure in it and some do not, and the gap between them is not about what they know. It is about what they can work out on the spot. That is fluid reasoning. Working memory sets how much you can hold in mind at once. Processing speed sets how fast you move through it. Neither settles whether you can actually solve the thing in front of you, because that takes a different function, the one that looks most like raw intelligence. It is also the one the entire brain-training industry claims it can sell you more of. On what it is, how it is measured, and what actually moves it, the honest answer is narrower than the pitch. If working memory is the mind’s RAM and processing speed is the clock, fluid reasoning is the processor doing the work: the cores computing an answer to input the machine has never seen. The defining word is novel. You are deriving the answer on the spot, from the structure of the problem itself rather than from anything in storage, which is why it travels across languages and cultures better than almost anything else we measure. The formal split is old and it has held. In 1963 Raymond Cattell separated general intelligence into two factors: [[fluid reasoning::Fluid reasoning: solving new problems on the spot by finding patterns and drawing inferences, without leaning on stored knowledge. Also called fluid intelligence, or Gf.]], the live problem-solving, and [[crystallized intelligence::Crystallized intelligence: the knowledge and skills you have banked over a lifetime, such as vocabulary, facts, and learned procedures. Called Gc. It keeps rising for decades while fluid reasoning falls.]], the knowledge you have banked. His student John Horn tested and extended it, and the distinction sits at the center of the modern [[Cattell-Horn-Carroll model::Cattell-Horn-Carroll model: the mainstream modern taxonomy of cognitive abilities, which sorts them into broad factors like processing speed, memory, and reasoning sitting beneath a single general factor.]] today. Fluid reasoning matters more than its one slot suggests, because it sits almost on top of [[g::g: the general factor of intelligence, the shared variance across every cognitive test. Fluid reasoning loads so heavily on it that the two are sometimes treated as nearly the same thing.]], the general factor. Load a battery of tests and fluid reasoning correlates with the common factor so tightly that some researchers treat the two as nearly the same thing. Measure reasoning well and you have measured most of what a full IQ test measures. The substrate is a coalition of regions. Reasoning runs on a frontoparietal network, the lateral prefrontal cortex working with the parietal lobe, and the sharpest evidence comes from damage. Duncan and colleagues mapped what they called the [[multiple-demand system::The multiple-demand system: a set of frontal and parietal regions that switch on for almost any hard, novel task, whether reasoning, working memory, or problem-solving. Damage there, specifically, is what drops fluid intelligence.]], the regions that switch on for almost any hard, novel task regardless of its content. When Woolgar and Duncan measured focal brain lesions against fluid-intelligence loss, it was damage inside those specific regions that predicted the drop, not frontal damage in general. Ten cubic centimeters of destroyed tissue inside the network cost several IQ points; the same volume outside it cost almost nothing. Here the hardware analogy earns a caveat. Your brain has no separable cores you could point to and count. The multiple-demand system is a distributed, overlapping network that also carries working memory and attention, not a dedicated arithmetic unit. Keep the metaphor, because it captures the thing that matters, that some part of the machine does the actual solving while the rest fetches and holds and paces. Just do not go looking for the chip. The prototypical test is the one you have almost certainly taken: a grid of figures with one cell blank, and a set of candidate pieces to complete it. That is [[matrix reasoning::Matrix reasoning: a grid of figures with one cell missing; you infer the rule connecting them and pick the piece that completes it. The prototypical fluid-reasoning task.]], and Raven’s Progressive Matrices is its most famous form, with the Matrix Reasoning subtest of the Wechsler scales and the Woodcock-Johnson fluid cluster close behind. The appeal is obvious. There are no words to know, no facts to recall. You either see the rule or you do not, which is why matrices get sold as culture-fair. That label is the first way the test fools you. A matrix is culture-reduced, not culture-free. It still asks you to read a grid left to right and top to bottom, to know that the blank wants completing, to be comfortable with the whole abstract convention of a puzzle on a page. Give the same figures to someone who has never sat a test like it and the score reflects the unfamiliarity as much as the reasoning. There is no such thing as a thought with no context; there is only context you happen to share with the people who wrote the test. The second way is timing, and it is subtler than it looks. Take a matrix test under a tight clock and you are no longer measuring only reasoning. You are measuring how fast you reason, which folds processing speed back into the score and, as Gonthier showed, can change the statistical character of the test itself. The untimed version measures something closer to how far you can get with no pressure. The timed version measures how much of that you can do quickly. They are related but they are not the same number, and a test that does not tell you which one it ran is not telling you what it measured. The third is the thing that makes reasoning tests both impressive and a little suspicious: they predict almost too well. Because matrix reasoning loads so heavily on the general factor, a forty-minute matrix test forecasts life outcomes about as well as a full IQ battery. That predictive power is real. It is also a reason for caution, because a single narrow puzzle is standing in for something enormous, and no one number should carry that much weight. Fluid reasoning has the least flattering age curve of any cognitive function. It peaks early and it falls, and the thing that rises to replace it is its own opposite. In Hartshorne and Germine’s large online study, the clean finding was that there is no single age at which the mind peaks; different abilities crest at different times. Processing speed tops out in the late teens. Working memory and reasoning hold into the late twenties, then slide. Vocabulary, the crystallized measure, keeps climbing into the sixties. Plot fluid and crystallized on the same axis and you get the defining picture of cognitive aging: two lines crossing, one falling as the other rises. Read it honestly, though. This is cross-sectional data, different people at different ages, so some of the early decline is a difference between generations rather than a change inside one person. Follow the same people over years and the drop starts later and runs gentler. But the shape is not in doubt, and it is more useful than depressing: the reasoning you use to solve the new is on loan, and the knowledge you build to handle the familiar is what you keep. Two things sharpen the picture. Fluid reasoning is state-sensitive on a timescale far shorter than aging, because it rides on the two functions that came before it. A bad night narrows the working memory it draws on and slows the clock it runs on, and the reasoning score falls with them. Much of what feels like a dull day is a well-slept function running on a starved one. And reasoning grows more genetically anchored as you age. The heritability of general ability climbs from around forty percent in childhood to two-thirds by late adolescence. That does not mean the environment stops mattering; the leading reading is that people increasingly steer toward the environments that suit them, so genes and surroundings pull in the same direction. Either way, the older you get, the less the number is something you can push around from outside, which cuts against the intuition the training industry sells. You can get a rough read at home, with more caveats than usual, because reasoning is the easiest function to fake a gain on. Skip the ad-funded IQ sites; use the International Cognitive Ability Resource, an open, public-domain set of matrix and reasoning items built by psychologists and hosted through the academic SAPA Project. Its short form runs sixteen items in under ten minutes and reports where you land against a large sample. Then hold the conditions still, because the number inflates easily. The worst offender is practice. Sit the same test twice and your score jumps; the reasoning has not improved, you have just seen the format before. The retest meta-analysis puts that inflation around five to six IQ points on the same form, and still around three on a fresh one. Timing matters too, in the way the measuring section warned: an untimed sitting at your kitchen table is not comparable to a timed norm, so match the condition or do not compare across it. And a sixteen-item test has a low ceiling, so if you are strong it cannot tell you how strong; it clusters everyone near the top and calls it a tie. Which leaves one honest use. Treat one sitting as a noisy baseline, not a verdict. If you want a trend, use a fresh form, keep the timing fixed, and leave enough months between sittings that you are not just remembering puzzles. The number is real enough to track and far too soft to defend to two decimal places. It tells you roughly where you sit and, more usefully, whether you are running the function well rested or into the ground. This is where the brain-training industry made its boldest bet. The landmark claim, from Jaeggi and colleagues in 2008, went past memory to reasoning itself: a demanding dual n-back task, they reported, raised fluid intelligence, and more training produced more gain. If true, you could drill your way to a higher ceiling on the function nearest raw intelligence. Dual n-back became a movement. It did not survive better studies, and the reason is worth keeping. The early trials used passive control groups, people who did nothing and knew it, so expectation and motivation could pass themselves off as a real gain. Put in an active control that keeps both sides equally invested, as Redick and colleagues did in 2013, and the transfer to reasoning disappears. The meta-analyses settled it: training reliably improves the trained task and its close cousins, and does essentially nothing for fluid intelligence. The cleanest cut is the one to remember. Weak studies showed a tempting effect; the well-controlled ones showed an effect indistinguishable from zero. One thing does move it, and the exception is instructive. The intervention with the strongest causal evidence for raising measured intelligence is the least glamorous one imaginable: staying in school. Pooling more than six hundred thousand people across natural experiments, Ritchie and Tucker-Drob found each additional year of education worth around three IQ points, an effect that persists for years. The contrast is the whole lesson. A game that looks like reasoning does nothing; years of actually reasoning about hard content do something durable. That points at the one move you can actually make. You cannot add cores, but you can keep feeding them genuinely new problems instead of the same ones dressed up, because the only thing that exercises reasoning is reasoning about something you have not already solved. Everything else is subtraction: stop running the cores you have on no sleep, on a full working memory, on an untreated condition that taxes the whole system. The ceiling is largely hardware and largely genetic, and it settles more firmly with age. You cannot raise it. You can live closer to it. Fluid reasoning is the function that flatters us most and yields least. It is the closest thing we have to raw intelligence, which is exactly why it draws the biggest promises and breaks them. Read it the way you would any measure of the mind: against your own baseline, under fixed conditions, as a rough vital and not a verdict. When the number sags, ask what you did to the working memory and speed underneath it before you conclude anything about the reasoning on top. And notice what the aging curve is really telling you. Fluid reasoning falls, but the line rising to meet it is everything you have stored: the vocabulary, the patterns already learned, the problems that are no longer new because you have solved them before. That is a different function, the knowledge you keep rather than the reasoning you spend.
