The Extraction Order

Skill is not a movement. Skill is the extraction of the invariants that determine a class of situations, followed by an action the situation already calls for. Train the extraction and the movement arranges itself. Train the movement and the extraction stays where it was, which means the skill stays where it was.

This is a claim about causes, and it is meant to be tested. The Extraction Order is the instrument built to test it: three offices, four measures, one law of dose, and a clause that states in advance what result would prove it wrong. It is addressed to two readers who believe they do different work. The professional athlete stands over a ball with water on the left. The chief executive sits across from a board with a number that does not add up. Each must see, before the event resolves, which kind of situation this is. Each has been trained almost entirely in what to do once they know. Neither has been trained in knowing.

Look at where practice time goes. It goes to the visible form of the movement: the takeaway, the wrist angle, the hip turn, the phrasing of the opening remark, the shape of the slide. Form is what a coach can see from outside, so form is what gets corrected, and correction is what gets paid for.

The cost is exact. Technique training trains the visible and leaves the extractable untrained. The athlete rehearses what they already know and, because every session begins and ends with instruction, learns to wait for instruction. The dependence is not a character flaw; it is the product the method manufactures. A player who can execute every position in the swing and cannot tell, from the lie and the wind and the slope, which shot the situation demands, has been trained in the answer and never in the question.

The same culture assumes that skill accumulates from repetitions of the movement. A thousand identical swings on a flat range mat teach one situation a thousand times, and the course never presents it. Two further habits compound the error: internal cueing, in which the coach names a micro-motion and the athlete attends to the body, and the calm-mind school, which teaches the zone as a calm mind reached by breathing, mindfulness, and routines. Both place the work inside the performer. Neither places it where the information is, which is outside, in the situation.

Extraction is not raw sensation topped with interpretation. That view, the enrichment theory, holds that a novice and an expert receive the same data and the expert adds more to it: more memory, more inference, more assumption. Gibson and Gibson took it apart with a single consequence: if extraction improved by enrichment, then the more the performer learned, the less their extraction would correspond to the information actually present. It does not. The expert “sees and hears more, but this may be not because he imagines more, or infers more, or assumes more, but because he discriminates more” (Gibson and Gibson 1955). The wine taster does not add memories to the wine. The taster discriminates more of the variables already in it.

This is the whole turn. Improvement is differentiation, not addition. Goldstone named its mechanisms: attention weighting, in which the important dimensions gain weight; imprinting; differentiation, in which stimuli once indistinguishable become psychologically separate; and unitization, in which a task that required detecting several parts is done by detecting one constructed unit (Goldstone 1998). Kellman and Garrigan define this kind of learning as “experience-induced changes in the way [people] extract information,” and they state the consequence plainly: expertise “depends heavily on implicit pattern recognition and selective extraction skills,” and “the burden of explanation placed on the learning of facts, rules, or techniques is often exaggerated” (Kellman and Garrigan 2009).

Consider the radiologist’s first glance. The film holds the same pixels for the resident and the attending. The attending does not reason longer. The attending sees the thing the resident cannot yet separate from its background. Consider the chess position shown for five seconds and then taken away. The master recalls far more of it than the novice. Scatter the same pieces at random and the advantage collapses, because what the master extracted was structure, organized into larger units, and random placement contains no structure to extract (Chase and Simon 1973).

An invariant, then, is the feature of a situation that holds across its variations and determines its class: the grain of the green that decides the break, the sequence in a forecast that decides whether the number is real. Extraction, causally defined, is the act of pulling the determining structure out of a situation before the situation resolves; here, that feature. Nothing that does not change extraction counts as extractive training.

One constraint follows. Learning occurred readily when the invariant that determined the classification was available in the learner’s representation, and did not occur when the trainable invariant was decorrelated from it (Garrigan and Kellman 2008). Learning follows the learner’s representation, not the physical stimulus. Pointing at the right answer in the physics of the ball is useless if the athlete’s extraction does not carry it.

The quiet eye is the most measured form of the claim. Vickers defines it as the final fixation or tracking gaze before the final movement, on a specific location or object, within 3 degrees of visual angle, held for a minimum of about 100 ms. Higher level performers hold it longer. Under pressure it shortens and the number of fixations increases (Vickers 2009). The eye is not calm because the mind is calm. The eye is still because it has found the one location that carries the information, and it stays there.

It can be trained. Causer, Holmes and Williams took twenty international level skeet shooters, matched for ability, and gave one group a four step preshot routine with three video feedback sessions on their own gaze and an expert model, across an eight week intervention. Quiet eye duration rose from 397 to 423 ms. Onset moved from 257 to 244 ms. Shooting accuracy rose from 62 to 70 per cent, and the improvement transferred to competition. The control group showed no change (Causer et al. 2011). The same approach was tested in surgery: quiet eye training improved surgical knot tying more than traditional technical training in a randomized controlled study (Causer et al. 2014).

Focus is the second line of evidence. Across fifteen years of work, an external focus of attention, on the effect in the world such as the target, the flight of the ball, or the implement, produces better learning and performance than an internal focus on the body’s movements and positions (Wulf 2013). “Keep the left arm straight” directs attention inward. “Land it past the flag” directs it outward.

Pressure is the third. Skills acquired with explicit, verbalizable rules break down under pressure; skills acquired implicitly do not break down in the same way (Masters 1992). Every rule installed in words is a rule the performer can reach for when frightened, and reaching for it is how the skill comes apart. Extractive learning “does not consist of learning declarative information,” because “structures extracted by experts in a domain often cannot be verbally explained” (Kellman and Garrigan 2009). What cannot be said cannot be grabbed in a panic.

Now the limits, stated with the same weight. In darts, a 2020 study trained a focus of attention specific to what the thrower sees and lengthened quiet eye duration without improving throwing accuracy. Gaze moved; outcome did not. In the clinic, anxious individuals allocate excess attention to threat cues (Bar-Haim et al. 2007), and a whole class of interventions was built to train attention away from threat. The pooled interventions did not outperform controls on anxiety or depression outcomes (Cristea et al. 2015). Attention changed; the condition did not.

These are the most important findings in the file. A proxy can be trained without the thing it stands for, and a system that measures only gaze or only attention will report success while the performer fails.

Extraction training as it is usually sold is an inadequate idea. It names the effect: see the game better, be present, trust your eyes. It deletes the cause: nobody says what is to be seen or how seeing it changes. It forbids the measure: the zone cannot be scored, only felt. And when the performer fails, every failure lands on the performer for not being ready. That is knowledge of the first kind, which in Spinoza’s account arises from hearsay, casual experience, and imagination, and which he names the sole cause of falsity; the second and third kinds are necessarily true (Ethics II, Proposition 40, Scholium 2).

The standard is set at the opening of the Ethics: the knowledge of an effect depends on, and involves, the knowledge of its cause (Ethics I, Axiom 4). An account of improvement that cannot state its cause is not a weaker account; it is no account. The adequate version supplies the cause. Improvement is the extraction of invariants, trained by discrimination with correction, specified as an effect in the world, and stripped of useless labour.

Method is not the enemy of this. Method, properly understood, is the thing itself. In the Treatise on the Emendation of the Intellect, method “is nothing else than reflective knowledge, or the idea of an idea,” and “there can be no method without a pre-existent idea” (TIE 38). Method must furnish, first, “a means of distinguishing a true idea from all other [ideas]”; second, “rules for [grasping] unknown things according to the standard of the true idea”; third, “an order which enables us to avoid useless labor” (TIE 49). These three offices are the architecture of this system.

Spinoza also answers the regress: to work iron you need a hammer; to make the hammer you need another tool; but this does not go back to infinity. The mind, by its native strength, makes intellectual instruments for itself, and from those operations makes fresh instruments, and so ascends (TIE 30 to 32). The coach who refuses to transfer a movement is doing exactly this. The coach gives one true idea, one discrimination the performer can actually make, and lets the performer build the next instrument from it. A transferred prescription is a ceiling. A transferred instrument is not.

The same ground disposes of the calm-mind school. Fear over the water is an affect, and an affect cannot be restrained or taken away except by an affect opposite to it and stronger (Ethics IV, Proposition 7). No one talks down fear by deciding to be calm, and a breathing routine does not supply the opposite affect. What supplies it is an adequate idea of the task, which is itself a greater power of acting, and joy is the passage to a greater power of acting. The performer who can see what the shot requires is not calmer by effort. The performer is more capable in fact, and the fear loses its object.

The first office trains the extraction of the invariant that separates two outcomes. It answers Spinoza’s first requirement, a means of distinguishing the true idea from all other ideas.

The mechanism is classification before outcome. The situation is presented; the performer calls its class before the outcome exists; the outcome arrives; the call is scored. Will this putt break left or hold its line. Will this pitch be a fastball or a change. Is this forecast built on a real pipeline or on a hoped-for one. The call must be committed, in words or by signal, before the ball leaves or the quarter closes. Afterward it is marked right or wrong, and the mark is recorded.

Without that score there is no learning, only rehearsal. Hindsight is always correct. The committed call is the only thing that can be wrong, and only what can be wrong can be corrected.

The instruction is this. Choose one pair of outcomes that the performer confuses. Build a set of situations in which one invariant decides between them and everything else varies. Present them in mixed order, demand the call before the outcome, and score every call. Begin with contrasts coarse enough for the performer’s extraction to carry, then narrow them as the error rate falls. This is differentiation made into a drill: two situations that once looked the same become separate (Goldstone 1998).

The second office states the target as an effect in the world. Never as a movement. Never as an avoidance. It answers Spinoza’s second requirement, rules for grasping unknown things according to the standard of the true idea.

The worked example is the feared-water session. A player has a pond down the left side of a hole and a history of finding it. The Extraction Order refuses all technical instruction. The coach asks only for the ball in a named part of the fairway: the right half. Then finer: the right third, past the sprinkler head. Then the coach reverses the demand and asks the player to hit the ball into the water. Then to hit it just barely into the water, a yard past the bank.

The player’s dispersion resolves without one word about the swing. The reason is causal. “Stay out of the water” makes the water the object of attention, and an anxious mind already allocates excess attention to threat (Bar-Haim et al. 2007). An avoidance is a target with a negative sign, and the body does not read the sign. “Put it in the right third” is an effect in the world, and an external focus of attention produces better learning and performance than an internal one (Wulf 2013). Asking for the ball in the water finishes the work: the feared object becomes a location like any other, and what can be hit on purpose can be missed on purpose. The divot points where the ball was sent, and the player reads the divot, not the elbow.

The executive’s version is identical in form. “Don’t lose the room” is an avoidance. “By the end of the second slide, the chair has asked about the margin” is an effect in the world: observable, scorable, and aimed where the information is, at the other people.

The instruction is this. Write every target so that an observer who has never seen the performer could say whether it happened. If the target names a body part, rewrite it. If it contains the word “don’t,” rewrite it. Then tighten it until the performer misses half the time, because a target never missed teaches nothing.

The third office removes what is rehearsed rather than learned. It answers Spinoza’s third requirement, an order that avoids useless labour.

Four things are subtracted. The first is the drill of the known: any repetition whose outcome the performer can already predict. It feels productive and trains nothing. The second is the internal cue: it moves attention inside, and every explicit rule is a rule available to break under pressure (Wulf 2013; Masters 1992). The third is the coach’s narration, which supplies the answer before the performer has extracted it; a supplied answer is a prescription, and a prescription is a ceiling. The fourth is any measure that does not move. A statistic that stays flat whatever the performer does is decoration. The mechanism is arithmetic: training time is finite, and every minute spent on these is a minute taken from discrimination.

The instruction is this. At the end of each week, list every component of practice. Beside each, write the measure it moved. Strike every component with nothing written beside it.

What is not measured is not trained. A method that forbids the measure is first-kind knowledge wearing a uniform. The Extraction Order keeps four instruments, and each yields a number.

The first is recognition latency: how early, before the event, the situation was identified. Present the situation on video or live, cut it off at fixed points before the decisive moment, and require the call at each cutoff. Recognition latency is the earliest cutoff at which the performer’s classification accuracy reaches eighty per cent across a set of at least forty trials, reported in milliseconds before contact, release, or decision. For the executive, the briefing is released page by page, and latency is the earliest page at which the call reaches the threshold.

The second is error rate: the proportion of committed classifications that were wrong. It is recorded for every discrimination set and reported as a percentage with the number of trials beside it. It is the plainest instrument and the hardest to fake.

The third is gaze or attention onset and stillness, which is the quiet eye. With an eye tracker, record the final fixation before the final movement: its onset relative to the start of that movement, in milliseconds; its duration, in milliseconds; and whether it held within 3 degrees of visual angle for at least 100 ms without interruption (Vickers 2009). Where no tracker exists the measure is cruder, and it should be called cruder: video of the eyes, frame counted.

The fourth is availability under stakes: did the skill appear when it counted. Take the error rate, or the sport’s performance measure, in practice and again under consequence: competition, or a live board meeting. Availability is the consequence score divided by the practice score. A ratio near one means the skill is owned. A ratio well below one means it is borrowed.

The third instrument is always read against the fourth. The dart study is the reason: quiet eye can be lengthened while accuracy stays flat. Gaze is evidence of extraction, not proof of it. Only the outcome under stakes proves it.

The intervention must match the exposure. A practice that occupies twenty minutes cannot answer a condition that runs twenty hours. If a person’s difficulty runs fourteen hours a day, a twenty minute remedy is not treatment; it is denial.

This is why the Extraction Order does not sell a routine. A preshot ritual lasts seconds; the habits it is meant to correct run through every hour in which the performer looks at anything. The executive who misreads rooms misreads them all day, and a weekend workshop is a sip against that tide. Dose is therefore specified in the block that follows, and daily exposure to the relevant situations is counted at day zero. If the training dose is a small fraction of that exposure, the prediction is modest, and it is stated modestly in advance.

The session is a ninety minute block in four movements.

The field occupies the first twenty minutes. The performer plays or works in the real environment, or the closest faithful copy, with no instruction and no targets beyond the ordinary ones. The coach records which situations recur and which calls go wrong, and corrects nothing.

The discrimination set occupies the next thirty minutes. The coach selects one confusable pair from the field and presents a mixed series of situations in which one invariant decides between them. The performer calls each one before the outcome. The coach records the call and the outcome and says only “right” or “wrong.” There is no explanation of why.

The scored set occupies the next thirty minutes. The targets are now specified as effects in the world, tightened until roughly half are missed, and something real is attached to the score: a public tally, a forfeit, a ranking against the performer’s own last session. This is not pressure inoculation, the field’s habit of making practice stressful to simulate competition. It is the consequence condition the fourth instrument requires.

The withdrawal occupies the last ten minutes. The coach steps away. The performer repeats a short scored set alone, calling and scoring without any voice but their own. Then, and only then, the performer may ask questions.

The coach’s conduct is fixed, line by line. The coach gives no internal cues. The coach does not name micro-motions: no elbows, no wrists, no hips, no breath. The coach does not explain during the set. The coach does not rescue: when the performer fails, the failure is scored and the next trial begins. The coach answers questions only after the withdrawal, and answers them by naming the invariant, not the movement.

Day zero is measurement: recognition latency on at least forty occluded trials, error rate on the confusable pair chosen from the field, quiet eye onset and duration where a tracker is available, availability under stakes from one practice and one consequence condition, and a count of daily exposure. The numbers are kept by someone other than the coach.

The first week trains coarse discrimination: five sessions on one confusable pair presented with its widest contrast. The target is a falling error rate. The second week narrows the contrast and moves occlusion earlier. The target is earlier recognition at the same accuracy.

The third week adds specification and stakes. Targets are rewritten as effects in the world and tightened; the scored set carries real consequence. The feared outcome, whatever it is, is made into a target at least once. The target is an availability ratio that holds.

The fourth week subtracts. The withdrawal lengthens until, in the last two sessions, the performer runs and scores the discrimination set alone. Every component that moved no measure has been struck.

Day thirty repeats the day zero protocol exactly: the same trials, the same occlusion points, the same pair, the same conditions, the same person keeping the numbers. The comparison is day thirty against day zero, and against a control condition of equal time spent on technique, where one can be arranged.

The method is false if the following occurs. After thirty days at the stated dose, error rate and recognition latency do not improve against day zero, or improve no more than in an equal block of technique training. The method is also false if those two improve and availability under stakes does not, because an extraction that does not appear when it counts has not been trained in any sense that matters.

Two further results do not falsify the method but limit it, and they are stated here so that no one can later claim them as surprises. First, transfer is limited. Learning follows the learner’s representation (Garrigan and Kellman 2008), so an invariant extracted on one course, against one opponent, in one kind of meeting, may not carry to another where the invariant appears differently. A new class of situations requires a new discrimination set. Second, gaze training sometimes moves gaze without moving outcome. The dart study showed it; the attention-training meta-analysis showed the same shape at scale (Cristea et al. 2015). If quiet eye lengthens while error rate and availability stay flat, the quiet eye was not the invariant for that task, and it is dropped.

A method that states its failure conditions can be corrected. A method that cannot fail cannot improve.

You cannot practise seeing what you have never been asked to call.

A target that names the body trains the body to wait for a coach.

All things excellent are as difficult as they are rare, and none of them is rare for want of repetition.

Tomorrow, before any outcome, say what will happen, and write down whether you were right.