Prediction register

Co-contraction grows with the stakes

If the disorder is the motor system stiffening a gesture that it should leave free, the stiffening should scale with the expected cost of the error. As that estimate is raised (by making the stakes and the score more salient, or with instructions that direct attention to the hand and to the release) antagonist co-contraction during the loose should rise with it: graded, monotonic and specific to the release, instead of appearing all at once or not appearing at all. The dosing variable is the antecedent estimate, measured independently of the outcome, not the score that results from it.

Registered on

17 July 2026

How to test it

Manipulate the expected cost of the error in graded steps (the stakes and the salience of the score, or instructions that direct attention to the hand and to the release with increasing insistence) confirming at each step, with self-report and an autonomic index, that the estimate has risen, and recording the electromyography of the forearm during the loose. The prediction is a monotonic increase of co-contraction with the antecedent estimate, hooked in time to the release and not present to the same extent in the slow draw and hold.

Read it in the essay

A continuous variable under a step outcome

If the abruptness of the onset is a threshold on a variable that rises continuously, and not a jump between two stable states, the underlying physiology should change smoothly while the visible gesture changes in a step. As the trigger is increased, co-contraction should rise gradually and monotonically, but the behavioural outcome, clean release versus freezing, should flip at a threshold value of that rising stiffening.

Registered on

17 July 2026

How to test it

Track co-contraction and the outcome of the release together while the trigger is increased in fine steps, and look for two things: a smooth and gradual increase of co-contraction accompanied by a categorical change of the release at a threshold; and the absence of the early-warning signatures of a true bistable transition, no critical increase in the variance of the underlying variable, no lengthening of the recovery time before the flip. These are the documented precursors of genuine coordination transitions, and here they should not appear.

Read it in the essay

The specificity to acts under the latency

The disorder should strike only the gestures too rapid for feedback, those whose duration is at or below the latency of visuomotor feedback, and spare those slow enough to be controlled with feedback, even when the expected cost of the error is held equal and measured independently in the two. And, as a consequence, slowing a vulnerable gesture beyond the feedback threshold should reduce its vulnerability with no reduction of threat.

Registered on

17 July 2026

How to test it

In the same affected athlete, compare gestures that differ in duration but are otherwise matched: the brief release against a deliberately prolonged and controllable version of the same finger action; or the rapid release against the slow draw and hold. Hold the expected cost of the error equal across the conditions, and verify with the antecedent measures that the estimate is the same, so that the duration of the gesture remains the only variable that changes. The disorder should stay confined to the sub-latency gesture. A stronger test lengthens the release itself, a slowed or differently timed loose, so that it crosses from below to above the latency of feedback, and asks whether the vulnerability falls as the gesture becomes correctable.

Read it in the essay

Co-contraction bound to context, not lesion

The motor signature of the disorder, antagonist co-contraction, should appear hooked in time to the release and only when the system treats the gesture as threatening: present when the trigger is present, absent when the gesture is executed in its predictive and relaxed mode. And it should be the same co-contraction that experience had removed, the stiffening that falls as a skill consolidates, here returning under threat.

Registered on

17 July 2026

How to test it

Record the electromyography of the forearm during the release in two conditions: with the trigger present (a target with a score, an instruction to control the release) and with it removed (a neutral target, an external focus). The co-contraction signature should not be constant, but switch with the condition (present in the first, absent or much reduced in the second) in the same athlete and in the same session.

Read it in the essay

A neural signature of de-automatization

The de-automatization of the gesture should be accompanied by a measurable change in cortical dynamics, in the direction of a greater involvement of verbal-analytic cognition with motor control; in particular, an increased communication between the left temporal region and the frontal one during the gesture.

Registered on

17 July 2026

How to test it

Use the established marker of the involvement of cognition with motor control: the coherence between the left temporal region (T7) and the medial frontal one (Fz), which is lower in experts than in the less skilled and is understood as an index of the engagement of verbal-analytic processing with the planning of movement. Compare the cortical activity of the affected athlete in the predictive condition (clean release, neutral target) and in the de-automatized one (disturbed release, target with a score or feedback instruction). The prediction is an increased T7-Fz coherence in the de-automatized condition: the neural correlate of a gesture monitored rather than predicted.

Read it in the essay

A dissociation between interventions, by mechanism

Interventions should help if and only if they change the mode of control, not if they simply change arousal. The manipulations that lower the estimation of threat or shift control away from the gesture (a neutral target, an external focus of attention) should abolish the disorder temporarily; those that lower arousal without changing the mode of control (relaxation, beta-blockade) should not; and an instruction to monitor and control the release should reinstate it.

Registered on

17 July 2026

How to test it

In affected athletes, compare four conditions on the same outcome (clean release versus freezing or snatch): a neutral target; an external focus of attention (on the target or on the arrow, not on the hand); a relaxation or anxiolytic manipulation; and a feedback-seeking instruction. The predicted ordering is that the first two abolish the disorder, the third does not, and the fourth reinstates it. The external focus is a well-defined manipulation known to promote automatic control; the neutral target is the one that practitioners already report as effective, and that removes the very centre to be controlled.

Read it in the essay

The effect is mediated by measured co-contraction

If the estimation of threat degrades the release by way of impedance, then antagonist co-contraction is not a simple correlate of the disorder, but its mediator: the path from the imposed cost of error to the ruined release should pass through it. Fix the co-contraction, and threat should lose its grip; supply the co-contraction, and threat should become superfluous.

Registered on

17 July 2026

How to test it

Two complementary manipulations. The first: raise the imposed cost of error while recording the co-contraction of the forearm and the outcome of the release, and formally verify whether the effect of the manipulation on the outcome is carried by the measured co-contraction, a mediation analysis in which the input-outcome path is reduced once co-contraction is entered. The second: manipulate the presumed mediator directly. Induce antagonist co-contraction of the string forearm in a neutral, low-threat condition, a neutral target, and ask whether the freezing and the breakdown of the timing appear with no threat present; and reduce co-contraction in a high-threat condition, with electromyographic biofeedback or targeted relaxation of the forearm, and ask whether the release is saved even if the centre and the score remain.

Read it in the essay

On real data TOTAL+ stays among the best tie-break rules

If I had to make a prediction today about what will survive contact with independent empirical data, I would say: the top of the ranking more than its fine ordering. I expect the closely spaced positions to be reshuffled, some weighted rules to give back part of the advantage obtained inside the model, and TOTAL+ to remain competitive to the point of emerging not only as the best rule at minimum informational cost, but as the best rule overall in the relevant empirical domain.

Registered on

11 September 2026

How to test it

Re-run the ranking of the fourteen tie-break rules from chapter 10 on independent empirical data instead of model-generated matches, and compare the resulting order with the one in the volume.Test criterion added on 2 October 2026, after registration.

Read it in the essay

Anomalous arrows become less frequent at higher levels

My hypothesis is that a substantial share of out-of-pattern shooting reflects execution errors or disturbances, and that ability consists partly in the capacity to reduce their frequency: I therefore expect ε to fall as one moves up in level, and the departure from the regular core not necessarily to disappear even at elite level, but to be on average more pronounced at lower levels.

Registered on

11 September 2026

How to test it

ε is the share of anomalous arrows, drawn from a wider group, in the volume’s mixture model. It is tested by estimating ε by level on real data; according to the volume this needs impact coordinates, because from the rings alone the frequency and severity of anomalous arrows are confounded.Test criterion added on 2 October 2026, after registration.

Read it in the essay

Breakdowns exist, for everyone (P-CAT)

Every archer produces a minority of shots that do not obey the same law as the ordinary shot: a few per cent, never zero. The book calls these shots breakdowns and gives three plausible origins: physiological (the release is a fine motor skill performed under load, and under load it can degrade in jumps as well as by degrees), technical (nocks, rests and limbs) and attentional (a lapse of attention in the middle of the release, which punches a hole in the group rather than widening it). The expected signature is an excess of impacts far from the group even within a single session, on a steady day, that a single bell curve does not contain.

Registered on

16 October 2026

How to test it

Counting distant arrows is not enough: a single distribution with heavier tails can produce as many as a mixture of two populations, and the signature cannot be seen in scores alone. The test needs the coordinates of every impact, a session under steady conditions and the shooting order together, over a long series (hundreds of arrows), with the definition of a breakdown and the two competing models (a continuous tail, or a mixture of ordinary shots and breakdowns) fixed in advance, with the date of the analysis. The prediction fails if the continuous-tail model predicts the impacts held out of the fit better than the mixture does, even when there are many distant arrows.

Breakdowns get rarer with level, not milder (P-EPS)

The frequency of breakdowns falls as level rises: from about 5–6 per cent at the lower end of the competitive scale towards about 2 per cent among the elite. Their severity, on the other hand, stays roughly constant. According to the model the best archers do not make smaller errors: they make the same kind of degraded execution less often, because learning makes breakdowns of the shot less frequent but not less severe, and part of the breakdowns (equipment, lapses of attention, chance events) does not belong to trainable technique and never falls to zero.

Registered on

16 October 2026

How to test it

On a cohort of archers at different levels, sessions are recorded with the position of every arrow in millimetres, and the frequency of breakdowns and their depth, measured in units of each archer's own group spread, are estimated separately. As level rises, the frequency of events outside the group must fall, not their relative depth. The prediction does not get its signature if severity also falls along with frequency, or if frequency and depth, once normalised for the group, both stay unchanged.

The day dominates, and it is lopsided (P-DAY)

Once the sampling floor (how much the total swings from chance alone) is taken out, the largest individual component of noise is the state that changes from one day to the next: it weighs more than the tails of single arrows and more than fine drifts. Days, moreover, are not symmetrical: bad days are worse than good days are good, because many causes can take ten points off in an afternoon and almost none can add them beyond the ceiling of the consolidated shot. The two parts stand together.

Registered on

16 October 2026

How to test it

The same reference session (72 arrows, with the position of every impact) is repeated on many days, and how much performance scatters from one day to the next is separated from how much it would move through sampling alone; five sessions are enough for a first breakdown of the variance, not for a verdict. The between-day component must dominate sampling, and the day-to-day deviations must have a tail towards the bottom. The prediction does not show its signature if the between-day component does not dominate, or if the deviations turn out roughly symmetrical.

The real floor sits a fifth above the bell curve (P-FLOOR)

At every level, the floor measured on an archer's totals under steady conditions exceeds by about 15–25 per cent the one the pure bell curve (the model without breakdowns) predicts for the same expected score. The difference is not constant along the scale: it is largest around the middle and smallest at the ends, because lower down the group is already wide and breakdowns add little, while among the elite a broken arrow has less room to hurt the result. The signature is a ratio between the observed personal floor and the bell-curve floor of about 1.2 on average, which itself traces a bell shape along the scale. The book specifies that the prediction is read on the floors, that is on the standard deviations, and not on the variances.

Registered on

16 October 2026

How to test it

Twenty rounds give a single archer a first indication, not a decision: over twenty rounds a standard deviation has a 95 per cent interval running from about 0.76 to 1.46 times the observed value. It needs a cohort of many archers at different levels, each with at least twenty comparable rounds recorded in the same format, and a simulation of the estimator carried out before the data are collected. The prediction fails if real floors turn out indistinguishable from those of the pure bell curve, or if the ratio is constant or rises steadily along the scale. It has to be decided alongside the ratio between the floor estimated on ends and the one estimated on arrows, because dependence between neighbouring arrows would inflate the ratio as if it were tail.

Within a round, arrows have no memory (P-IID)

Each arrow is a draw without memory: the previous error does not raise the probability of the next one, and there is no such thing, as a mechanism in its own right, as a cursed end. Shooting is a discrete task, a programme that closes and restarts with every arrow, and the cluster of errors an archer feels as contagion may be the kind of sequence the mind picks out even in pure chance. The signature: in series where the shooting order is kept, with the state of the round held steady, a breakdown is not followed by an excess of further breakdowns. That the state does not move within the round is a separate assumption, which the book declares on its own and which is not part of the bet.

Registered on

16 October 2026

How to test it

Totals alone cannot decide it: positive dependence between neighbouring arrows and an independent distribution with a heavy tail inflate the total in the same way. It needs long series with the coordinates of impacts and the real order of the arrows. The prediction fails if a breakdown is followed by a systematic excess of further breakdowns, that is by contagion. The book considers it the costliest prediction in the register: with an average correlation of just 0.01 between arrows, the predicted floor at 630 would rise from 8.8 to about 11.5 points, and every figure that depends on independence would have to be redone.

The tail runs deeper than the bell curve (P-DEPTH)

Measured in units of the personal floor, the depth of an archer's worst round clearly exceeds what a Gaussian distribution predicts, and the difference grows as level rises. Two reasons are proposed: a round that collects several breakdowns can sink further than sampling from a Gaussian core allows; and near the theoretical maximum the room above the average shrinks while the room below stays wide. The alternative is that this second effect, the ceiling of the scale, is enough on its own to explain the whole tail.

Registered on

16 October 2026

How to test it

The tail in points is compared with the tail in the coordinates of the group, in millimetres: the part due to the ceiling of the scale must vanish when moving to the geometry of the group, while the part due to breakdowns must remain. If the deep tail disappears in millimetres, the ceiling is enough on its own and the prediction does not get its signature. It needs personal archives with at least 20 comparable rounds per archer, plus sessions with the position of every arrow for the check in coordinates.

A collapse round is a crater, not a bad afternoon (P-BLACK)

In the same archer's worst rounds, ends are far more scattered among themselves than in their best rounds, and the best rounds are also the most even. The proposed mechanism adds the day and the breakdowns together: the state of the day shifts performance in a diffuse way, while breakdowns are sharp events that sink a single end, and chance can pile more than one into the same round. The alternative is the spiral, in which the error changes what comes next and calls up further error.

Registered on

16 October 2026

How to test it

A raw comparison of worst and best rounds favours the prediction from the start, because choosing the lowest total also chooses part of the scatter that produced it. The signature only counts against a reference built with the same selection: series of rounds are simulated from the model without breakdowns, the worst and best rounds are picked from them by the same rule, and the difference in scatter is compared with the one observed. It needs end-by-end scores from at least 20 comparable rounds, with the selection rule declared in advance and part of the data held back for the check. Telling the crater from the spiral also needs series with the shooting order: if a systematic excess of further breakdowns follows the first one, the spiral gains weight.

The floor changes with the bow (P-DIV)

For the same expected score and conditions, the floor curve changes from one division to another: compound quieter, barebow noisier. The difference does not come from the arithmetic of the scoring rings, which cannot tell which bow shot the arrows, but from the shape of the distribution and its tails: in compound the mechanical release removes one of the main sources of breakdowns, the one tied to the fingers, and changes the tail above all; in barebow the lack of a sight adds an aiming error on both axes.

Registered on

16 October 2026

How to test it

Floor curves are built separately, division by division, from personal archives with a known round format, and compared at the same expected score and conditions: they must not all collapse onto the same curve. The prediction does not show its signature if, once score and conditions are controlled, the curves overlap within the uncertainty; in that case the difference seen on the range depends only on how the groups of archers are made up.

Level is stable (P-REL)

Latent level, that is the real ability underneath the observed scores, measured over windows of at least eight comparable rounds, is very stable from one window to the next. Once enough averaging has reduced the noise of sampling and of single days, a persistent personal component emerges: consolidated technique changes over months, days and sampling over days or a single round. The alternative is that there is only the moving average of a state that keeps wandering.

Registered on

16 October 2026

How to test it

On a personal archive kept in the same format over a season, averages are calculated over successive windows of at least eight rounds. They must correlate with each other more than the sampling floor and the day-to-day variability alone can explain. If, once those two components are accounted for, successive windows show no additional persistence, the prediction does not get its signature.

Day-to-day drift is vertical (P-DIR)

The centre of the group moves from one day to the next mainly in the vertical, while within a single day the group stays almost round. Elevation depends on launch speed, draw length and release timing, which can change from day to day, and on a judgement that has to be rebuilt every time; horizontal alignment rests on steadier references, such as the shooting line, the torso and the lateral set-up. The alternative is a drift with no preferred direction, spread in a similar way over the two axes.

Registered on

16 October 2026

How to test it

The reference session (72 arrows, with the position of every impact) is repeated on several days. The two signatures must appear together: the ellipse built on the centres of the different days must have its major axis vertical, and the groups within each day must stay almost round. The prediction does not show its signature if the groups within the session are also systematically vertical, or if the drift between days has no preferred axis.

Precision and consistency are two separate axes (P-AXIS)

Precision, that is how tight the core of the group is within a round, and consistency, that is how little that core changes from one round to the next once the floor is taken out, are largely independent across archers. Precision depends on fine control of the ordinary shot, which repetition refines; consistency depends on whatever interrupts, degrades or shifts that shot: attention, routine, equipment, sleep, tuning, the state of the day. The fraction of breakdowns is not a third axis but one of the components of consistency. The alternative is that ability is a single dimension.

Registered on

16 October 2026

How to test it

On a cohort, the two axes are measured with reference sessions repeated over several days. The correlation has to be calculated with level held fixed: both axes improve as level rises, and a correlation over the whole population would mostly measure general ability. The prediction requires that, once level is controlled, knowing how tight the core is says little about how stable it will stay from one day to the next; if the correlation stays high with level held fixed, the prediction fails.

The shape of the group is more stable than its size (P-SHAPE)

The ratio between the axes of the group repeats from one session to the next better than its overall spread: the individual shape of the group is a more stable trait than its size. Shape depends on the architecture of the shot, that is on which link in the technical chain is the main limit and on how that asymmetry shows up on the target, and it changes slowly with technical work; size absorbs more of the condition of the day. The alternative is that shape and size are equally unstable.

Registered on

16 October 2026

How to test it

With reference sessions repeated over several days, the variability of each measure is split into a stable part and a part that changes from session to session. The stable share must be larger for the ratio between the axes than for the size of the group. The prediction does not show its signature if shape and size are equally stable, or if the ratio between the axes varies as much as the spread or more.

Vertical error has a ballistic component (P-VERT)

Vertical error grows with distance more than proportionally and carries the signature of the spread in launch speed; horizontal error grows almost in proportion to distance, as expected of a mainly angular error. The reason is ballistic: the arrow's drop depends on the square of the time of flight, and small differences in speed turn, over distance, into vertical differences that do not follow the law of angular error. The alternative is that both axes grow in proportion to distance.

Registered on

16 October 2026

How to test it

The same archer's group is measured at three or more distances, with the position of every arrow, and the two curves are compared as angles: beyond about 50 metres vertical and horizontal spread must start to diverge, with the vertical growing faster. The prediction fails if the two angular curves stay parallel or if the vertical grows less than the horizontal. Even when the signature appears, it does not prove the cause on its own: getting closer to the cause needs the arrow-by-arrow relation between launch speed and vertical deviation.

The X count measures fine centring (P-XRATE)

At the same score, the percentage of Xs measures mainly the quality of fine centring, that is how close the group stays to the centre, more than the ability to group, and it varies between archers more than chance would produce. Two archers with the same average can therefore differ consistently in their share of Xs because they differ in fine tuning, not necessarily in spread. At high levels the score saturates and loses sensitivity to centring, while the X keeps telling it apart; and tuning is a job separate from the shot. The alternative is that the X is simply a consequence of grouping.

Registered on

16 October 2026

How to test it

Using the coordinates of impacts in reference sessions, together with the archer's own scorecards, the share of Xs is compared at the same score with two quantities: the offset of the centre of the group from the centre of the face, and the spread of the group. The share of Xs must track the offset of the centre more than the spread; if it mainly tracks the spread, the prediction does not show its signature.

The quiet eye foreshadows the arrow (P-QE)

The duration of the final fixation on the aiming point, that is how long the gaze stays still there just before the release, grows with level. Within the same archer, moreover, arrows preceded by a shorter fixation produce worse impacts, and under pressure the fixation tends to shorten. The proposed reasons: the motor programme uses a stable visual reference to set the last parameters of the action, and a long fixation is the visible sign of attention directed outwards. The three parts stand together: the prediction fails if any one of them fails.

Registered on

16 October 2026

How to test it

Gaze, recorded with eye-tracking glasses, and the position of every impact are recorded together. The central signature lies within the same archer, on a steady day: shorter fixations must systematically precede worse impacts. Comparing levels tests the growth with experience, and a session in which the stakes are raised in graded steps tests the shortening under pressure. If duration does not predict the outcome within the same archer, the explanation that the fixation is merely a by-product of level is sufficient, even if stronger archers fixate for longer.

Eye dominance leaves a horizontal signature (P-EYE)

Cross-dominant archers (dominant hand and dominant eye on opposite sides) who do not correct for it show a specific excess of horizontal offset and spread. Those who remove the conflict by closing or covering the dominant eye that is not aiming are no different, at the same level, from archers who are not cross-dominant. The dominant eye carries more weight in locating the target: when it is not the aiming eye, perceived alignment can slide sideways and leave a signature mainly on the horizontal axis.

Registered on

16 October 2026

How to test it

On a cohort, with eye dominance and the correction strategy recorded, sessions with the position of every arrow are compared at the same level. The prediction can fail in two independent ways: if the horizontal excess does not appear in cross-dominant archers who do not correct, or if correcting does not remove it.

Level carries over from indoor to outdoor, regularity does not (P-TRANSFER)

Moving between indoor and outdoor, general level carries over to a large extent, regularity much less. Level depends mainly on the technical quality of the shot, which stays the same in both settings; a large share of outdoor regularity depends instead on being able to read and handle the wind, which indoor shooting neither trains nor measures. The alternative is a symmetrical transfer: an archer who is regular indoors is regular outdoors to the same extent as they keep their level.

Registered on

16 October 2026

How to test it

On personal archives covering both the indoor and the outdoor season, the book estimates how much of the level carries over from one to the other and how much of the variability specific to outdoor shooting does: the first coefficient must be high, the second markedly lower. A wind session, with several archers on the same line and the wind measured in sync with the impacts, checks that the remaining variability is tied to reading the wind and is not generic noise. The prediction does not show its signature if level and regularity carry over with similar strength, or if outdoor variability has no link with wind handling.

A round has a profile over time (P-TIME)

A round is not uniform over time: it has an entry cost at the start and, beyond an individual load threshold, a progressive decline. The profile is predicted to be smaller at the top and increasingly marked further down the levels. In the first minutes neuromuscular activation can improve fine performance, while with load come tremor and lapses of attention; and the threshold of decline can be trained, so stronger archers should have a flatter profile and a more distant threshold. The alternative is a round that stays flat from start to finish.

Registered on

16 October 2026

How to test it

It needs series that keep the order and timing of shots, and sessions extended beyond the length of a competition, often past 150 arrows, to find after how much load a lasting decline appears and which axis gives way first. The average profile must turn out not flat, with an opening phase different from the middle stretch and a decline beyond a threshold, and its shape must change with level; the two parts stand together. If the decline appears just the same in a low-load control session, what is being measured is the afternoon, not fatigue.

A breakdown leaves a tail of disturbance (P-REC)

After an arrow outside the group, a measurable part of the disturbance carries over into the next end instead of ending with the shot that produced it. The size of this tail defines a personal recovery coefficient, and the residual disturbance decreases with experience: a glaring error raises activation and shifts attention onto the outcome just produced, and the routine, which experience builds and consolidates, is there precisely to absorb it. The alternative is pure independence even across breakdowns.

Registered on

16 October 2026

How to test it

In long series with the real order of the arrows, a breakdown must be followed by a small excess of breakdowns or decline in the arrows immediately after, followed by a return to the ordinary level, and this tail must become shorter or weaker with experience. The predicted effect is subtle: if the excess is large and feeds on itself, the picture is one of contagion; if no memory of the error appears, independence is enough and the prediction fails.

Form has a short memory (P-PERS)

Form on the day can last a few competitions, but not long enough to become a stable trait on which to build long-term forecasts. Being in form is real, but its predictive value is much shorter than everyday language suggests: sleep, neuromuscular tone and activation fluctuate over hours or days, while what stays stable for months belongs to level. The resemblance between one day and the next fades within about 2–3 competitions; the book declares this figure as an assumption. The alternative is long, readable cycles of form.

Registered on

16 October 2026

How to test it

With reference sessions repeated across the whole season, the book measures how much the state of one day resembles that of the following ones. The resemblance must fade quickly. If the dependence stays strong over gaps longer than two or three competitions, a persistent component compatible with cycles of form emerges, and the prediction fails.

Trainability is almost never a stable trait (P-TRAIN)

In the great majority of groups, the apparent difference in response to training remains compatible with the hypothesis that everyone responds in the same way. What the coach reads as 'he responds, he doesn't' comes mainly from three effects: regression to the mean, when archers are chosen or judged from a partly noisy starting point; selective dropout, because those who respond little, or believe they do, leave more readily; and the meeting between a coach and a person at a particular moment in their life, a real effect but a different thing from a stable trait.

Registered on

16 October 2026

How to test it

It needs cohorts with random assignment, a baseline measure precise enough to reduce regression to the mean, and dropout handled by counting every assigned archer in the analysis, including those who leave. If trainability were a trait, the same archers would stay among the strongest and weakest responders in parallel groups or independent replications. The prediction fails if, once regression, selection and dropout are controlled, that ranking repeats as a stable trait would instead of reshuffling.

Growth is overestimated (P-DEV)

Growth observed over short windows tends to be larger than real long-term growth, and the bulk of the technical gain comes in the early stages, while later improvements become steadily smaller. Regression to the mean and the selection of those who stay on the path inflate curves read from raw data; and once the shot is consolidated and automatized, every step forward costs more work and yields less. The alternative is linear growth, readable directly from the observed curves.

Registered on

16 October 2026

How to test it

On long longitudinal archives, kept in the same format, with cohorts followed over several years, a model is used that separates change in level from measurement error and allows for those who drop out. Corrected slopes must come out lower than raw ones, and the progress curve must show diminishing returns; the two parts stand together. If corrected curves match the raw ones and improvement stays linear, the prediction does not show its signature.

The first end pays an entry cost (P-WARMUP)

The first end pays an almost universal entry cost, which a good warm-up should reduce until it disappears: fine neuromuscular performance improves in the first minutes of activity, and the eye needs the first arrows to adapt fully to the task. The alternative is a profile already flat from the first end, as you would expect of a fully automatized shot that is available straight away.

Registered on

16 October 2026

How to test it

On personal end-by-end scorecards, with the warm-up recorded, the first end must sit below cruising level, and on days preceded by a full warm-up the gap must shrink or disappear. If the first end does not sit below cruising level, or the warm-up does not reduce the gap, the prediction does not show its signature. It is one of the entries that can be decided with a single archer's scorecards and a spreadsheet, together with sessions recording the position of every arrow.

The day can be read early, a little (P-EARLY)

The first arrows say something real about the day, beyond the points they are worth: not much, because the floor of six arrows is high, but more than nothing. If the day sets the width of the group, every window of the round is a sample of it. The alternative is independent windows, in which the first end says nothing beyond the points it is worth. The signature is a correlation between the first end and the rest of the round above that of pure sampling.

Registered on

16 October 2026

How to test it

It needs end-by-end scorecards from many competitions, with several days per archer, the individual average effect removed and a structure that keeps arrow, end, day and archer apart: with a single day per person, differences between people disguise themselves as the state of the day. Thresholds and windows are fixed on one part of the data and checked on the rest, and the comparison is made against a reference built with the same dependence between arrows. If the correlation does not exceed that of pure sampling, the prediction fails.

Rust is tuning (P-RUST)

Short breaks cost almost nothing; after long ones the main problem is getting the system back in tune rather than recovering a lost shot. Consolidated motor skills are tenacious memories, while the fine match between equipment, eye and aim deteriorates faster: it is the difference between forgetting how to ride a bicycle and finding it again with flat tyres. The alternative is a technical loss in the strict sense, growing with the length of the break.

Registered on

16 October 2026

How to test it

On personal archives around periods of rest, recovery in the first sessions back must be very fast, too fast for a rebuilding of the shot from scratch. If the return is slow and proportional to the length of the break, technical loss remains plausible and the prediction does not show its signature.

The seasonal wave is almost all composition (P-SEASON)

The rise and fall of scores over the year depends mainly on the calendar, the type of competition and the weather. The share due to a real periodisation of the archer's state, comparing them with themselves under comparable conditions, should be small: many important causes change along with the season and get mixed up with it. The alternative is an annual wave that stays visible even with the same competition and conditions.

Registered on

16 October 2026

How to test it

On personal archives covering several years, comparing the archer with themselves at the same type of competition and under the same conditions, the seasonal wave that remains must be a small fraction of the raw swing observed over the year. If a large wave remains even with the same competition and conditions, the prediction fails.

Pressure is a vector (P-PRESS)

Pressure does not change a single quantity: it can shift the centre of the group, widen it, increase breakdowns or alter shot timing, and different people show different combinations. The final score adds these effects together and can hide very different mechanisms. The three routes proposed are the classic explanations of performance under pressure: excessive activation widens the group through tremor; conscious monitoring of an automatized shot increases breakdowns; distraction changes timing and centre. There is no reason for the same person to show all three. The alternative is a single mechanism, the same for everyone.

Registered on

16 October 2026

How to test it

In training, the stakes are raised in graded steps, from nothing at stake to increasing stakes, checking that the pressure is actually felt, and centre, width, breakdowns and shot timing are measured separately. For the same overall drop, different individual profiles must emerge: some widen without shifting, some add breakdowns without widening, some change mainly their timing or centre. If a single mechanism accounts for everyone, the prediction fails.

The favourite erodes through composition, not psychology (P-LEVEL)

Round after round, the advantage of the higher-ranked archer shrinks, but only to the same extent as the gap between the contenders narrows: the bracket concentrates increasingly similar archers, and when the gap is small any format discriminates less. With the gap held fixed, the erosion must disappear. The alternative is a psychological effect of the stakes, whereby the favourite loses ground in the later rounds even against opponents just as far behind.

Registered on

16 October 2026

How to test it

It needs complete bracket archives with known qualification rankings, and the gap between contenders declared in advance as a control variable. If, with the gap controlled, the effect of the round disappears, the composition of the bracket is enough; if it survives, the prediction fails, and it would be the first serious evidence of an effect of the stakes specific to the favourite. The book points out that P-FINAL says the same thing and needs the same archive: a single outcome decides both.

The deciding set is not of higher quality (P-DECIDE)

In the sets that decide a match, the average quality of shooting does not rise compared with the personal average of the two contenders. Reaching the fifth set means the two stayed close in the previous sets, and two weak archers can be as evenly matched as two strong ones; selection acted on the sets already shot, not on the deciding one. In the model the expected difference is zero, and with a different make-up of the population it can even turn negative. The alternatives are a real rise, from moderate activation helping the automatized shot, or a real fall, from cracking under pressure.

Registered on

16 October 2026

How to test it

The deciding set is compared with the personal average of the same two archers, controlling only for variables fixed before the match, such as ability and probability of winning. The quality of the match must not be controlled after the fact, because that could manufacture the sign of the effect. It needs a session in which the stakes rise in graded steps, and situations in which the stakes change for reasons unrelated to the two archers. The prediction fails if the deciding set turns out consistently above the two archers' personal average.

In a shoot-off the group decides, not character (P-TIE)

The share of shoot-offs an archer wins is explained by the width of their group and by specific practice of the single high-stakes arrow; once these components are accounted for, no stable residue remains that could be put down to a general trait of holding up under pressure. The book states it conditionally: for the same number of shoot-offs shot and the same level of opponents, the share of shoot-offs won rises with the archer's level, and once that level is properly included, it explains all of the rise.

Registered on

16 October 2026

How to test it

The raw share of shoot-offs won mixes level, specific practice and selection, which all push in the same direction, so the raw correlation decides nothing. It needs shoot-off archives with the qualification rounds matched, so that each archer's group width is known. The prediction fails if, once group width and volume of competition are controlled, a residue of wins remains that repeats when it is measured again on the same archer.

After a disaster comes a rebound above average (P-BOUNCE)

After an end much worse than one's own average, the next end scores above the personal average, not merely at the average. The distinction matters: if arrows are independent, after an extreme end the next one returns to the average through regression alone, and that is not a rebound but an absence of memory. The rebound is the excess above that expected return, and the proposed reason is attentional: a glaring error narrows attention onto the shot for a few minutes and reduces the ordinary spread.

Registered on

16 October 2026

How to test it

On end-by-end scores from at least 20 rounds per archer, in a cohort of different levels, the average excess of the following end above the personal average is measured, with its interval. The comparison average must be estimated leaving out both the selected end and the next one, otherwise one ends up reading one's own arithmetic; the rules for cases where the worst end is the last of the round, or where ends tie, are fixed in advance. The prediction fails if the following end only returns to the average (pure regression) or stays below it (a spiral).

The rebound is earned (P-RESIL)

Recovery after a disastrous end is not handed out at random: it appears above a certain threshold of automatization and grows with level. Below that threshold, after the disaster an archer stays close to the low level just shown, or gets worse still. The routine that absorbs activation and detaches attention from the outcome is a product of technical development, and an archer who has not yet automatized the shot pays twice: first the error, then the cost of having to manage it. The alternative is that resilience is mainly a trait of character, independent of technical level.

Registered on

16 October 2026

How to test it

On the same data as P-BOUNCE, end-by-end scores in a cohort of different levels, the size of the rebound must change with level: positive and growing among the strongest, to the point of changing sign at the lower end of the scale. The two predictions are tested on the same data, but one can fail without the other. This one fails if the size of the rebound does not depend on level.

Clutch is small (P-CLUTCH)

Once level, opponent and selection are taken out, the stable part of over-performance in high-stakes moments is small compared with the noise of the few occasions available. Truly weighty occasions are few and already heavily selected, ideal ground for building reputations out of noise; and the known theories of pressure explain at most who loses less, not who improves beyond their usual level. The alternative is a real, stable talent for performing better when the stakes rise.

Registered on

16 October 2026

How to test it

Whether high-stakes over-performance repeats in the same archer is measured on independent occasions: with repeated real occasions or, as a controlled substitute, with a training session in which the stakes rise in graded steps. If those who over-perform once tend to do it again beyond what chance would produce, the prediction fails.

Release times have two humps (P-PANIC)

In a competitive population, with nothing at stake, the distribution of the time between the sight entering the gold and the release has two humps instead of one: some archers release with a very short and barely variable latency, others with longer and far more scattered times, with a sparsely populated zone in between. The proposed reason is associative learning: a stimulus that reliably precedes a response ends up calling it up, and the gold can become the trigger. The prediction concerns only the shape of the distribution: the effectiveness of treatments and health questions are left out.

Registered on

16 October 2026

How to test it

Release times are measured in training, first with nothing at stake and then with increasing stakes, with the rule that classifies the two groups declared before looking at the data. The signature is a two-humped distribution that survives controlling for division, equipment and level, together with a group elongated in the direction from which the sight enters the gold. The prediction fails if the times form a single continuum with a long tail, or if the two humps are explained by division or equipment.

Wind has a shared part and a personal one (P-WIND)

A large share of drift in the wind is common to the whole shooting line; what remains is individual and behaves as a relatively stable parameter of the archer. The average wind mainly shifts the centre of the group, while gusts widen it. Slower and lighter arrows pay more, which accounts for part of the differences between divisions; and reading the wind, choosing when to shoot and how much to correct is a skill that can differ consistently between archers. The alternative is that wind produces only individual noise, with no recognisable common component along the line.

Registered on

16 October 2026

How to test it

Several archers shoot together on the same line, on several days with different winds, with the wind measured by anemometers synchronised with the impacts. The common shift of the field is separated from each archer's residuals by estimating the two variances and their covariance together, because archers who read the wind poorly miss more precisely when the wind is strong; until that is done, individual susceptibility is expressed as an interval. The prediction fails if no factor common to the line emerges, or if individual residuals do not repeat from one session to the next.

There is a wind you cannot feel that still moves the group (P-GUST)

There is a band of wind the archer does not perceive but which is already enough to shift the centre of the group measurably: the damage starts before the wind is felt. On the line, wind is judged mainly through the skin and through visual cues such as flags, and the threshold at which these cues become obvious can be higher than the force needed to move an arrow a few centimetres at 70 metres. The alternative is that the threshold of perception and the threshold of effect coincide, with no blind band at all.

Registered on

16 October 2026

How to test it

Using the same wind session (several archers on the same line, anemometers synchronised with the impacts), each archer's own perception is recorded alongside the instrument readings. The signature is a systematic shift of the common component even in the windows the archers judge to be calm. If the common component does not shift in those windows, the prediction fails.

Cold costs, and it costs below a threshold (P-COLD)

Below a certain temperature performance gets worse, and the drop increases as the temperature falls; above that threshold the effect is negligible. The mechanism is physiological: fine finger dexterity deteriorates as the hand cools, and the release is one of the most exposed components, hence the expected signature mainly on the vertical axis. The alternative is a continuous effect, growing gradually as the temperature falls, with no threshold.

Registered on

16 October 2026

How to test it

In sessions recording the position of every arrow, with the temperature noted window by window and, where possible, the temperature of the hand measured directly, the curve of performance against temperature must show a recognisable knee, together with an increase in vertical spread below the threshold. If the curve falls continuously, with no knee, the prediction fails.

Rain does not matter (P-RAIN)

At equal visibility, rain produces no measurable drop in performance. Neither modern equipment nor the shot is affected to any relevant degree, and most of what is put down to rain would depend on factors that often come with it: poor visibility, wind and cold. It is a bet that openly goes against common sense. The alternative is a real effect of rain, mediated mainly by visibility or grip.

Registered on

16 October 2026

How to test it

In sessions recording the position of every arrow, with conditions noted window by window, visibility, wind and temperature are controlled for. The prediction fails if, once these factors are taken out, an effect of rain remains. It is tested on the same data as P-COLD, but the two remain separate bets, because the outcome of one does not decide the other.

The heartbeat window (P-HR)

The slowing of the heartbeat just before the release distinguishes levels, and within the same archer arrows shot without that slowing tend to produce worse impacts. The slowing may be a sign of attention directed outwards; and each heartbeat causes small movements of the chest, so lowering the rate in the decisive window removes a source of disturbance. The alternative is that it reflects only general physical condition, with no predictive value for the single arrow. The book registers it for completeness, without developing it in the text.

Registered on

16 October 2026

How to test it

Heart rate is recorded with a chest strap synchronised with the shot, in sessions recording the position of every arrow and in sessions where the stakes rise in graded steps. The signature is a relation, within the same archer, between the size of the slowing before the release and the quality of the impact. If that relation is absent, the prediction fails.

Breathing writes in the vertical (P-BREATH)

When breathing departs from its usual pattern, the first expected effect is an increase in vertical spread, before horizontal spread. The chain running from the diaphragm through the chest and shoulder to the bow acts mainly in the vertical plane, and many aiming disciplines converge on releasing during the pause after breathing out. The alternative is that a sufficiently expert shot fully compensates for changes in breathing, leaving no trace on the impact. Registered for completeness, not developed in the text.

Registered on

16 October 2026

How to test it

With a breathing strap synchronised with the shot and the position of every arrow recorded, the test looks for an excess of vertical spread in arrows shot outside the usual breathing pattern. If arrows shot outside the pattern show no vertical excess, the prediction fails.

Balance is projected onto the target face (P-SWAY)

Body sway while aiming grows together with spread, and part of the decline caused by fatigue and pressure passes through balance control. If the relation is causal, specific balance work must carry over to shooting as well. Sway introduces an angular component that distance amplifies, like a beam of light that moves further the further away it is projected. The alternative is that the bow arm and the rest of the chain compensate for the sway before it reaches the target. Registered for completeness, not developed in the text.

Registered on

16 October 2026

How to test it

With a force plate underfoot, in sessions where the stakes rise in graded steps and in sessions extended to fatigue, the test looks for a relation between sway and spread that holds even with technical level controlled, and for a share of the drop due to fatigue or pressure that is explained by the increase in instability. If the relation disappears with level controlled, or the drop does not pass through balance, the prediction fails.

Short sleep widens the group and leaves the centre where it was (P-SLEEP)

After a sleepless night the group widens without its centre moving. It is a different signature from that of a tuning problem, which shifts the centre while leaving the width more stable. Sleeping little degrades fine motor control and sustained attention, two components needed to repeat the release well. The alternative is that short sleep produces only a generic bad day, with no signature distinguishable from other sources of variability. Registered for completeness, not developed in the text.

Registered on

16 October 2026

How to test it

With reference sessions repeated over several days and a sleep diary, treating the length of sleep as a dose, the test looks for an increase in spread with the centre essentially unmoved on days of short sleep. If short sleep shifts the centre, or leaves no signature distinguishable from other sources of variability, the prediction fails.

An everyday dose of caffeine widens the group (P-CAF)

In a rested archer, an ordinary dose of caffeine widens the group without shifting its centre, with a signature consistent with increased tremor, rather than improving the shooting. With the dose both alertness and tremor rise, but in a slow, fine and already automatized shot the second should weigh more than the first. The alternatives are no effect, through habituation to caffeine, or a positive effect in sleep-deprived archers, in whom the extra alertness has something to make up. Registered for completeness, not developed in the text.

Registered on

16 October 2026

How to test it

Sessions recording the position of every arrow are repeated at declared doses, with the dose assigned before the session and not reconstructed afterwards. The signature is a wider group on full-dose days, with the centre unmoved. If the group does not widen, or the centre shifts, the prediction fails. It stays separate from P-SLEEP: the two cross precisely where lack of sleep changes the effect of caffeine.

Mixed practice pays off in competition (P-CI)

For the same volume, mixed practice (alternating distances, faces and tasks, never two identical blocks in a row) gives worse scores during training but better ones in competition than blocked practice. Blocked work lets the archer reuse a solution still held in memory, mixed work forces them to rebuild it each time, and that is the ability competition calls for. The serious alternative is that in the archer's single, highly consolidated shot mixing adds nothing, because classic interference arises mainly from switching between different skills.

Registered on

16 October 2026

How to test it

A mixed regime and a blocked regime are compared, with a design declared in advance and at the same volume, measuring outcomes in competition and some time after the end of the intervention, not straight away. The two faces stand together: worse performance in training, better in competition and in delayed tests. If either is missing, the prediction fails.

The gap between training and competition is a personal parameter (P-SPEC)

The gap between training average and competition average, corrected for the floors of both, behaves as a stable personal parameter. Its size depends on how much training and competition differ for that archer, and making the two settings more alike reduces it: what is learnt transfers better the more conditions, rhythm and inner state resemble each other. The part of the gap left once that resemblance is accounted for is the component of holding up in competition. The alternative is that the gap is explained entirely by sampling, because competitions are few and therefore noisier than training sessions.

Registered on

16 October 2026

How to test it

On an archive kept in the same format, with every session classified as training or competition and the conditions noted, the floor-corrected gap must stay stable across seasons and shrink in archers who make training more like competition. The two parts stand together: if the corrected gap is not stable, or does not shrink as the settings are brought closer, the prediction fails.

Overcorrection is widespread (P-CORR)

A sizeable share of archers correct more than the optimal correction calculated by the book's model allows, and pay for this overreaction with more spread, mainly on the vertical axis. The error just seen is vivid and the noise is abstract, so the mind tends to treat every deviation as a signal to correct; and archers are not usually given an explicit measure of the optimal correction, so experience gets close to it only by chance. The alternative is that experience alone is enough to calibrate the correction.

Registered on

16 October 2026

How to test it

The test measures how much the archer moves the shot after seeing an error, with the order of arrows recorded and a deviation deliberately imposed, and compares it with the optimal correction derived from their personal drift, measured with sessions repeated over several days. The prediction fails if the estimated correction does not exceed the personal optimum, or if the distance from the optimum does not come with additional spread that grows along with it.

The favourite's erosion in real brackets (P-FINAL)

In real brackets the favourite's advantage shrinks round after round to the same extent as the gap between the contenders narrows: it is an effect of composition, not necessarily of pressure. As the bracket advances, the archers left are ever closer in level, and the expected advantage of the better one shrinks. The alternative is a psychological finals erosion, which would bring the favourite down even when the gap between the two stays constant.

Registered on

16 October 2026

How to test it

On bracket archives with known qualification rankings, the gap is held fixed: the erosion must disappear. If it remains, an effect specific to the round or the stakes emerges and the prediction fails. The book states that P-LEVEL and P-FINAL say the same thing and need the same archive: a single outcome decides both.

The set system acts as a shield in real matches too (P-SHIELD)

In real matches between known profiles, the set system protects against the effect of the occasional breakdown in the way the book's model predicts: the set confines the catastrophic arrow to a single set, instead of letting it weigh on the whole match as in cumulative scoring. The shield benefits the archer who has some breakdowns, so the signature is that, at the same expected value, the more regular archer wins less than 50 per cent of the time, and even less under the set system than in cumulative scoring; in the model the figures are 45.4 per cent with sets and 48.5 per cent cumulative. The alternative is that this protection does not appear in real matches.

Registered on

16 October 2026

How to test it

On a personal archive of matches with the qualification rounds matched, so that each archer's profile is known, the test estimates how often the more regular archer wins at the same expected value, separately under the set system and in cumulative scoring. If the frequency does not stay below 50 per cent, or is not lower with sets than with cumulative scoring, the prediction fails.