The angle is not a place

“Step out at forty-five degrees.”

It is one of the most common things you hear in a boxing gym. It sounds precise. It contains a number, a direction, and almost a promise: take the right step and you will end up in the right place.

The problem is that an angle does not exist on its own.

Forty-five degrees relative to what? To the initial line between the two boxers? To the direction of the opponent's feet? To their shoulders? To the path of the jab? To the centre of the ring? And above all: while you move, does the other boxer stand still?

Boxing does not happen on a stationary sheet of paper. It is a geometry that reacts. Every position changes the available actions, and every action forces both boxers to recompute position, distance, orientation and balance.

That is why an angle is not a place you arrive at. It is a relation you have to create before the other boxer manages to correct it.

This is not a scientific law. It is a model: a structure simple enough to be understood, drawn, simulated and — eventually — measured. The interesting part starts right here.

Picture two boxers facing each other. Boxer A takes a diagonal step. If B stays oriented toward the old position, A has changed the relation between the two bodies: A can see a line that did not exist a moment earlier, while B has to rotate to recover it.

Now run the same scene again, but let B follow the movement immediately. A's step is identical. The path is identical. The number is identical. The advantage is gone.

In a third scene A moves far less, but does it while B has shifted weight forward, lengthened the step and is finishing a right hand. The geometric displacement is smaller, but the correction demanded of the opponent is far more expensive.

Research on dynamical systems applied to combat sports proposes to look at the two opponents as a single interpersonal synergy: the perceptions and actions of both athletes are coupled, and attacking and defending possibilities emerge from the relation rather than from the isolated qualities of one body. Krabben, Orth and van der Kamp identify interpersonal distance as a candidate control parameter and note that finer order parameters are needed, “such as combatant's relative orientation or relative center of mass” [2].

That is exactly the gap this essay tries to slide a model into: if relative orientation is a candidate variable and nobody is yet measuring it systematically in boxing, we may as well define it properly before pretending we understand it.

You do not own an angle. You own it only until the other boxer cancels it.

To make the problem measurable we can, for a moment, reduce the boxers to two points on a plane. A sits at (xA, yA), B at (xB, yB). The distance between them is the formula everybody has already met at school.

But centimetres do not tell the whole story. A hundred and eighty centimetres can be an offensive distance for a tall boxer with a long reach and a dead distance for a shorter athlete. The useful measure has to be scaled to the person: we normalise distance by a characteristic length LA, for instance A's functional reach.

This is not a blackboard invention. In 2006 Hristovski and colleagues asked eight novice boxers — aged twenty-one to twenty-three, fresh out of a two-semester course in elementary technique — to hit a heavy bag fixed to the wall. They prescribed no specific punches, only that the boxers maximise efficiency and variety. The metre in front of the bag was calibrated into ten ten-centimetre segments, and at each distance every athlete threw sixty punches.

The result is that jabs, hooks and uppercuts did not appear with the same probability at every distance: there were critical values of scaled distance at which a whole class of punches appeared in or dropped out of the repertoire. Around a scaled distance of roughly 0.6 the authors located the region of maximal metastability — the region where a boxer can switch between punch types with the greatest flexibility [1].

Distance does not only measure how far the target is. It measures which decisions are still credible.

Two points are useful but incomplete. A boxer has a direction: feet, hips, shoulders and guard are not oriented the same way at every instant. Two athletes can sit at the same distance and have completely different options.

So we give each boxer an orientation vector, oA and oB, and call rAB the vector running from A to B. The signed angle between A's facing and B's position is:

The same formula, applied to B with the vector rBA, gives βB. At this point we are no longer only asking “where are the boxers?”, but how much A faces B, how much B faces A, and which of the two has to correct their structure more to become operational again.

From these two measures we can build a very simple indicator.

If A has |βA| = 8° and B has |βB| = 38°, then G = 30°. A is looking almost straight at B; B has to rotate thirty degrees more to rebuild a comparable alignment. This is the kind of position a gym would call “taking the angle”.

But the number alone does not award the advantage. A might be out of range. A might have crossed feet. A might be falling sideways. A might have won a nice photograph without owning a single available punch. That is why G has to be read together with at least two other conditions.

The third variable is the hardest. Balance cannot be deduced from the position of a point: it needs information about the feet, the centre of mass and the phase of the movement. Biomechanics helps explain why. Dinu and Louis analysed the cross, hook and uppercut with seventeen inertial sensors on twenty-three boxers — fifteen elite and eight junior — and found systematic differences not only in the force produced but in how body segments contributed: juniors compensated with the shoulder, elite boxers recruited elbow and pelvis better [3].

A 2025 study measured lower-limb contribution with dual force plates in ten amateur boxers, before and after a nine-and-a-half-minute fatiguing circuit: punch force dropped by an average of 4.26%, with the largest reductions on the cross and the left hook [7]. The useful conclusion is not that a single correct biomechanical position exists. It is more cautious than that.

An angle without a base is a photograph. It is not yet an action.

Geometry becomes tactics when it produces a delay. Suppose A leaves the line at some instant. B starts to rotate, resets the feet, recovers the target, and becomes sufficiently aligned again later. The difference between those two instants is the realignment window.

To measure it for real you need a threshold declared before the analysis: we treat B as realigned when their angular error falls back under a tolerance ε.

This definition is still incomplete without looking at the base: a boxer can have the torso apparently pointed at the target and still be unable to transfer weight, defend or counter. But τ lets us ask a better question than “how many degrees did it create?”.

For how long did the movement force the opponent to chase a relation that had already changed?

This is where the check hook and the pivot start to make sense. The hook is not just a punch that lands while the other boxer comes in; the pivot is not just an elegant turn. The combination tries to leave the opponent oriented toward a position that no longer exists.

So far these are words with symbols in them. To see whether the model says anything, I wrote a kinematic simulation: A exits along a 17° arc around B's position, at constant tangential speed (about 1.4 m/s), staying oriented on the target. B is a tracker with only two parameters — a reaction latency and a maximum angular velocity — and is run in three different ways. The pivot lasts 0.35 seconds and starts at 0.30. Integration step 5 milliseconds, horizon 2 seconds, initial distance 1.65 metres, functional reach 0.90 metres, tolerance ε = 12°.

None of these constants is measured on real athletes. They are editorial assumptions, declared in the code and in the report. But once they are fixed, everything else — βA, βB, G, ρA, τ — is computed from the geometry at every step, not drawn by hand.

The third result is the one I care about. Against the opponent committed forward, peak asymmetry exceeds what the same step achieves against a completely motionless opponent: 22.0° versus 17.0°. A's step is identical in all three cases. The difference is that B was still investing along the old line.

And there is a second detail the model only makes visible because it was forced to be explicit: a pure pivot around the opponent preserves distance. In the first two scenarios ρA stays nailed to 1.83. It only changes in the third, and not thanks to A: it is B who, by walking forward, closed the distance on themselves.

The full script — simulation, τ measurement and generation of all three figures — can be downloaded here: numpy and matplotlib are enough to rerun it and reproduce exactly the figures quoted above.

The visible angle is in space. The usable advantage is in time.

We are used to imagining a fight as a race of speed: who starts first, who moves the hand faster, who gets there first. But the time of an action is not the speed of the punch.

Piorkowski, Lees and Barton compared jabs, crosses and hooks in ten competition-standard athletes, captured with an eight-camera optoelectronic system, thrown either as a single maximal punch or inside a combination. Hooks reached higher contact velocities than straight punches, but with significantly longer delivery times. And the single maximal punch hit 9.26 m/s against 7.49 m/s for the same technique thrown out of synch inside a combination [6].

These are laboratory numbers, not free sparring. They do not tell us which punch “is better”. But they remind us that speed, execution time and tactical availability are three different things. A theoretically fast punch is unusable if the body first has to rotate, recover the feet or find the target again. A slower punch can arrive earlier because it starts from a structure that is already loaded.

Geometry creates or removes these times before the punch even starts.

A favourable position is not there to produce more actions. It is there to improve the quality of the actions available.

Dunn and colleagues analysed twenty-six amateur boxers across nineteen bouts at the 2015 Australian national championships, elite male divisions. Winners did not throw more punches than losers: total volume was similar. They were more accurate. 33% of their punches were scored as landing, against 23% for the losers, and completely missed punches fell from 27% to 17% [4].

The sample was small and the judging system studied was the amateur ten-point must system. We cannot turn these percentages into a universal rule. But the principle is consistent with our problem: a good angle is not judged by the number of steps or punches it generates, but by the improvement in the quality of the relation.

A 2025 study of the twelve finals of the 2023 IBA Women's World Championships recorded 1,323 offensive and 1,456 defensive actions. The winners performed fewer offensive actions than their opponents (635 against 688) but with consistently higher effectiveness ratios across all three rounds. And — this is the part that concerns us — pivoting was the action most strongly associated with winners, with a standardised residual of +4.77 [5].

And in any case, association does not prove that pivoting makes you win. The best athletes may pivot well because they already possess superior timing, reading, balance and precision. The causal direction is not established. But the result makes the question worth studying.

Does the movement really create a better position, or is it just the visible sign of a broader technical ability?

The first is hunting for the perfect angle. No number works independently of stance, distance, reach, speed and the opponent's state. Forty-five degrees can be too many if they take you off your base. Ten can be enough if the other boxer is already committed. Thirty can be useless if the opponent turns with you. The correct angle is the smallest change that produces an operational difference.

The second is confusing lateral movement with advantage. Moving sideways does not automatically take you off the line of attack: you can move laterally and stay perfectly readable, you can even accompany the opponent's realignment and leave the symmetry intact. That is exactly the REACTIVE scenario of the simulation, the one where G never crosses the threshold.

The third is stopping the analysis at the best frame. A photograph can make a position that lasts a tenth of a second look dominant. The important variable is not the peak of G, but its evolution: G(t). A good movement produces a rapid rise in asymmetry and holds it long enough to allow an action. A decorative movement produces a spectacular peak and collapses immediately.

To give a project like this any weight you have to avoid the most common failure of sports writing: mixing scientific results, technical experience and metaphors as if they were the same thing. It is better to declare the levels separately.

These variables do not constitute a system for awarding points or predicting a winner. They exist to turn vague words — “line”, “angle”, “off-axis” — into more precise observations.

The first study does not need laboratory sensors. It needs definitions better than a coach's memory. You could start from technical sparring filmed with a fixed, elevated camera: the top-down viewpoint reduces perspective distortion and lets you mark the approximate centre of each boxer, the feet, shoulder orientation, the start and end of the displacement, the moment of the punch and its outcome. Each exchange becomes one row.

The hard part is not computing the formulas. It is deciding consistently when a pivot begins, when it ends, what “clean punch” means and how to recognise a stable base. Thomson, Lamb and Nicholas built a notational analysis template for amateur boxing and assessed its reliability: repeated observations by the same analyst were solid, with agreement between 80% and 100%, while agreement between different analysts was distinctly less impressive, even though it improved with tolerance margins [8].

Data does not become objective because you put it in a spreadsheet.

There would be no need to present the work as settled truth. It would be enough to call it what it is: an exploratory study.

A model always risks falling in love with its own variables. We could measure an angle perfectly and miss a feint. We could compute distance and fail to see that a boxer is tired. We could record shoulder orientation and miss a hand controlling the line of sight, a frame on the forearm, a foot blocking the exit.

The fight is bigger than its representation. The job of the mathematics is not to reduce it to five numbers. It is to force us to specify what we believe we are seeing.

If we cannot answer, maybe all we saw was a good-looking movement.

Boxing is often narrated through places: the centre, the ropes, the corner, the inside. But none of those places holds value on its own. The centre is useful as long as it lets you choose. The ropes are dangerous as long as they reduce your exits. The inside favours whoever manages to operate first. An angle exists only as long as it asymmetrically changes what the two boxers can do.

That also changes how you train it. Drawing an X on the floor and stepping onto it is not enough. You need a task in which the opponent can react, follow, anticipate or lose the line. You need to check the final position. You need to know whether the movement produced a punch, an escape, a new entry, or only the need to start over.

The angle is not the point you arrive at. It is the time you force the other boxer to spend before you become available again.

You do not need to move far. You need to change the relation early enough to be ready before the other boxer has finished correcting it.