The Ontology

What kinds of thing exist in a grappling exchange

Ask what a grappling exchange is made of and most answers arrive as a list of positions and techniques. The ontology answers one level down, with an inventory of kinds of thing, sorted into three levels. At the physical level: two embodied participants and the things around them; the stable properties of bodies and room; the state that changes as the exchange runs; what each person can command; what each body does on its own; what gravity and the mat contribute; the rule by which the whole system moves to its next state; the paths it takes; and the policies that steer it. At the constitutive and contest level: what makes an interaction grappling at all, and what a particular ruleset makes legal, valuable, terminal or resettable. At the agent level: what each person can observe, what they know, what they can reliably execute, and what they decide. From those the ontology derives the adversarial relations the rest of the theory is built on, reachability, enforceability and defensive viability among them.

Positions, techniques and invariants are on none of the three levels. They are derived descriptions over the state, the policies, the transitions and the equivalences beneath them, which is why a theory built on this ontology can disagree with another theory about what advantage is while agreeing entirely about what exists. The inventory, in the ontology’s own closing words:

Positions, techniques and invariants are not treated as primitive physical objects. They are derived descriptions over deeper state, policy, transition and equivalence structure.

Two bodies, one system

The physical level begins with two people rather than one. The exchange contains two principal adaptive participants and the physical things they act through and against: the mat, a wall or a cage, clothing and equipment. Attacker and defender are not kinds of participant. They are local roles, read off what each person’s current policy is doing and where the terminal relations point, and the description stays symmetric until a relation gives one side a role. The ontology also declines to assume a standard body. Size, limb number, joint range, strength, mobility, fatigue and prosthetic embodiment may all differ, and a mechanically active prosthesis whose actuator a person can command counts as part of that person’s control system.

What follows from treating the pair as one system is the sentence the whole page rests on: the next physical state is produced jointly. Each person’s commands contribute, and so do the responses each body makes without being told, the passive mechanics of gravity, inertia, tissue and the mat, the stable properties of both bodies and the room, and whatever disturbs the system from outside. When a sweep lands, the bottom player’s drive, the top player’s failing post, gravity and the friction of the mat produced the next state together. The ontology does not describe what one grappler did as though the other were furniture, and it does not treat a result as the product of one input.

The next physical state is co-produced by both participants' controllable actuation, automatic organism response, passive mechanics, system properties and external disturbances.

What is stable, and what is changing

The ontology sorts what exists into six broad kinds rather than one: the physical things; the relatively stable properties of bodies, materials and the room; the state that evolves during the exchange; the primitive physical relations and laws, from actuation constraints to the transition itself; the constitutive and contest specifications that say what counts as grappling and what a ruleset makes of it; and the agent-level processes of capability, observation, information, belief and selection. Two of these matter most to what follows here, and they are the two most easily run together: what is stable about a body and what is currently true of it.

Stable properties, over the horizon being analysed, are things like mass, segment lengths, baseline joint limits, tissue properties, morphology and baseline force capacity, and, for the room, gravity, the friction of the surface, the compliance of a wall and the stiffness of a cage. The state is everything that materially changes as the exchange runs: where each body is and how it is moving; what is connected to what, the coupling and contact between the two; each organism’s current activation, force reserve, local and global fatigue, injury and tissue stress; and the state of the surface and materials. The line between them is drawn by the horizon. A limb’s length is a property; fatigue in the fifth minute is state; a joint limit changed by an injury during the round moves from one column to the other.

That is where fatigue, conditioning and injury enter the theory. They live in the state, and the state modifies what this athlete can currently execute, how fast a transition happens, how robust a structure is and what is safe. In the fifth minute of a hard round a guard player has the same physically available actions they had in the first minute, and fewer executable ones. Strength programmes, recovery planning and periodisation stay downstream: real, important, and not part of what grappling consists of. Diagnostically it is the difference between “the escape is unavailable” and “the escape is available and this athlete can no longer execute it”, and the second is a conditioning or pacing problem wearing a technical disguise.

Dynamic organism state can modify executable capability, transition dynamics, robustness and injury-relevant constraints. Strength, conditioning, fatigue, recovery and injury can therefore affect realised grappling mechanics without making periodisation, S&C programming, recovery planning or safety governance foundational ontology objects.

What each person commands, and what nobody does

What a participant can actually command is narrower than the word “control” suggests. The controllable actuation space is made of actuator-level quantities: joint torque, muscular activation, grip force, stiffness, controlled loading and unloading, release, and coordinated commands across several limbs. A hook, an underhook, a grip or a body lock is not, in itself, one of those. It is an achieved mechanical relation, produced by a policy acting through the actuators and kept in being by the same means. For a coach this changes the grammar of a correction. An underhook is a result an athlete keeps producing against an opponent working to undo it, and when it goes, something changed in the actuation on one side or in the mechanics between them.

Two further sources of force sit outside anyone’s commands. The organism produces automatic responses: startle, automatic postural corrections, reflex contractions, unconscious reflex movement, generated by the body and not selected by the current deliberate policy. And passive mechanics contribute without any organism-level selection at all: gravity, inertia, tissue elasticity, passive ligament tension, the reaction of a wall, environmental contact, material tension. The ontology draws these lines functionally, and says so; it claims nothing about the neuroscience of volition. Whether automatic response needs to stay a category of its own, or can be absorbed once the organism’s state is described finely enough, is one of the questions it keeps open.

A hook, underhook, grip configuration or body lock is not necessarily itself an actuator input.

Paths, not snapshots

A state is a snapshot. The exchange is a path through states, and the ontology treats the path as an object worth analysing even though it is generated by the transition and is not primitive. Two paths that arrive at similar gross states can differ in fatigue, residual momentum, tissue stress, the coupling history now encoded in the state, remaining force reserve and what each side is set up to do next. Two passers who reach the same top position, one after a long scramble and one from a clean pass, are in the same position and in different states, and the next exchange is where the difference can show.

What steers the path is a policy: a causal rule for selecting actuation as state and time change. It has to be causal, which means it cannot condition what an athlete does now on an opponent’s choice that has not yet been made. Movement, maintenance, waiting, release, switching between branches and coordinating several limbs are all policies. A technique is later represented as a learned executable family of policies, and the availability of a policy in the physics comes before any athlete’s ability to run it.

The current ontology adds the minimum a sequence needs before it counts as a policy at all. An athlete attacks the armbar from closed guard, loses it, attacks the triangle, loses it, and attacks the omoplata. That sequence may be a game or it may be three unrelated attempts, and the difference is whether the choices were conditioned on the opponent. “If they posture, the triangle; if they stack, the omoplata; if they pull the elbow free, back to mount” is a policy. Switching to the next attack after each one fails is, by itself, only chronology. That is why drilling a chain in a fixed order can produce a fluent athlete with no game, and why the evidence a coach should look for is branching on the opponent’s response rather than smooth repetition.

A causal policy requires action selection or continuation to be conditioned by relevant physical state and/or observation-information state over an appropriate temporal horizon. Mere chronological switching after success or failure is insufficient, by itself, to establish a coherent causal policy, although it may be compatible with one.

What makes it grappling

Physics alone cannot say which interactions count as submission grappling, and a definition built on outcomes alone pulls in neighbouring combat sports. So the ontology specifies the domain in four parts. The interaction is constituted through bodily positioning, direct or equipment-mediated coupling, force transmitted through grips, hooks, frames, clamps and supports, control of movement and orientation, and transitions generated through grappling mechanics; striking is not one of its constitutive mechanisms. The participants are opposed in a control and escape relation: each preserving useful control of their own body, establishing or preserving control of the opponent, escaping or preventing the opponent’s control, and shaping the other’s future options through grappling mechanics. A submission relation exists where one participant’s policy creates or maintains a process capable of producing injury, physiological incapacity, or a recognised concession based on the credible expectation of either; that relation is mechanical and causal, and it is distinct from whether a ruleset ends the contest the moment it is recognised. And a recognition convention decides what counts as decisive, which is what lets catch-and-release, positional rounds and submission-recognition formats be grappling without a single tap.

That is why the same physics runs in every room. A retention drill, a positional round from a half-passed guard and a match in which the guard is being passed contain the same physical entities and relations under different task, opponent and contest constraints, and practice design, pedagogy and curriculum sit downstream of the physics. If a drill lets the passer use only one route, it has removed a constraint the match will restore, and the coach should know which one. The same discipline lets the ontology describe a grappling interval inside a mixed contest, for as long as contact, control, transition and termination are what govern events, without pretending to be a theory of striking or of the whole fight.

Drilling, situational practice, constrained sparring, open sparring and competition can instantiate the same physical entities and relations under different task, opponent and contest constraints. Practice design, pedagogy, session structure and curriculum remain downstream unless they encode a physical-state, information, capability, rules or selection claim.

Rules change the value of a state, never its mechanics

A straight ankle lock has the same mechanics in every gym in the world. Whether it is legal, whether the reap that leads into it is legal, whether the takedown before it scored, and whether the clock is running out are facts about a contest, and the ontology gives the contest a structure of its own: what is legal or prohibited and what is penalised; what value is assigned to states, trajectories, durations, events and accumulated score; what formally ends the contest or is recognised as an event, from a tap or a referee’s stoppage to time expiry, disqualification or a win on score; and how it starts and restarts, including overtime positions and out-of-bounds resets. Rules shape the task without rewriting the physics beneath it.

Three notions therefore stay apart. A control has a physical effect where it changes or preserves physical state. A policy is constitutively relevant where it materially alters or preserves the control and escape relations, the submission relations, or the capacity to interrupt or sustain them. And a policy has strategic value where, under these rules and this athlete’s objectives, it improves the expected outcome through scoring, time, risk or the future states it prefers. Strategic value is not part of the mechanical ontology at all, which is what explains a guard retention that is mechanically relevant and scores nothing, a low-risk stall that is strategically optimal, a heel hook that is physically possible and illegal, and an adjustment that is physically real and constitutively irrelevant. Physical possibility is not legality; physical effect is not scoring value; a mechanically stronger action is not automatically the strategically better one. Where the rules of a target contest enter the theory in earnest is on the competition readiness page.

Rules, scoring, legality, time and victory conditions can change the strategic value of a physically identical state or action without changing its mechanics. Physical possibility is not legality; physical effect is not scoring value; mechanical superiority is not automatically strategic superiority.

What each person can see, can do and chooses

Nobody in a grappling exchange has the whole state. Each participant has an observation relation, fed by vision, touch, proprioception, balance, hearing, the clock and score, and the referee’s signals, and it can omit a hidden grip, the opponent’s exact activation and fatigue, a limb that cannot be seen, and intent. Out of observations past and present, learned expectations and memory, each participant assembles an information state, which is where incomplete knowledge, hidden information, belief, perceptual delay and deception all live. The ontology does not require this to be a probabilistic model. It requires only that it exist, because hidden information, delay and deception are ordinary parts of grappling and need a formal home.

Being able to do something is a third thing again. Mechanical availability is not the same as learned ability, so the ontology gives each participant a capability specification of their own: technical repertoire, motor coordination, competence at switching policy, reliability under resistance, learned timing, motor economy, the capacity to run a coordinated branch. Acute restrictions such as fatigue, injury and the force left in reserve are deliberately kept out of it. They belong to the physical state, and the two are never counted twice. An athlete can reliably execute a policy from a configuration, over a horizon, at a declared standard of reliability, or they cannot, and what “reliably” should mean empirically, whether success rate, control error, performance against a class of responses, transfer or repeatability, is left open on purpose.

Even together, availability, capability and a recognised opportunity do not decide what a participant does. Selection depends on the information state, the family of policies this athlete can execute, the value the rules put on outcomes, and the person’s own preferences, risk tolerance and tactical objective. Two athletes with the same picture of the moment and the same executable options choose differently when one of them is ahead on points with a minute left. The sentence the whole agent level reduces to:

mechanical policy availability, executable capability, information and policy selection are different things.

Four facts inside one moment

The first of the ontology’s seven clauses fixes that distinction in four words that grappling language runs together whenever someone says “you had the armbar”. From mount, with the opponent’s arm isolated, the armbar may be physically available: the transition is possible from this exact arrangement of bodies. Whether this particular athlete can execute it, reliably, against resistance, is a second fact. Whether they recognised that it was there, in the moment, with their opponent moving, is a third. And whether they selected it, rather than keeping the mount and hunting the choke, is a fourth.

Keeping these apart is the single most useful diagnostic habit the theory offers. A coach who sees the missed armbar and says “take the armbar next time” has assumed that the problem was selection. It may have been recognition, in which case the athlete needs to see the arm earlier; or capability, in which case the athlete needs the mechanics under resistance; or the arm was never genuinely isolated and the armbar was not available at all. Each of those is a different week of training. The clause itself:

Physical availability means an action or transition is physically possible from the current physical state. Executable capability means the participant can realise it with sufficient reliability under relevant constraints and resistance. Observed or recognised opportunity means the participant's information state represents it as available or useful. Selected action means the participant actually chooses or instantiates it. None implies the others.

Names are compressions

Named positions, named techniques, systems, curricula and coaching methods are all real and all useful, and they are all derived: they are names people gave to recurring arrangements of something deeper. A coach who says “side control” has compressed a situation into one word, and the compression is useful, which is why every coach uses it. What the ontology adds is the state underneath the word: which connections each grappler has kept and which they have lost, what each of them can still reach, and what that leaves each of them able to do. Two situations that share the word can differ in that state, and the difference tends to show up in what follows.

The ontology also says how a name and the formal state relate, and the relation is looser than a dictionary’s. A formal position class is derived only from an exact equivalence between configurations, at a declared resolution and horizon, and conventional names sit over those classes in several ways: one name may contain several mechanical classes, several names may map to one, and a conventional category may be better treated as a cluster of approximately similar states. That keeps every word a coach uses while limiting what a word may imply. A technique is a learned executable family of policies, and invariants are derived as well.

That is a practical claim, and it changes what a coach can say. “You were in side control” names the arrangement; “your near elbow was captured and your hips were flat, so the recovery had become slow and easy to interrupt” names the state, and the second statement can be trained. Neither is wrong. The second is the one the rest of the theory works with.

This retains conventional vocabulary without treating it as primitive ontology.

What stays apart

Beneath the seven clauses runs a list of pairs the theory refuses to collapse. Each one is a place where ordinary grappling language uses a single word for two different facts, and where merging them produces confident wrong advice.

• Physical effect and strategic value. A sweep that turns the opponent over has the same effect under every ruleset; whether it scores, and what it is worth, is a separate fact.
• Reachability, enforceability and defensive viability. Being able to get to the back, being able to keep it against everything the opponent does, and being able to survive underneath someone who has it, are three different properties. The theory of control is built on this distinction.
• Opponent causal contribution. A back take that happened because the opponent turned away under pressure and one that happened because you forced the turn are different events, even when the final position is identical.
• Non-contest practice. Drilling and sparring are the same physics under different constraints, and not a different subject.
• Exact equivalence and approximate similarity. The word “kimura” is used for a grip, a control position, a sweep and a submission. Those situations share a hand configuration and differ in what the grip is being used to do: the target process and the consequence. The theory needs a more precise account of what is happening than the shared name gives it.
• Mechanical terminality and contest termination. A choke can be mechanically finished before the tap and tapped before it is mechanically finished. The two often coincide. When they do not, everything downstream depends on having kept them separate.

The list as the theory keeps it:

Physical availability, executable capability, observation/information, selected action, physical effect, strategic value, reachability, enforceability, defensive viability, opponent causal contribution, non-contest practice, exact equivalence versus approximate similarity, mechanical gating, and mechanical terminality versus contest termination remain distinct. Named positions, techniques, systems, curricula and coaching methods remain derived.

Why the types matter

Position, technique, mechanics, capability, perception, decision and rule value are seven different kinds of thing on this page, and the reason to keep them apart is practical. For diagnosis, a missed escape is a failure of availability, of executable capability, of recognition or of selection, and a position name says which of those it was no more than a street name says why a car stopped. For coaching, each of those failures wants a different week: mechanics under resistance for capability, seeing the arm earlier for recognition, a different choice under the same picture for selection, and for availability no drill at all, because the arm was never there. For technical explanation, a technique is a policy family rather than a shape, so explaining one means saying what it is conditioned on and where it branches. For development, what develops is the athlete: executable capability, recognition and selection are properties of a person, while availability is a property of the state, so a syllabus organised around what an athlete must become able to do is organised around the right kind of thing. And for ruleset transfer, mechanics and physical effect travel unchanged from one contest to another while legality and value do not, so what has to be learned again across rulesets is selection and valuation, and what does not is the physics.

Where the foundation stops, and the rule that keeps it honest

Pedagogy, institutions, lineage, curriculum, periodisation, governance and commercial organisation are outside the foundation, with one condition. They are outside it only while the claim being examined encodes no hidden statement about physical state, information, capability, rules or selection. A lineage’s teaching that “the hips must face the opponent in this guard” is a mechanical claim in pedagogical clothing, and the ontology treats it as one.

The rule that keeps this honest is that “out of scope” is a classification and never an escape. A hard case cannot be waved out of the theory to protect it, and a case declared outside the scope contributes nothing to the theory’s support.

Pedagogy, institution, provenance, curriculum, periodisation, governance and commercial organisation are outside the foundational ontology only when the assessed claim does not encode a hidden physical, informational, rules, capability or selection claim. OUT_OF_SCOPE is a scope classification, not an evidential refuge, and contributes zero positive universality evidence.

What the rest of the theory builds on this

Two things, mainly. The first is a description of a grappling state that does not depend on anyone agreeing what to call the position. That is what lets Adversarial Viability Control Theory define control as a relation between two people rather than as a label, and what lets the Control Grammar name the relations that make up such a state. The ontology already derives the relations that page works with, possible reachability, enforceability, defensive viability, interruption, response restriction and mechanical terminality among them, and hands them over defined. The second is the four-way split: because availability, capability, recognition and selection are four facts and not one, the capability architecture can say precisely what an athlete needs to become able to do, and the developmental layers can hold evidence about each separately.

That is the whole theory in miniature. Split what is usually fused, describe the state beneath the name, and every layer above gains something specific to say.

Open problems and standing

The account has been tested against a wide range of submission-grappling material and held: that is strong support across the tested domain, and it is not proof of completeness, so a future case the categories cannot describe would force a repair. What remains open is mostly measurement, and one question of parsimony. Knowing that recognition is distinct from capability is not the same as having a reliable way to observe either in a live round, and whether automatic organism response needs to remain a category of its own is undecided. The theory does not pretend otherwise. Its own list:

Automatic-organism-response minimality; empirical observation/information operationalisation; executable-capability operationalisation; response-restriction metrics; mutual or uncertain terminal races; interacting causal attribution; and human inter-rater validation remain open.