The capability architecture

The unit of organisation

A syllabus organised by position lists what can be done from each place. The same control problem can appear in several of those places, and a syllabus organised by position will list it several times under different names, or miss that it is one problem at all. A technique catalogue answers “what can I do from here”. A capability architecture answers “what am I unable to do yet, wherever it comes up”, and those two questions produce very different maps of the same sport.

The architecture is a reference and coverage graph of those problems. It is deliberately not a coaching sequence and not a priority engine: it says what exists to be developed, and leaves the decision about what this athlete works on next to the coaching method.

Ten families

The capabilities group into ten families, and every family is derived from a structure in the Control Grammar. What each is derived from is kept visible, because a mechanical relation, a derived diagnostic, a dynamics annotation and a game predicate are different kinds of thing, and flattening them would lose the distinction the grammar exists to make.

• Establish and Deny Access from a primary mechanical relation: ACCESS Getting a hand to the far hip before the pass, or denying the choking hand a path to the neck.
• Preserve and Reduce Connectivity from a primary mechanical relation: CONNECTIVITY Isolating an arm from the torso for the armbar, and keeping your own elbow connected under mount.
• Create, Preserve and Break Useful Coupling from a primary mechanical relation: COUPLING Making a body lock that moves their hips when yours move, and breaking a seatbelt before it becomes one.
• Navigate Dynamic Transitions from a dynamic state annotation: DYNAMICS Winning the scramble after a failed shot rather than freezing in it.
• Build Enforceable Transitions from a game predicate: REACHABLE and ENFORCEABLE Turning a back take that is merely reachable from the front headlock into one you can force.
• Preserve and Deny Reorientation from a derived diagnostic: task-relative orientation and reorientation, read through access, connectivity, dynamics, mechanism and viability Denying the re-face from back control, and keeping your own turn alive under side control.
• Restrict and Interrupt Response from a game predicate: RESTRICTED and INTERRUPTIBLE Taking away their meaningful answer to the pass, and breaking into their choke before it takes effect.
• Organise Support and Load from a primary mechanical relation: SUPPORT-LOAD Loading a passer onto their own front knee, or restoring your own posts under a pin.
• Create and Escape Terminal Processes from a game predicate, with its typed endings kept distinct: TERMINAL, with the typed termination events kept distinct Building a finish that no timely defence survives, and escaping before that point arrives.
• Preserve and Restore Viability from a game predicate: VIABLE Staying non-losing under a heavy pass, then turning the recovered frame into a guard.

Fifty problems

Inside the ten families sit fifty capability nodes. Each carries a title, the target problem it names, the family of opponent responses it has to survive, the surfaces it appears on, and its place in the minimum complete game. They are connected by fifty-one prerequisite relations and twenty-five co-development relations, which say which problems have to be solved before others and which develop together.

“Universal core” has a precise and modest meaning here: a shared reference structure of capability and coverage, common to every grappler. It does not mean an identical sequence, an identical emphasis, an identical policy network or an identical game. Two athletes can satisfy the same fifty problems with technical solutions that share almost nothing.

Two axes: the problem and the surface

The architecture also carries seventeen surface domains as a first-class layer: the places grappling happens as people actually talk about it. That gives two organising axes that must not be confused. A position family is not a capability family. Any correspondence between the two is a transformation, and a map that merely renames a position family into a capability family has collapsed the distinction.

The minimum complete game runs across both axes as a coverage floor rather than a priority order: thirteen mappings that say which problems must not be starved in broad development, and not which ones an elite competitor should work on this month.

A reference, not a sequence

The graph carries six developmental stage labels, from orientation through to state engineering, and they are descriptors of resolution rather than a compulsory order. A capability can be trained and assessed at whatever resolution the athlete and the safety constraints permit, and stage ranges are guidance about where that usually sits, never a bar. What actually gets worked on is a projection: the coach selects an active subset of the graph for one athlete or one cohort, at a chosen resolution, based on the athlete, the opponent, the ruleset, the environment and the time horizon, without rewriting the graph underneath.

That is the difference between a syllabus and a reference. A syllabus tells you what week three contains. A reference tells you what exists, so that week three can be chosen for the person in front of you.

From a failure to a capability

The grammar turns a failed action into a failure signature. The architecture lets that signature retrieve candidate capabilities: an armbar that failed because the arm was never isolated and one that failed because the stack arrived first point at different families of candidate nodes. Retrieval is not selection. A signature may surface candidates; it never bypasses the coaching method’s judgement about which of them is worth this athlete’s next month.

Evidence about a capability is kept in five dimensions: recognition, execution, adaptation, retention and transfer, with cross-context transfer as its own question. The architecture claims no universal threshold on any of them. How much transfer counts as enough is a coaching judgement, and later an empirical one, not a number the graph supplies.

Attachments and exemplars

A capability on its own is not teachable. Worked exemplars attach to capabilities through a separate object, and the attachment carries the pedagogical role the exemplar is playing: a primary seed, an alternative, a response branch, a recovery, a transfer exemplar or a context variant. Six roles, and being the primary seed means only that it is the initial worked example when seeding is chosen. It does not make it canonical, universal or currently preferred. A reference exemplar may resolve as a single choice, as a set partitioned by context, or as several co-preferred options, and every one of those is replaceable.

An exemplar is identified by the policy it expresses rather than by its name: the state you enter from, the control proposition, the mechanism where it matters, the execution signature, the response family and the exit. Two techniques sharing a name with different policies are different exemplars; two names for one policy are one exemplar. That is how the architecture avoids becoming a technique list wearing a new vocabulary, and the retained rule underneath it is blunt: a capability requirement is not a technical exemplar, and there is no universal technique set.

Which exemplar an athlete gets is a selection, and the inputs to it are explicit: morphology, current capability, ruleset, opponent, environment and time horizon. Two athletes satisfying one capability may do it with different exemplars, chosen for different bodies, opponents and rulesets, and the capability is the same. The networks of exemplars and their links are local implementation systems. They are never a second syllabus, and structure that starts to look universal is a signal to review the reference above it, not to promote the network.

What the exemplar library is

The reference library that populates the architecture holds sixty-one exemplars and sixty-three attachments, and every one of the fifty nodes has at least one. Those are coverage figures. Being in the library establishes nothing about pedagogical superiority, athlete mastery, optimal sequence or a universal canon, and the counts are a description of what has been populated rather than a target to optimise.

Over the library, the architecture explicitly permits derived views: orders of operation, dilemmas, branches, recoveries and transitions drawn over exemplars and their links. The position map on this site is a shape of that kind, organised by position and transition. That is an observation about shapes and not a correspondence between entries, and no such correspondence has been written.

What this changes

For diagnosis, it changes the question from “which move did they get wrong” to “which problem can they not yet solve”, and the second question has an answer that can be looked for across surfaces. For coaching, it separates the problem from the solution: the method selects a capability to develop and only then an exemplar to develop it with, chosen for this athlete’s body, opponents and rules. For development, it is the vocabulary in which the curriculum states its requirements and the evidence is kept: not “knows forty techniques” but “can restore connectivity under load, on four surfaces, against resistance”.