Jiu-Jitsu has no language.
We built one.

Positions are symbols. Connections are algebra. Videos become notation.

Closed guard position — real frame from ViCoS dataset
Raw frame
Closed guard with keypoint overlay
34 keypoints detected
Frame from the University of Ljubljana dataset.
Closed Guard
Human description

“Bottom player wraps both legs around the opponent’s torso and locks the feet.”

Algebra
CGRD := {
  ¬Z0(Op.Ba)
  CON(Me.Le+, Op.To, d1, −)
  CON(Me.Le−, Op.To, d2, +)
  CON(Me.Fo−, Me.Fo+, d3, 0)
}

Two legs control the torso. Feet form a closed loop.

Back control position — real frame from ViCoS dataset
Raw frame
Back control with keypoint overlay
34 keypoints detected
Same dataset. Same gym. Different position.
Back Control
Human description

“Control from behind. Legs wrap the torso. Feet do not cross.”

Algebra
BCTR := {
  FacingAligned
  ¬Z0(Op.Ba)
  CON(Me.Le+, Op.To, d1, −)
  CON(Me.Le−, Op.To, d2, +)
  ¬CON(Me.Fo−, Me.Fo+)
}

Same leg control on the torso. Different topology: no loop.

CGRD and BCTR share the same connections. They differ by one constraint:
CON(Me.Fo−, Me.Fo+, d3, 0) — feet locked — vs ¬CON(Me.Fo−, Me.Fo+) — feet open.
One line of algebra distinguishes two entirely different positions.

You just saw structure emerge from a single frame.

A language can describe.

The right language makes new things possible.

Roman numerals could represent numbers.
Arabic numerals made calculation possible.

BlawkOps does the same for Jiu-Jitsu.

From techniques to structure

Techniques

Jiu-Jitsu can be approached through sequences of moves.

Each technique is learned individually.

Techniques + Structure

Structure adds a layer underneath.

Techniques can be organised through invariants.

Knowledge transfers across positions.

This is an extension, a new layer that makes existing knowledge more powerful.

A logographic system

In logographic writing systems, symbols represent meaning directly. Instead of building meaning from sequences, structure defines meaning.

Jiu-Jitsu works the same way.

rest
person
+
tree

A person leaning against a tree. Rest.

Meaning emerges from structure. Strokes are production details. Radicals carry meaning.

In logographic writing

Characters share radicals.
You can read new words by recognising their structure.

In Jiu-Jitsu

Positions share invariants.
You can understand new positions by recognising their radicals.

This is how fluency emerges.

Decomposing a position

Closed Guard
CGRD

Radicals = what defines it

  • Legs control the torso
  • Feet form a closed loop
  • Opponent elevated

Lose a radical, lose the position.

Strokes = how you perform it

  • Grip on the collar or wrist
  • Angle of hips
  • Hand placement
  • Head position

Strokes matter for efficiency. Not for identity.

Radicals define the position.
Strokes define how efficiently you perform it.

What structure adds to teaching

Techniques

Sweeps, escapes, submissions — each learned as a sequence of steps.

This builds execution.

+
Structure

Positions, invariants, radicals — a stable vocabulary underneath the techniques.

This builds understanding.

Structure gives techniques a place to live.
Knowledge becomes transferable. New positions become readable.

This is an extension of how Jiu-Jitsu is taught — a new layer, built on top of existing practice.

A structural approach

1
Position

Can you name it? Mount, Guard, Back, Side, Turtle.

2
Radical

What makes it that position? Mount = hips over hips + knees inside.

3
Transition

How does one position become another? Which radical changes?

4
Technique

Details last. Grip, angle, pressure. Stroke order.

Example
MNT

“Hips over hips. Knees inside the elbows. Opponent’s back on the ground.”

High Mount, Low Mount, Grapevine Mount are the same character in different fonts.

Striking
Alphabetic

Letters sequenced into words.
Jab → Cross → Hook.

vs
Jiu-Jitsu
Logographic

Characters recognised by structure.
Guard. Mount. Back.

Position = unit of meaning
Radical = structural invariant
Stroke = execution detail

One structure, multiple positions

DLR
CON(Me.Le−, Op.Le+, d, )
Left leg hooks opponent’s right leg. Wrap direction: negative.
helicity flip
SLX
CON(Me.Le−, Op.Le+, d, +)
Same leg, same axis. Wrap direction: positive.
axis substitution
LSSO
CON(Me.Le−, Op.Ar+, d, +)
Same wrap direction. Target moves from leg to arm.

Three positions. One structure. Two transformations.

This is how structural understanding works.

The system

Image
Keypoints
17 joints × 2 athletes
Connections
CON tuples from proximity
Frame
Facing, ground, brackets
Radical
Bottleneck scoring
Position
CON is the only primitive relation. A typed connection: attacker wraps axis, with depth and helicity. Everything else is derived.
Frame constraints encode spatial geometry: who faces whom, who is grounded, knee brackets. They narrow the match without adding new primitives.
Cycles and closures distinguish positions that share connections. Closed guard = feet locked (h=0). Back control = feet open. One line of algebra, completely different position.
Temporal persistence smooths frame-by-frame noise. 8-frame persistence + closure memory. Label flicker: 0.21 → 0.01 switches per frame.

Data & validation

120,279
labelled frames
34
keypoints / frame
12
radicals defined
1
primitive

Built on the Brazilian Jiu-Jitsu Positions Dataset from the ViCoS Lab, University of Ljubljana. 10 combat positions, 18 classes, MS-COCO keypoint annotations.

Hudovernik & Skocaj (2022). Video-Based Detection of Combat Positions and Automatic Scoring in Jiu-jitsu. MMSports’22, Lisbon. CC BY-NC-SA 4.0.

The first empirical validation of a symbolic BJJ language.
No neural classifier. Positions recognised from algebra alone.

What becomes possible

Once Jiu-Jitsu has a language, we can:

Translate positions directly from video

Any match becomes a typed sequence of symbols. Searchable. Comparable. Storable.

Compare techniques structurally, not stylistically

Not “this looks similar” — algebraically identical, or different by exactly one operation.

Teach invariants instead of memorising sequences

Learn 12 radicals. Recognise every position built from them.

Discover equivalences across positions

DLR and SLX are the same structure with opposite helicity. The algebra proves it.

Build machine-readable grappling

Automated scoring. Training analytics. Opponent scouting from competition footage.

A formal representation makes all of this possible.

Positions are objects. Transitions are transformations.

In mathematics, this system is a category.

Positions are objects. Transitions between positions are morphisms.
A match is a path through a space of structures.

DLRSLX
Helicity flip
Same structure, different helicity.
SLXLSSO
Axis substitution
Same structure, different attacking limb.
MNTBCTR
Frame change
Same connections, different facing.

Each pair is the same object under transformation.

This is how experienced practitioners see connections across positions.
They are operating on structure.

About

BlawkOps was not built in a laboratory.

It came from conflict, obsession, fatherhood, mathematics, and years of trying to understand violence without glorifying it.

I grew up tasting violence early. Later, mathematics became the language through which I tried to understand the world: abstraction, structure, geometry, systems, and invariance.

Then I became a father.

My son changed the direction of everything.

I walked into the gym and onto the mats because of him — to learn how to protect, to guide, and to teach discipline, calmness, responsibility, and control before fear or chaos could teach harsher lessons instead.

Brazilian Jiu-Jitsu became more than a martial art. It became a study of pressure, leverage, timing, asymmetry, adaptation, failure, and patience.

The struggle of learning it as an adult — the humiliations, repetitions, injuries, confusion, and rare moments of clarity — pushed me toward building a symbolic framework for understanding it differently.

BlawkOps emerged from that pursuit.

A language for grappling. A mathematical notation for positional relationships, transitions, orientation, control, movement, and structure. An attempt to bridge martial arts, geometry, category theory, systems thinking, and computation.

The ambition was never only to teach fighting on the mat.

It was to make it possible to teach aspects of fighting on a whiteboard — to analyze movement as structure, and to transmit intuition through symbols, diagrams, abstractions, and systems without losing the human reality underneath them.

This project would not exist without the people who shaped the journey.

To my mentors, B and A: thank you for the patience, knowledge, guidance, support, and countless hours shared on the mats. A mentor transmits far more than technique. They transmit discipline, character, perspective, and resilience over time.

To E.C.: thank you for the inspiration that abstract mathematics and category theory can illuminate structures far beyond pure theory.

And to my partner in life — the real fighter — thank you for carrying burdens no one sees, for the quiet strength behind the scenes, and for standing beside this journey from the beginning.

Above all, this project is for my son.

Everything transmitted across generations begins as an attempt to protect something you love.

We are defining a language for Jiu-Jitsu.