Portfolio · Biological Architectures
The Tardigrade
The minimum biological case exhibiting non-linear state resolution by settling rather than branching. Eight legs, two eyes, four ganglia, one shared neuropeptide field. Twelve levels of cognition stacked on the same primitive, 12/12 pass, no training loop.
The Tardigrade
A geometric nervous system that walks, turns toward light, and climbs a fourteen-level tower of cognition — without a single branching decision at the engine.
The Question
Two animals sit either side of an ancient fork in the tree of life. Both are Ecdysozoa. Both molt.
- C. elegans took the microchip path. Three hundred and two neurons, mapped nerve-by-nerve since 1986. A graph. Efficient. Rigid.
- Tardigrades kept the ancestral complexity. A ganglionic brain. A neuropeptide bath between the cells. A graph and a field.
The question this experiment tries to answer: what does the field give you that the graph alone cannot?
The Answer, Short Version
Under the VINE paradigm, the field gives you non-linear state resolution — the ability for identical inputs to produce different outputs depending on where the system has been. Branching without branches. Decisions by settling.
The tardigrade here is the minimum biological case that exhibits this.
Eight legs, two eyes, four body ganglia, one shared neuropeptide field.
No if, elif, or else at any decision surface. Every commitment is
a basin in geometric space, found by geometric settling.
The Architecture
BRAIN
│
┌──────────┼──────────┐
│ │ │
G1 G2 G3 G4
┌─┴─┐ ┌─┴─┐ ┌─┴─┐ ┌─┴─┐
L1a L1b L2a L2b L3a L3b L4a L4b
neuropeptide field
— shared drift, no glass —
Three structural commitments, each a VINE primitive:
- Glass connections. A signal transmitted one-way, carrying its primary value alongside a silent state dimension that accumulates geometric history over time. One-way, like an ion channel. The silent dimension never contributes to the output; it is pure state, carried forward.
- The VINE perceptron. A single-layer geometric unit that
resolves non-linearly separable inputs not through hidden layers but
through geometric settling on a bounded manifold. Decisions are
basins found by settling — not branches written in advance. This
is the VINE replacement for
if/elif. - The field. A bath. No glass. Every ganglion speaks into it; every ganglion feels it. The attractor is a specific mathematical constant.
What Emerges
Three behaviours were tested. None were programmed. Each is an emergent property of the geometry under stimulus.
1. Gait — no walk command given
A constant bilateral light is fed to both eyes. No command says "walk". Yet within fifteen timesteps, the legs settle into the tardigrade's natural gait: paired alternation, anterior-to-posterior wave.
From the clean re-run:
Leg pair alternation: 100%
Anterior→Posterior wave: ✓ YES
Four pairs of opposite-sign activations, with amplitude decaying from G1 to G4. The field, once excited, produces the walking rhythm as its natural resolution.
2. Phototaxis — no turn rule given
Light applied to the left eye only. The bilateral differential flows
through the ganglion chain as a lateral bias. Left-side legs push
harder; the body turns right; the tardigrade moves toward the light.
No if left > right anywhere in the loop. The geometry of the
differential is the decision.
| Scenario | Left legs | Right legs | Result |
|---|---|---|---|
| Light LEFT only | 1.08 | 0.56 | → toward L |
| Light RIGHT only | 0.56 | 1.08 | ← toward R |
| Equal light | 0.83 | 0.83 | STRAIGHT |
3. Behavioural modes — no category code
Brain output scales with total drive. The mode label (IDLE, ASSESS,
APPROACH, AVOID) is produced by asking which named basin the scalar
is closest to on a one-dimensional manifold. That is the VINE
replacement for the if/elif reporting chain you would otherwise
write — a single argmin over distances, no branching logic.
The Tower
A further twelve cognitive levels are stacked above the tardigrade's base behaviours. Each level adds one capability, each level tested with a pass condition, each level running under the same geometric engine. The full walk, re-run on the cleaned code, returns 12/12 passing on the first attempt:
| Level | Capability | Result |
|---|---|---|
| 3 | Concept formation | pass |
| 4 | Conjunctive features | pass |
| 5 | Depletion (adaptive unlearn) | pass |
| 6 | Trajectory (transformation) | pass |
| 7 | Universal rules | pass |
| 8 | Planning (trigger & wait) | pass |
| 9 | Multi-step sequences | pass |
| 10 | Resource hoarding | pass |
| 11 | Object permanence | pass |
| 12 | Environmental signs | pass |
| 13 | Shelter & weather | pass |
| 14 | Emergent tool use | pass |
Selected phrases from the run log:
✓ SUCCESS! Learned object permanence: deposited food PERSISTS
✓ SUCCESS! Learned to hoard: pick during abundance → survive scarcity
✓ SUCCESS! Emergent tool use observed: collision physics pushed food into zone
Each of these is a behavioural capability a standard reinforcement agent would take many hours and many reward-shaping passes to reach. Here, each is a re-use of the same VINE perceptron settling step with a different basin topology. No training loop. Change the basin; change the behaviour.
What This Proves, What It Does Not
It proves: A single geometric mechanism — the VINE perceptron settling step — scales from leg rhythm to object permanence without qualitative change. The same arithmetic handles phototaxis and shelter-seeking. A new level costs hours, not a re-architecture.
It does not claim: That this specific demonstration produces the XOR basin separation as a single scalar. The canonical XOR proof is held separately — a single-layer perceptron, two scalar weights, one epoch, verified against published results. The tardigrade inherits the same perceptron but uses it to coordinate a body, not to pass a logic test.
It does not require: Training data. A loss function. Backpropagation. Gradient descent. Any of the standard neural-network furniture. The system has a personality (fixed random weights), a memory (accumulated potential), and a drift (settling step). Nothing else.
Why a Tardigrade
Because the tardigrade is the smallest case where the answer to the opening question is yes — field geometry gives you the behaviours a graph cannot. C. elegans, stripped of the field, sits at the floor and cannot solve XOR. The tardigrade, with field intact, climbs fourteen levels. The difference is the bath.
The rest of the paradigm follows. Every VINE module — the ERP, the language platform, the cyber threat detector, the horse-care system just commissioned — runs on the same primitive. What you can build for a tardigrade you can build for anything that decides.
Raychell Langan · NEXICOG Ltd · Hampshire, UK