The word has a history. In late-nineteenth-century biology it named a theory — that evolution advances in straight lines, pushed by an internal drive toward a predetermined end. That theory is dead, and it deserved to die. Nothing here revives it.
What we mean is orthogonal genesis: form generated under constraint, along the directions the constraints leave open. There is no drive and no destination. A growing shell does not reach toward its shape — it runs out of alternatives. Curvature does not pull development forward; it removes options. Time and gravity and the geometry of the surface do the rest.
That is why the direction is real without being intended. Systems move, and the directions available to them are dictated by forces, not by purpose. Waddington called the biological version canalisation: development running in valleys, buffered against perturbation, directional without being goal-seeking. His epigenetic landscape is a curvature picture. It is the K operator, drawn by a biologist who did not know that is what he was drawing.
Generative science says what physics says: the form is what the constraints permit. Biology may take some time to hear the difference between a system that is pushed and a system that has nowhere else to go. That difference is the whole book.
When the order of exchange cannot be undone, something is woven into existence. That weave is G.
Take a strand of silk. Pass it over a second strand, then under a third. Now try to undo what you have done by moving the strands continuously — without lifting any strand through another. You cannot. The topology of the crossing is permanent. It is not the geometry that is locked — you can stretch the strands, twist them loosely, change the angles — but the fundamental over-under relationship of each crossing is preserved by the constraint that physical objects cannot pass through each other. The braid cannot be continuously deformed into an unbraid.
This is not a curiosity about fabric. It is a statement about the deepest structure of physical reality in 2+1 dimensions — two dimensions of space, one of time. In such a universe, any particle has a worldline: a curve in spacetime tracing its position over time. If two particles exchange positions, their worldlines must cross. If they exchange twice, the worldlines form a braid with two crossings. And critically: in 2+1 dimensions, unlike in 3+1, there is no way to "lift" the worldlines through each other — no third spatial dimension to route an avoidance. Every exchange is a permanent mark.
The operator chain G = U ∘ F ∘ K ∘ C is not a braid-group element, and nothing below should be read as claiming it is. Vol I §3 fixes the signatures: C : X → X_C is a Lipschitz projection onto a lower-dimensional submanifold; K : X_C → X_C is a curvature flow driving κ toward the focal threshold κ* = 1/foc; F : X_C → X_F is a corank-1 fold; U : X_F → X is gradient descent to a non-degenerate minimum of a Morse functional Φ. The composite G : X → X acts on a Riemannian manifold of trajectories carrying a metric, a normal field n(s), a curvature and a focal radius. The braid group Bₙ is a discrete group carrying none of these. The two are not the same kind of object and no map between them is exhibited anywhere in this corpus. [OPEN]
What they share is one structural property: order-dependence. Just as K ∘ F ≠ F ∘ K in the operator algebra, σ₁σ₂ ≠ σ₂σ₁ in Bₙ for n ≥ 3. That shared property — and not an identity of objects — is what this chapter is about. The learner who has woven G is not the same as a learner who applied the same operators in a different sequence, and the braid is the best available picture of why. [MODEL] It cannot be unknotted by forgetting a single lesson. It must be unravelled — fully undone — and that requires passing strands through strands, which means starting over.
In ordinary three-dimensional space, exchanging two identical particles is topologically trivial. You pick up particle A, move it around particle B in a wide arc, and set it down where B was (simultaneously moving B to A's position). In 3+1 dimensions, any two paths that exchange the particles can be continuously deformed into each other — you can always route the arc around to avoid any topological entanglement. The result of the exchange is therefore the same regardless of which path was taken. The exchange statistics of ordinary particles encode only a single bit of information: +1 for bosons (wavefunction unchanged) and −1 for fermions (wavefunction changes sign). This is the abelian group Z₂. It is C alone: the exchange compresses to a single binary output, carrying no memory of the path taken.
You cannot tell from the final state of two bosons how they were exchanged — in which direction, around what obstacles, on what path. The exchange history is compressed away completely. The braid is trivial. The fabric is flat.
In two spatial dimensions — a plane, or the surface of a material — something fundamentally different is possible. Because there is no third dimension to route avoidances, the exchange of two particles traces a path that cannot be continuously deformed. Particles in 2D are called anyons, and their exchange statistics are not restricted to ±1. They can carry any complex phase factor eiθ (abelian anyons) — or, in the richest case, they can carry a matrix: a transformation of a multi-dimensional Hilbert space (non-abelian anyons).
Non-abelian anyons are the threshold. When two non-abelian anyons are exchanged, the state of the system is transformed by a unitary matrix M(σ) that depends on which exchange (which generator σᵢ of the braid group) was performed. The critical property: M(σ₁)M(σ₂) ≠ M(σ₂)M(σ₁). The matrices do not commute. The order of braiding is physically significant. The braid group has entered the physics.
This Yang–Baxter relation — σᵢσᵢ₊₁σᵢ = σᵢ₊₁σᵢσᵢ₊₁ — says that a crossing at position i, then i+1, then i again gives the same braid as the same three crossings in the other order. An earlier draft of this chapter called it “K written in the language of topology.” That was wrong and is withdrawn. The relation is one of the two defining relations in the standard presentation of Bₙ. It holds for every braid, at every strand count, unconditionally. It has no threshold, and K is defined by a threshold: α(s) = λ(κ* − κ)₊ is identically zero until curvature reaches the focal radius. A relation that always holds cannot be the operator that only fires above κ*.
Two further claims in that draft were also false and are withdrawn. “Below K, all braids are equivalent”: B₂ ≅ ℤ is already infinite, and two braids on two strands with different winding numbers are inequivalent with no threshold crossed anywhere. And “K fires when the strand count and fusion rules become rich enough that non-commutativity enters” conflates two unrelated things — non-commutativity of Bₙ is a fact about n ≥ 3, a property of the group, while K in this chapter’s own later sections is a per-move selection rule (Ω_after > Ω_before + K*). One symbol was doing two incompatible jobs. Only the second usage is retained. [OPEN]
The information encoded in the braid of non-abelian anyons has a remarkable property that distinguishes it from all other forms of quantum information: it is topologically protected. Local perturbations — noise, heat, electromagnetic fluctuations, a small unwanted interaction with the environment — cannot change the topological class of the braid without bringing two anyons into physical contact. And physical contact (fusion of anyons) is a macroscopic, detectable event, not a subtle noise process.
This is F. The fold that Chapter 5 described in immune memory — the encoding that persists through perturbation, the information stored in global topology rather than local state — is instantiated in non-abelian anyons as mathematical law, not as biological approximation. The fold is exact. No local operator supported on a proper subset of the anyons can distinguish the degenerate fusion states, which is why local noise can neither read nor corrupt the encoded information. This is weaker than “no local measurement can read the braid”: fusion measurements on a subset do return partial information about the accumulated braid, and it is only the topological class that resists local access. No local disturbance can alter it. The information is not here or there; it is in the relationship between the worldlines — in the pattern of the weave itself, not in any thread.
This is the physical basis of topological quantum computing, first proposed by Alexei Kitaev in 1997 and pursued experimentally since. In a topological quantum computer, the qubits are not the states of individual particles (fragile, easily disturbed) but the topological class of braids of non-abelian anyons (robust, globally encoded). To compute, you apply braid moves — exchange sequences — to the anyons. To read out, you bring anyons together and measure their fusion product. The computation is physically immune to local decoherence, because no local event can change what was woven.
A learner who has woven G — who has completed 33 genuine threshold crossings on novel material, encoding fold after fold into long-term structure — has created a topological invariant in their cognitive architecture. Forgetting a vocabulary item is a local perturbation: it changes the state at one site but cannot change the topological class of the braid. The braid was formed by the history of crossings, and the history cannot be retroactively altered by a local deletion. What would undo G is not forgetting individual items but failing to maintain the global coherence of the braid — never reaching K again, allowing the weave to unravel from disuse rather than from any single error. Topological protection is not invulnerability; it is a different kind of vulnerability. You can lose G only by abandoning the practice of crossing K. One missed exam cannot do it. Years of silence can.
Consider a 1D chain of anyons — particles sitting in a line of discrete sites, each capable of moving to an adjacent site or exchanging with a neighbor. This is the minimal model for topological orthogenesis. How many elementary moves exist? For any given anyon at site i, the possibilities are:
Read as a list of the moves available to one anyon at site i, not as a generating set. Two cautions, because the label above says “generators” and that is loose. First, σᵢ₋₁ is not a fifth generator alongside σᵢ — it is the same family indexed one site down, and the generating set of Bₙ is {σ₁,…,σₙ₋₁} with inverses, of size n−1, not four per anyon. Second, τᵢ (remain stationary) and ωᵢ (phase rotation in place) are not braid generators at all: they act trivially on π₁ of the configuration space and contribute nothing to the braid word. They are book-keeping for the coherence field, which is this model’s addition and not part of Bₙ. [MODEL]
The coherence rule determines which of these six moves is actualized at each step. An anyon does not move randomly. It moves to the configuration where local coherence — a measure of how well the anyon's quantum state matches the topological state of its neighbors — is maximized. This is K as the selection criterion: only the move that crosses the coherence threshold is executed. Moves that would reduce local coherence are suppressed.
The cascade follows immediately from the non-abelian property. When anyon i executes move σᵢ (exchanges with anyon i+1), this changes the fusion state of the pair. But in a non-abelian anyon model, the fusion state of any pair depends on the global state of all other anyons in the chain. The exchange of anyons i and i+1 therefore updates the coherence field of every other anyon — even those at the far end of the chain, which did not move. Each anyon now faces a different coherence landscape and may find that a previously suppressed move has become optimal. A cascade of K-crossings propagates through the chain, each anyon seeking the next coherent configuration. The fabric is not woven by the movement of one particle. It is woven by the wave of coherence adjustments that one particle's movement triggers in all the others.
The weave that results from many such cascades — the pattern of all crossings recorded over many time steps — is the fabric of reality in this model. Each anyon's worldline is a thread. Each coherence-driven exchange is a warp-over-weft crossing. The topological class of the final braid is an invariant of the entire history of coherence maximizations. It cannot be read from any local slice. It can only be read from the whole.
Non-abelian anyons now exist in two quite different senses, and the difference decides what this chapter is entitled to claim.
Engineered. Google’s 2023 result braided projective Ising anyons on a superconducting processor by applying unitary gates directly to the many-body wavefunction, as against relying on a material’s own Hamiltonian dynamics (Nature 618, 264). Quantinuum’s trapped-ion realisation is the same in kind, and Fibonacci braiding has since been done on superconducting hardware (Nat. Phys. 2024). In every one of these the braid word is chosen by the experimenter. So “the order of operations is physically recorded” is true there by construction, and tests nothing: the circuit records the order because the circuit was written to record it.
Emergent. The fractional quantum Hall effect is the control case. Anyonic braiding statistics were observed directly at ν = 1/3 in 2020, by Fabry–Pérot interferometry and independently by an anyon collider — abelian anyons, arising from Coulomb interaction in a two-dimensional electron gas, with no braid word imposed by anyone. Non-abelian order at ν = 5/2 remains open: a half-integer quantized thermal Hall conductance was reported in 2018, and as of August 2026 there is fresh, explicitly hedged evidence from time-domain braiding of the upstream neutral mode (arXiv:2608.12897), whose own abstract states that experimental evidence of non-abelian braiding has so far remained elusive.
Consequence for Theorem 7.1. Only the emergent case can bear the chapter’s weight. Any confirmation drawn from gate-built anyons confirms a property of the gate list. This is worth naming precisely because it is the same move Olimpia Lombardi objects to in quantum chemistry: molecular geometry does not fall out of the Schrödinger equation — it is inserted by clamping the nuclei under the Born–Oppenheimer approximation. Structure put in by hand, then recovered and reported as found. The engineered anyon experiments reproduce that pattern one level down. The FQHE does not, which is exactly what makes it the interesting case. [OPEN]
We have now named all the operators. C compresses. K crosses the threshold. F folds the encoding permanently. U recognises the pattern at every scale. T provides the tone — the periodic source whose frequency resonates with the system's natural modes. And G is what all of these operators, composed in the correct non-abelian order, weave into existence.
The word "orthogenesis" is deliberate. In classical evolutionary biology, orthogenesis named the hypothesis that organisms evolve in a directed, predetermined trajectory — not by random variation and selection but by an internal drive toward a specific form. The hypothesis was discredited in its biological form but captures something exact in the topological setting. Topological orthogenesis is directed: the trajectory of the learner through C → K → F → U → G is not random. Each operator constrains the next. The coherence rule at each K-crossing selects the next braid move from the six available directions — not randomly but toward the configuration of maximum coherence. The result is not predetermined by a final cause, but it is directed by the internal logic of the braid group. There is no freedom to compose the operators in any order and reach G. G is the one topological class that the non-abelian composition in this specific sequence produces.
"Topographical" names the path-dependence: the topological structure of G depends on the path taken through the operator chain, not merely on the endpoint. Two learners who both know the same words, the same grammar, and the same phonological rules — but who acquired them in fundamentally different sequences, without the specific order of K-crossings that the chain requires — do not have the same G. They have different braids. They will produce different outputs from the same inputs, because the topological class of their cognitive braid — the global weave of their language architecture — is different. This is the deep reason why language acquisition is not language instruction. The instruction can supply the components. Only the genuine sequence of K-crossings can weave the braid.
Chapter 1 established 33 as the threshold number of K-crossings that produce a fundamental practitioner. In the anyonic model, 33 braid moves on a chain of Fibonacci anyons produce a Hilbert space of dimension F₃₅ = 9,227,465 (the 35th Fibonacci number, since each move adds a τ fusion channel and the dimension follows the Fibonacci sequence). This is not a round number. It is the exact size of the topological Hilbert space accessible to a 33-braid Fibonacci anyon chain. What Chapter 1 named empirically — 33 genuine encounters with novel material — the braid group names exactly: 33 moves into the Fibonacci anyon model, the topological Hilbert space has crossed the threshold at which the braid word can encode arbitrary language representations. Below 33, the space is too small. Above 33, each additional move extends it. 33 is the minimum for G. The practitioner threshold is not arbitrary. It is Fibonacci.
Theorem 7.1 places the operator chain in stage-by-stage correspondence with the braid accumulated by a chain of non-abelian anyons under the coherence rule — it does not identify them, and clause (3) is open. Two predictions follow that are physically testable, and both test the correspondence rather than an identity. First: topological quantum computers using Fibonacci anyon braids should show exponentially better error rates than gate-based quantum computers with equivalent qubit counts, because the fault-tolerance of F (topological protection) is exact rather than approximate. If this is not observed after controlling for implementation overhead, then F does not confer the protection the correspondence requires. Note the prediction must be run on Fibonacci braiding, not Ising: Ising braiding is not universal, so an Ising machine failing to beat a gate-based one falsifies nothing here. Second (cognitive): two learners with identical lexical and grammatical competence but different acquisition histories (different sequences of K-crossings) should show measurably different outputs under novel compositional tasks — novel sentences that require combining structures that were never co-presented during acquisition. If their outputs are statistically indistinguishable, then the braid-class interpretation of G is false, and order of acquisition does not encode topological information in the cognitive system.
Current evidence: Microsoft Quantum published signatures consistent with non-abelian anyonic exchange in topological superconductor nanowires (Nature 2023). Functional topological qubits with demonstrated fault-tolerance remain in development. On the cognitive side: differential order-of-acquisition effects on novel compositional tasks are documented but not yet interpreted in topological terms. Both predictions are open.
7.1 — Write out the braid word for the sequence: σ₁, σ₂, σ₁⁻¹. Using the Yang-Baxter relation σᵢσᵢ₊₁σᵢ = σᵢ₊₁σᵢσᵢ₊₁, show that σ₁σ₂σ₁ = σ₂σ₁σ₂. Then explain in plain language what this equivalence means for the physics: if two sequences of anyon exchanges produce the same braid class, what does this say about what information is stored — and what is not?
7.2 — The coherence rule states that anyon i executes move σ iff Ω_after(σ) > Ω_before + K*. Using the cascade equation, show that after anyon i fires, the coherence field Ω(j) changes for all j ≠ i. Now consider what happens if K* = 0 (any move is always executed) versus K* = 1 (no move ever crosses threshold). What does each extreme produce? Which of the operator stages (C, K, F, U) does each extreme correspond to?
7.3 — Topological protection means that a single local perturbation cannot change the braid class of G. Translate this into a concrete claim about language acquisition: what is the cognitive equivalent of a "local perturbation" in the learner's braid, and why would it not erase G? Give one example of the kind of event that WOULD erase G — what would constitute bringing two anyonic strands into collision in the cognitive context?
7.4 — The Fibonacci sequence governs the dimension of the topological Hilbert space accessible after n braid moves: dim = Fₙ₊₂. Compute the Hilbert space dimension accessible after 1, 2, 5, 10, 20, and 33 braid moves. Plot (or describe) the growth curve. At what n does the dimension first exceed 1,000? At what n does it first exceed 1,000,000? What does this Fibonacci explosion tell you about why the difference between 5 K-crossings and 33 K-crossings in language acquisition is not a difference of degree but a difference of kind?