The ISA Opcodes

Five opcodes, one vocabulary — a common language for quantum circuits, living cells, financial crises, and problems in mathematics. A working notation, not a proved-minimal basis.

Name change (2026): This document uses the current canonical names. Mapping from previous names: ORBIT → RESOLVE, LABEL → PROJECT, BIND → FUSE, MERGE → JOIN. FLIP, TWIST, CUP, LINK, SNAP, HALT, THERMAL unchanged. Published papers use the previous names; the legacy mapping table at §Full opcode table preserves all prior names.

The opcodes
The H^k tiers
Opcode incarnations across domains
Opcode reference
H³ and beyond (open — speculative)
The full opcode table
Theoretical foundations
The ISA trilogy and the Baez threefold way
The named ISAs
Relationship to other graphical calculi
Further reading

The opcodes

Opcode symbols: each canonical opcode has a formal symbol (used in LaTeX papers) and an outreach emoji.

Opcode	Formal	Emoji	Categorical morphism	Abstract role
RESOLVE	△	🔬	Comultiplication $\Delta: A \to A \otimes A$ · trace $\mathrm{tr}(\theta_A): \mathbf{1} \to \mathbf{1}$	1-to-many diagonalisation; spectral decomposition
PROJECT	▼	🎯	Multiplication $\mu: A \otimes A \to A$ · unit $\eta: \mathbf{1} \to A$	Many-to-1 evaluation; sector selection
FLIP	†	↩️	Dagger $(-)^\dagger$ · counit $\varepsilon_A: A^* \otimes A \to \mathbf{1}$ (CUP sub-role)	Orientation reversal; time-reversal; duality
TWIST	∮	🌀	Ribbon element $\theta_V: V \to V$; topological spin	Phase / monodromy; 1-to-1 with memory
FUSE	⋈	💎	Associator $\alpha_{A,B,C}: (A \otimes B) \otimes C \xrightarrow{\sim} A \otimes (B \otimes C)$; $F$-matrix	Non-associative fusion; recoupling; entanglement

Graphical calculus legend: 🕷️ present in ZX calculus · 🕷️* partial (related ZX construct, not full ISA semantics) · 🐸 present in 731 Frog Calculus · unmarked = ISA-native (no dedicated graphical symbol).

RESOLVE’s arity, settled. This page previously gave RESOLVE three different arities: $\Delta: A \to A \otimes A$ (1→2) in the opcode table, a closed scalar loop $\mathbf{1} \to \mathbf{1}$ (1→1) here, and Pachner $1 \to 4$ in the full table. These are not the same operation. RESOLVE is $\Delta: A \to A \otimes A$, arity 1→2 — the comultiplication, which is what the Frobenius identity PROJECT∘RESOLVE = id requires. The scalar-loop reading describes $\mathrm{tr}(\theta_A)$, a different morphism that happens to be built from RESOLVE and its dual; the Pachner 1→4 is a statement about triangulations, not about the morphism’s arity. RESOLVE is intentionally unmarked in the graphical-calculus legend because ZX assigns it no dedicated named spider. In ZX it appears as a disconnected bubble (no named spider); in the Frog Calculus it is a closed triangulation loop with no interior vertex. Neither calculus assigns RESOLVE a node: RESOLVE closes a loop rather than opening one.

The Frobenius identity PROJECT∘RESOLVE = id (evaluate after diagonalising = identity) is the spider identity. FLIP∘CUP = id (reverse an arrow, then close it = identity, compact closure) — where CUP is the counit sub-role of FLIP; see §CUP below.

The H^k tiers

Every opcode lives at a specific cohomological degree — the superscript k in H^k is not a power but a degree in the de Rham / sheaf cohomology sequence. The technical names:

H⁰ — zeroth cohomology: global sections, conserved quantities, classical observables
H¹ — first cohomology: connections, Berry phases, obstructions to global triviality
H² — second cohomology: characteristic classes, Chern numbers, genuine topological charges

In one sentence: H⁰ is counting (RESOLVE/PROJECT/FLIP), H¹ is interference (TWIST), H² is entanglement (FUSE).

Tier	Opcodes	One-word meaning	Technical meaning
H⁰	RESOLVE, PROJECT, FLIP	Counting	Global sections; classical observables; no phase
H¹	TWIST	Interference	Berry phase; connection; obstruction to triviality
H²	FUSE	Entanglement	Chern class; non-Abelian holonomy; topological charge

The ladder H⁰ → H¹ → H² is modelled on the de Rham sequence with d∘d = 0. This is also why FUSE∘TWIST ≠ 0 but FUSE∘FUSE = 0 in the ISA — the chain complex structure of the opcodes is the same object as the cohomology sequence. (See Theorem 3 below.)

The tier of a physical system is its minimum opcode requirement — the lowest H^k needed to describe it exactly:

Tier	Chemistry	Quantum computing
H⁰ only	DFT works fine	Clifford circuits; classically simulable
H¹ enters	CCSD sufficient	Still Clifford + cheap corrections
H² enters	DFT fails; need CASSCF	Gottesman-Knill breaks down; need magic states

The deeper point: DFT and Clifford simulation both fail at the H¹→H² boundary because they are both H⁰/H¹ approximations meeting the same H² obstruction. The cohomological degree is what makes that precise — see Weyl Chamber Homology for the proof that the chemistry Grassmannian and the quantum-computing Weyl chamber carry the same Bredon H² class (Euler characteristic 2).

Terminology note — legacy opcode names: Earlier papers and drafts use a twelve-opcode vocabulary. The current five-opcode names (PROJECT / RESOLVE / TWIST / FUSE / FLIP) consolidate those as follows: SPLAT → LABEL → PROJECT, SPLIT → ORBIT → RESOLVE, FLOP → CUP (FLIP fermion sub-role, name unchanged), BIND → FUSE, MERGE → JOIN. The sections below retain the legacy name FLOP where it clarifies categorical structure (especially the FLIP/FLOP distinction in the AZ tenfold way); treat CUP as the canonical name for that sub-role.

Opcode incarnations across domains

The same abstract opcode appears with different in/out counts depending on the domain. Each row is a physical domain; each column is one of the five canonical opcodes. Within each cell, the two sub-roles separated by · correspond to the legacy SPLIT/ORBIT sub-roles (RESOLVE column), SPLAT/LABEL sub-roles (PROJECT column), and FLIP/CUP sub-roles (FLIP column). Read across a row to see how a domain implements the full ISA; read down a column to see the same abstract operation across completely different fields.

Domain coordinate table — the three structural axes that determine which ISA a domain uses, plus whether its symmetry group is Abelian (if ✓, 6j symbols collapse to trivial phase factors and FUSE is weak or absent; if ✗, genuine recoupling and non-trivial FUSE). The suggested ISA follows from these axes. Most domains are Forge — this is the point: finite-β thermodynamic computation is the generic case; Boolean logic (Origami) and quantum mechanics (Meld) are the cold and imaginary limits of the same framework.

Domain	Semiring	Symmetry group	Abelian?	β-regime	Suggested ISA
3-manifold	ℤ / ℝ₊	SU(2) / G₂	✗	β→∞ (combinatorial)	Origami / Frog (if G₂)
Spectroscopy	ℝ	SU(2) / SO(3)	✗	finite β	Forge
Quantum info	ℂ	U(2ⁿ)	✗	β=it	Meld
Chemistry	ℝ → ℂ at CI	point group ⊂ G₂	✗	finite β → β=it at conical intersection	Forge → complex β at CI
Nuclear	ℝ	G₂ (always H²)	✗	finite β	Forge + Frog (tensor force mandatory)
Finance	ℝ₊	GL(n) / U(1)	✓	finite β	Forge
Condensed matter	ℂ / ℤ₂	U(1) → SU(2) → G₂	✓/✗ (phase-dependent)	β=it → β* snap	Forge → Meld (phase-dependent)
Turbulence	ℝ	SDiff(ℝ³)	✗	finite β	Forge
Biology	ℝ → ℂ	point group ⊂ G₂	✗	β ≈ β* (physiological)	Forge
Statistics / ML	ℝ₊	GL(n)	✓	finite β	Forge
MCMC / sampling	ℝ₊	—	✓	finite β	Forge
Causal inference	ℝ₊	DAG automorphisms	✓	finite β	Forge
Dynamical systems	(max,+) → ℝ	Diff(M)	✗	β→∞ → finite β	Origami → Forge
Number theory	ℚ_p · ℂ · 𝔸	GL(n) / G₂	✗	all β	full β-plane

Domain	RESOLVE 🔬	PROJECT 🎯	FLIP ↩️	TWIST 🌀	FUSE 💎
3-manifold	1 tet → 4 tets · closed triangulation loop	4 tets → 1 tet · face/edge colouring	1 tri → 3 tris · 3 tris → 1 tri	Dehn twist	non-Pachner obstruction
Spectroscopy	1 rep → CG irreps · closed G-orbit on weight lattice	CG sum → 1 rep (3j) · quantum number assignment	raising $J_+$ · lowering $J_-$	CG phase $(-1)^j$	Racah recoupling (6j) — H³ in extended ISA
Quantum info	1 qubit → register · feedback in quantum circuit	many states → 1 outcome · stabiliser projection	dagger / time-reversal · cup / partial trace	Berry phase / ribbon	$F$-matrix; non-Abelian anyon
Chemistry	CASSCF diagonalisation; NOON decomposition · G-walk / Galois step; CASSCF macro-iteration	3j/CG projection to density · orbital symmetry label; spin-state; point-group irrep	time-reversal; particle-hole; raising · Born rule for density; lowering	Berry phase on reaction path; Maslov index at conical intersection	tensor force; strongly-correlated bond; FeMoco; DFT failure
Nuclear	shell-model diagonalisation; Nilsson basis · closed shell (magic number); Nilsson orbit	9j evaluation; nuclear matrix element · $J$, $T$, parity quantum numbers	time-reversal; parity doubling · particle-hole in shell model	spin-orbit coupling (strong; mandatory); nuclear CG phase	tensor force $S_{12}$; mandatory in every nucleus
Finance	1 exposure → risk factor legs · closed risk cycle	risk factor legs → net P&L · scenario / regime selection	long ↔ short position · Born rule on exposure	convexity correction; drift	H² snap event (systemic crisis)
Condensed matter	Bogoliubov transform; band diagonalisation · hopping on lattice; Fermi sea orbit	3j / spectral projection · double occupancy $D$; order parameter; symmetry sector	particle-hole conjugation $C$ · fermionisation (Jordan-Wigner)	Berry phase; Chern number; BKT vortex	Mott transition (U/t snap); superexchange ring; topological order
Turbulence	large eddy → two smaller eddies · Kolmogorov cascade $k \to 2k$; inertial range	two sub-eddies dissipate at Kolmogorov scale · pressure Leray projector enforcing $\nabla\cdot u = 0$	—	vortex stretching $\omega \to \omega + (\omega\cdot\nabla)u\,\delta t$	blow-up conjecture: RESOLVE fails to close (NS unsolved)
Biology	CASSCF-like active-site diagonalisation · RESOLVE on Ramachandran torus; protein fold search; metabolic cycle	projection to electron density; tertiary fold evaluation · point-group label of active site; spin-state; cofactor oxidation state	time-reversal of reaction; particle-hole in redox · Born rule on conformational ensemble	Berry phase on reaction path; Maslov index at TS; spin-orbit (RuBisCO SOC problem)	chaperone-assisted H² fold; proofreading QEC; FeMoco (nitrogen fixation)
Statistics / ML	E-step (marginalise joint → conditional); multi-head projection · EM iteration (Fisher-Rao geodesic); Markov chain RESOLVE; attention token orbit	M-step (reconstruct parameters); head aggregation · convergence criterion ($\beta^*$ snap); energy eigenvalue; attention entropy	dagger on sufficient statistic · trace over latent variables; Born rule on posterior	$\alpha$-connection correction (curved exponential family); softmax Berry phase	multimodal posterior; phase transition in learning (grokking); non-Abelian Fisher tensor
MCMC / sampling	Markov chain step $x \to x’$; ergodic average over $\pi$	energy evaluation $U(x)$; accept/reject eigenvalue	Metropolis accept/reject: $\alpha = \min(1, e^{-\beta\Delta U})$	HMC leapfrog: symplectic integrator accumulating momentum phase	NUTS U-turn criterion: H² snap when Hamiltonian trajectory doubles back
Causal inference	DAG marginalisation $P(Y) = \sum_X P(Y\vert X)P(X)$ · observational RESOLVE; Markov blanket boundary	observational distribution fixed point · DAG structure label; backdoor criterion; instrumental variable sector	time-reversal of causal arrow · trace over latent confounders	do-calculus: graph surgery $\mathrm{do}(X=x)$ mutilates edges	counterfactual: $P(Y_{X=x}=y \vert X=x’, Y=y’)$ — two parallel worlds, non-local FUSE
Dynamical systems	soliton emergence: smooth hump → $n$ solitons (Lax eigendecomposition) · quasi-periodic orbit on $\mathbb{T}^n$; Lorenz lobe winding	soliton collision and re-emergence; Marchenko reconstruction · Lax eigenvalue $\kappa_n$; Lyapunov exponent; rotation number	time-reversal symmetry; Lax pair adjoint	resonance: KAM island chains; lobe-switching in Lorenz; Rankine-Hugoniot shock speed	cantorus at last KAM torus destruction $\varepsilon^*$; strange attractor (RESOLVE fails to close)
Number theory	spectral decomposition of $L^2(G(\mathbb{Q})\backslash G(\mathbb{A}))$; Hecke eigendecomposition · Apéry recurrence; rational points on $E(\mathbb{Q})$; $\pi_1(C)$ monodromy	L-function evaluation $L(s,\pi)$; Euler product · Hecke eigenvalue $a_p(E)$; quantum number of automorphic rep $\pi$	Langlands duality $G \leftrightarrow G^\vee$ · Abelian reciprocity ($GL_1$); class field theory	Tate twist; monodromy of local system; Selmer group $\mathrm{Sel}_n(E/\mathbb{Q}) \in H^1$	Apéry H² obstruction ($\zeta(3) \notin \mathbb{Q}$); Tate-Shafarevich $\Sha(E/\mathbb{Q})$; RH = H² zero-free region

Nuclear note: unlike every other domain, nuclear systems are H² by default. FUSE (the tensor force $S_{12}$) is mandatory even for the deuteron — the simplest nucleus. There is no H⁰ or H¹ nuclear bond. Chemistry reaches H² only in hard cases (FeMoco, bond-breaking); nuclear physics starts there and never leaves.

Opcode reference

RESOLVE 🔬 (formerly ORBIT, originally SPLIT) 🕷️

Symbols (Origami ISA): formal △ (hollow upward triangle — CT Δ analogue; open/branching, fan-out into components) · outreach 🔬

One wire becomes two (or one tetrahedron becomes four).

    │
    │
  ──┴──
  │   │
  │   │


String diagram	Comultiplication $\Delta: A \to A \otimes A$ — one wire splitting into two
Pachner move	$1 \to 4$ (one tetrahedron replaced by four sharing a central vertex)
Category theory	Coproduct / comultiplication of a bialgebra or Hopf algebra
Algebra	Coproduct $\Delta(E) = E \otimes K + 1 \otimes E$ in quantum group $U_q(\mathfrak{sl}_2)$

Where RESOLVE appears:

Domain	Instance	What splits
Quantum mechanics	Fourier / Bogoliubov transform	Single mode → momentum modes
Angular momentum	Clebsch-Gordan decomposition	Product representation → irreducibles
Nuclear physics	Racah recoupling	3-body state → sum of 2-body products
Langlands programme	Hecke eigendecomposition	Automorphic form → Hecke eigensheaves
Quantum error correction	Stabiliser expansion	Logical qubit → physical qubit register
Finance	Factor decomposition (PCA on yield curve)	Portfolio → risk factors

Key role: RESOLVE is always the diagonalisation step — the moment a composite object is resolved into its irreducible pieces. Every Fourier transform, every change of basis, every spectral decomposition is a RESOLVE.

PROJECT 🎯 (formerly LABEL, originally SPLAT) 🕷️

Symbols (Origami ISA): formal ▼ (solid downward triangle — filled/converging; fan-in onto a target) · outreach 🎯

Two wires become one (or four tetrahedra become one).

  │   │
  │   │
  ──┬──
    │
    │


String diagram	Multiplication $\mu: A \otimes A \to A$ — two wires merging into one (or a cap: one wire curling down to nothing)
Pachner move	$4 \to 1$ (four tetrahedra sharing a vertex collapsed to one)
Category theory	Counit $\varepsilon: A \to k$ of a Frobenius algebra; or the evaluation map $A^* \otimes A \to k$
Algebra	The $3j$-symbol / Clebsch-Gordan coefficient; the POVM measurement map

Where PROJECT appears:

Domain	Instance	What gets projected
Angular momentum	$3j$-symbol evaluation	CG amplitude → scalar (6j recoupling belongs at H³/RECOUPLE)
Quantum gravity	Ponzano-Regge face amplitude	Spin foam triangle → 3j amplitude
Quantum information	Character POVM measurement	State → outcome probability
Bethe ansatz	Scalar product of Bethe states	Rapidities → norm
Langlands programme	L-function evaluation $L(s, \pi)$	Automorphic form → complex number
Finance	Portfolio valuation	Risk factor exposure → P&L scalar

Key role: PROJECT is always the evaluation step — the moment a structured object is projected to a number. Every inner product, every measurement, every partition function evaluation is a PROJECT.

The Frobenius axiom $\mathrm{PROJECT} \circ \mathrm{RESOLVE} = \mathrm{id}$ (the counit-comultiplication identity) is the algebraic statement that diagonalisation followed by projection is the identity — you get back what you put in. This is the Pentagon identity in disguise, and it is simultaneously the Biedenharn-Elliott identity of angular momentum, the no-arbitrage condition in finance, and the topological invariance of Ponzano-Regge amplitudes.

TWIST 🕷️* ∮ 🌀

A wire acquires a phase (a curl or loop in the diagram).

Symbols (Origami ISA): formal ∮ (closed-loop integral — phase accumulated around a circuit) · outreach 🌀

    │
   ╭╯
   │
   ╰╮
    │


String diagram	Ribbon element $\theta_V: V \to V$ — a wire looping through a full $2\pi$ twist (a curl)
Pachner move	Gauge move (no change in triangulation topology; changes the phase of the amplitude)
Category theory	The ribbon element / twist morphism of a ribbon category; the natural isomorphism implementing the topological spin
Algebra	$\theta_j = q^{j(j+1)}$ (topological spin of a spin-$j$ anyon in $\mathrm{SU}(2)_q$)

Where TWIST appears:

Domain	Instance	What acquires the phase
Topological phases	Berry phase / Chern number	Wavefunction under adiabatic loop
Anyons	Topological spin $\theta_a = e^{2\pi i h_a}$	Anyon under $2\pi$ rotation
Phase transitions	BKT / TWIST failure at $\beta^* = \tfrac{1}{2}$	Quantum dimension $d_{1/2}(\beta) \to 0$
AZ classification	Chiral zero mode (S symmetry)	Edge state phase
Langlands programme	Monodromy of a local system	Parallel transport around a loop on the curve
Weil conjectures	Riemann hypothesis (zero-free region)	Zeta function zeros stay off the critical line

Key role: TWIST is the gap / topology check — it encodes whether the system is in a topologically non-trivial phase. TWIST failure (the amplitude $d_{1/2}(\beta)$ reaching zero) is the universal signature of a phase transition across all models in the $\mathrm{SU}(2)_q$ family. See BKT Transition / TWIST Failure for the full treatment.

FLIP 🕷️ † ↩️

A wire reverses orientation (arrow pointing down instead of up).

Symbols (Origami ISA): formal † (dagger — anti-involution, time-reversal) · outreach ↩️

    ↑         ↓
    │   →     │
    │         │


String diagram	Dagger functor $(-)^\dagger: \mathcal{C} \to \mathcal{C}^{op}$ — all wire orientations reversed; or the pivotal structure $V \cong V^{**}$
Pachner move	$1 \to 3$ (one triangle replaced by three sharing a central vertex) — the 2D orientation reversal
Category theory	The dagger / adjoint functor; the pivotal structure on a ribbon category; an anti-involution
Algebra	Anti-unitary operator squaring to $\pm 1$; transpose of the Cartan matrix (root orientation reversal)

Where FLIP appears:

Domain	Instance	What gets reversed
Quantum mechanics (AZ)	Time reversal $T$; $T^2 = +1$ (real) or $T^2 = -1$ (quaternionic)	Time coordinate
PT symmetry	Anti-unitary $\mathcal{T}$ in Bender-Boettcher PT quantum mechanics	Time
Topological phases	Kramers degeneracy ($T^2 = -1$, class AII/CII)	Kramers pairs
Langlands programme	Langlands duality $G \leftrightarrow G^\vee$ (root-system orientation reversal)	Long roots ↔ short coroots
Braiding / anyons	Charge conjugation; anti-particle	Anyon ↔ anti-anyon
ZX-calculus	Wire reversal (upward ↔ downward arrow)	Computational direction

FLIP fixed points (self-dual groups where FLIP = identity): $GL_n$, $G_2$, $F_4$, $E_8$. The self-duality of $G_2$ under FLIP is the 731 theorem (Paper 271). In the Langlands programme, these self-dual groups are the most symmetric — and the hardest — cases.

Key role: FLIP is the duality / orientation opcode. Any time a computation has a “left–right” or “past–future” symmetry, FLIP is the operation that implements it. The distinction between real ($T^2=+1$) and quaternionic ($T^2=-1$) FLIP is the distinction between the Origami-ISA column and the Meld-ISA column of the Baez threefold way.

CUP 🕷️ (FLIP fermion sub-role — formerly FLOP)

A wire curls under into a cup (fermionisation / Born rule).

  │   │
  │   │
  ╰───╯


String diagram	Frobenius co-unit evaluation / cup: $A \otimes A \to k$ — two wires meeting at the bottom in a cap
Pachner move	$3 \to 1$ (three triangles sharing a vertex collapsed to one)
Category theory	The Frobenius co-unit; the trace map $\mathrm{tr}: \mathrm{End}(V) \to k$; the Born rule
Algebra	Jordan-Wigner string; Majorana fermion creation/annihilation; particle-hole conjugation $C$

Where CUP appears:

Domain	Instance	What gets fermionised
Condensed matter (AZ)	Particle-hole symmetry $C$; $C^2=+1$ (Majorana) or $C^2=-1$ (complex fermion)	Particle ↔ hole
1D quantum models	Jordan-Wigner transform	Spin chain ↔ fermion chain
Quantum gravity	Trace / inner product in LQG	Spin network state → amplitude
Finance	Born rule / expectation value	Density matrix → portfolio expectation
Langlands (abelian)	Class field theory / $GL_1$ reciprocity	Hecke character → Galois character

CUP and the division algebra ladder:

CUP producing a Majorana co-unit ($C^2=+1$): lives at the $\mathbb{R}$-rung (Origami ISA, GOE, Dyson $\beta_D=1$)
CUP producing a complex fermion co-unit ($C^2=-1$): lives at the $\mathbb{H}$-rung (Meld ISA, GSE, Dyson $\beta_D=4$)
No CUP: lives at the $\mathbb{C}$-rung (Forge/Meld ISA, GUE, Dyson $\beta_D=2$)

Key role: CUP is the fermionisation / Born rule sub-role of FLIP. It is present in every model where particle statistics matter. Its sign ($C^2=\pm1$) is the deepest structural label in the AZ tenfold way — the distinction between Majorana (self-conjugate) and Dirac (complex) fermions. Legacy name: FLOP.

PROJECT (sector sub-role) 🕷️ ▼ 🎯

A wire passes through a projector (sector selection).

Symbols (Origami ISA): formal ▼ (solid downward triangle — filled/converging; fan-in onto a target) · outreach 🎯

    │
  ┌─┴─┐
  │ e │   (e² = e)
  └─┬─┘
    │


String diagram	Unit morphism $\eta: \mathbf{1} \to A$ — a dot (the Frobenius algebra unit); creates a wire from nothing
Pachner move	No direct Pachner counterpart; it is the colouring operation that labels edges/faces before Pachner moves act
Category theory	The unit of the Frobenius algebra $(A, \mu, \eta, \Delta, \varepsilon)$; state preparation; the map $\mathbf{1} \to A$ selecting the initial sector
Algebra	Gauge fixing; stabiliser eigenstate preparation; sector selection; the Satake isomorphism

Where PROJECT appears (sector sub-role):

Domain	Instance	What gets labelled
Quantum error correction	Stabiliser projection	Logical qubit sector
Gauge theory	Gauge fixing (Lorenz, Coulomb, …)	Physical Hilbert space
Anyons	Anyon type assignment to worldlines	Topological sector
Bethe ansatz	Vacuum selection (reference state)	Pseudovacuum sector
Langlands programme	L-function / automorphic representation $\pi$	Hecke eigenvalue
PT symmetry	Parity sector projection $\mathcal{P}$	Even / odd parity eigenspace
Finance	Scenario / regime selection	Market state

PROJECT failure = PT phase transition. When PT symmetry spontaneously breaks (Bender-Boettcher), eigenstates of $H$ are no longer eigenstates of $\mathcal{PT}$: PROJECT can no longer project onto definite-parity sectors. The parity sectors mix at the exceptional point.

Key role: PROJECT is the sector / gauge / colour opcode. It is always the operation that selects which subspace of the full Hilbert space the computation lives in. Every gauge-fixing, every stabiliser projection, every quantum number assignment is a PROJECT.

FUSE 🐸 ⋈ 💎 (formerly BIND)

Three wires enter a vertex (non-Abelian fusion; associator).

Symbols (Origami ISA): formal ⋈ (natural join / bowtie — two registers fused into an entangled pair) · outreach 💎

A naming note: FUSE is the one opcode that lives exclusively in the 🐸 Frog Calculus — the trivalent vertex, the non-associative fusion, the G₂ structure that ZX calculus cannot express. Meanwhile PROJECT (formerly LABEL/SPLAT), FLIP, and CUP (formerly FLOP) — which sound exactly like things a frog does — are all firmly in the 🕷️ ZX spider calculus. The name FUSE was chosen for its resonance with fusion categories, F-matrices, and anyon fusion, and to eliminate the collision with the monad bind operator (»=) in category-theoretic and functional-programming contexts.

  │   │   │
  │   │   │
  └───┼───┘
      │
      │


String diagram	Associator $\alpha_{A,B,C}: (A \otimes B) \otimes C \xrightarrow{\sim} A \otimes (B \otimes C)$ — three wires, non-trivial crossing structure; or the trivalent vertex of a fusion category
Pachner move	Not a standard Pachner move — it is the obstruction to Pachner invariance; its presence signals non-associativity
Category theory	The associator of a monoidal category; non-trivial when the category is only quasi-monoidal (quasi-Hopf algebra, braided fusion category with non-trivial $F$-matrices)
Algebra	Octonion associator $[e_i, e_j, e_k] = (e_i e_j)e_k - e_i(e_j e_k)$; the $F$-matrix of a fusion category; the 4-Majorana coupling $\gamma_i\gamma_j\gamma_k\gamma_l$

Where FUSE appears:

Domain	Instance	What fails to associate
Non-Abelian anyons	$F$-matrix / recoupling coefficient	$(a \times b) \times c \neq a \times (b \times c)$ in fusion
Octonions / $G_2$	Octonion associator; Furey’s ladder operators	$e_i(e_j e_k) \neq (e_i e_j)e_k$
Topological phases	Fidkowski-Kitaev $\mathbb{Z} \to \mathbb{Z}_8$ collapse	4-Majorana interaction
Interacting fermions	SYK four-body coupling	Non-factorising 4-fermion vertex
Langlands (non-Abelian)	Non-commuting Hecke operators at different primes	$[T_p, T_q] \neq 0$ for $GL_n$, $n \geq 2$
p-adic Langlands	Pentagon failure in p-adic Hodge theory	Non-associative p-adic completions

FUSE in finance: The interbank network accumulates systemic risk in H¹ — the cycle topology of mutual exposures, which is non-trivial even though balance-sheet arithmetic is Abelian. FUSE marks the H² snap event: the moment when those H¹ cycles become globally inconsistent and cannot be unwound bilaterally. Systemic risk is measured in H¹; systemic crises (2008 GFC, LTCM) are H² snap events. See Papers 397–398.

FUSE theorem (The Opcode Rosetta Stone, Paper 447): A gapped topological phase has non-Abelian anyonic order if and only if its minimal ISA programme contains FUSE. Associative phases are FUSE-free; non-associative phases require FUSE.

FUSE and the division algebra ladder:

No FUSE: associative computation — $\mathbb{R}$, $\mathbb{C}$, or $\mathbb{H}$ rung (pentagon holds, $\alpha = \mathrm{id}$)
FUSE present: non-associative — $\mathbb{O}$-rung; $G_2$, $E_8$; Furey’s octonionic Standard Model programme; 731-ISA regime

The canonical definition (Paper 591, Definition 4.1): FUSE $= \alpha_{A,A,A} \neq \mathrm{id}$ in the ISA magmoidal category. FUSE present $\Leftrightarrow$ pentagon coherence axiom fails $\Leftrightarrow$ non-trivial $F$-matrix.

This equivalence is definitional, not a result. FUSE is defined as the non-trivial associator, and a category with a non-trivial associator is magmoidal by definition — so “FUSE requires a magmoidal extension” restates the definition rather than proving anything. The corresponding theorem in Paper 591 is a tautology and is recorded there as one.

That does not make the opcode useless: naming the associator, and pinning it to a definite categorical object, is what lets the same term be used across octonions, $F$-matrices and Majorana couplings without ambiguity. But the content is in the dictionary below, which is checkable case by case, not in the definitional equivalence above. Note: Fibonacci anyons have non-trivial $F$-matrices but still satisfy the pentagon (they are a fusion category); they live at the $\mathbb{H}$-rung boundary, not the $\mathbb{O}$-rung.

The Fidkowski-Kitaev collapse ($\mathbb{Z} \to \mathbb{Z}_8$) is FUSE insertion: promoting a CUP-only programme (free Majorana chain, $\mathbb{C}$-rung) to a CUP+FUSE programme ($\mathbb{O}$-rung) collapses the integer winding-number classification to $\mathbb{Z}_8$, because $8$ is the Cayley-Dickson period at the octonion rung.

Key role: FUSE is the non-associative opcode. Its presence or absence is a syntactic, computable test for non-Abelian anyonic order — no modular tensor category computation required. It is the hardest opcode to implement and the most powerful: systems with FUSE can encode computations that FUSE-free (associative) systems cannot.

H³ and beyond (open — speculative)

The Origami ISA currently terminates at H² (FUSE). The orbital simplex construction (Paper 719) and the Ponzano-Regge / Turaev-Viro frameworks suggest a coherent H³ tier whose primitive is the 6j symbol evaluated on a tetrahedron — not a new mathematical object, but a new ISA depth.

The correct simplex-to-symbol ladder (per Turaev-Viro consensus):

H^k	Simplex	Recoupling symbol	ISA opcode
H⁰	vertex	identity / single spin	RESOLVE/PROJECT/FLIP
H¹	edge	R-matrix / braiding	TWIST
H²	triangle	3j / Clebsch-Gordan	FUSE (associator = non-trivial F-matrix)
H³	tetrahedron	6j / F-matrix amplitude	RECOUPLE (proposed)
H⁴	4-simplex	15j symbol	(Crane-Yetter; not yet proposed)

Key clarifications:

FUSE is defined as the non-trivial associator α_{A,B,C} ≠ id — this is the categorical H² object. The F-matrix of a fusion category is this associator; it is an H² primitive.
The 6j symbol as a Racah recoupling amplitude (the scalar evaluated on a tetrahedron) is an H³ object in the Ponzano-Regge / Turaev-Viro sense: it is a 3-cocycle evaluated on a 3-simplex.
The 9j symbol is not a new simplex-level primitive; it is a circuit of three 6j (RECOUPLE) gates. Mac Lane coherence guarantees that all higher recoupling reduces to sequences of F-moves.
The domain table rows for spectroscopy/chemistry labelled “6j” under PROJECT or FUSE should be understood as H³-level operations that the current three-tier ISA approximates by collapsing into FUSE.

Whether RECOUPLE is one opcode or several is open: the Frog ISA (H⁰–H³) already accommodates this tier, and EMIT is the proposed H³ primitive in the QFT / amplituhedron context. These may be domain-specific faces of a single H³ opcode, or genuinely distinct primitives. See Paper 719 for the orbital simplex argument; see Papers 207/281 (Frog Calculus) for the graphical calculus.

For a full derivation of how Pachner moves map to quantum symbols — including the Biedenharn-Elliott identity and why 9j is a RECOUPLE circuit — see Pachner Moves and Quantum Symbols.

The full opcode table

Opcode	Legacy names	Graphical calculi	String diagram	Pachner move	AZ symmetry	Division algebra	Langlands
RESOLVE	ORBIT, SPLIT	🕷️	$\Delta: A \to A \otimes A$ (split)	$1 \to 4$	—	All rungs	Hecke eigendecomposition
PROJECT	LABEL, SPLAT	🕷️	$\mu: A \otimes A \to A$ (merge)	$4 \to 1$	—	All rungs	L-function evaluation
TWIST	—	🕷️*	$\theta_V: V \to V$ (curl)	Gauge move	$S$ (chiral)	All rungs	Monodromy of local system
FLIP	—	🕷️	$(-)^\dagger$ (wire reversal)	$1 \to 3$	$T$ (time reversal)	$\mathbb{R}$ / $\mathbb{H}$	Langlands duality $G \leftrightarrow G^\vee$
↳ CUP	(FLIP sub-role, formerly FLOP)	🕷️*	$\varepsilon_A: A^* \otimes A \to \mathbf{1}$ (cup)	$3 \to 1$	$C$ (particle-hole)	$\mathbb{R}$ / $\mathbb{H}$	Abelian reciprocity ($GL_1$)
FUSE	BIND	🐸	Associator $\alpha_{A,B,C}$ (trivalent)	Obstruction	—	$\mathbb{O}$ only	Non-Abelian Hecke interaction

Theoretical foundations

Why five opcodes?

The Origami ISA is a working vocabulary, not a proved-minimal one. The opcodes are chosen so that every one names a real morphism in a traced symmetric monoidal category with Frobenius structure (TSMC — a monoidal category with a trace operation closing loops in the string diagram), plus one magmoidal extension for the non-associative case.

UNPROVED (flagged 2026-08-04). Earlier versions of this page asserted that this set is the minimal magmoidal extension of the free TSMC — “the smallest opcode set that is both TSMC-complete and magmoidal-complete”. No proof of that claim exists anywhere in this corpus, and no reference was ever given for it. It is the only statement that would answer “why exactly five?”, so its absence matters. Until someone proves it, the honest position is that five morphisms were chosen because they span the structures of interest — which is a legitimate design choice, but not a theorem. See “What a minimality proof would require” below. Every opcode except FUSE is a named morphism in the TSMC + Frobenius structure (the “spider calculus”). FUSE is the unique opcode that requires a magmoidal extension: it encodes a non-trivial associator, realised physically as G₂/octonion symmetry.

What a minimality proof would require. Making the claim precise is most of the work, and it is worth stating what is missing so the gap is actionable:

Fix the target class. “TSMC-complete” needs a definition: complete for what — generating every morphism of the free TSMC on one self-dual object? Every morphism of some named category of physical processes? Without this the claim has no truth value.
Fix the generation rules. Composition and tensor alone, or also trace, dagger, and the Frobenius equations as rewrites? Minimality is meaningless until the closure operations are pinned down.
Prove sufficiency — every morphism of the target class is generated. For the Frobenius/spider fragment this is already known: the normal-form theorem for special commutative Frobenius algebras (Coecke–Pavlović–Vicary; Lack’s composing-PROPs) says every connected diagram reduces to a spider. So the sufficiency half is standard and citable rather than new.
Prove independence — for each opcode, exhibit a morphism generated by the full set but not by the rest. This is the half that is genuinely open here, and it is where a proof would live. Independence is normally shown by constructing a model (a functor to some concrete category) that satisfies the equations of the smaller set but not the larger.

Step 4 for FUSE looks the most tractable: any non-associative model — the octonions being the obvious one — should separate FUSE from the associative fragment, since the associator is non-trivial there and trivial everywhere in a monoidal category by coherence. But note that x_E5 found the naive pentagon fails for octonionic labels (1176 of 2401 labelings), so the associator must be supplied as explicit coherence data before any such argument runs.

The five opcodes form a completeness hierarchy: each tier lifts the ISA to the next level of the cohomological (H^k) computational tower, and no opcode at level k can be simulated by any combination of opcodes at level k−1. The H^k tiers are not merely a grading — they are the homology groups of a genuine chain complex (see Theorem 3 below).

Monoidal categories underlie all of mathematical physics for the same reason: any system in which operations compose in parallel and in sequence — quantum circuits, Feynman diagrams, tensor networks, representation theory, the Langlands correspondence — is an object in some monoidal category. The twelve opcodes are the universal generating morphisms of that structure, extended to include the non-associative (magmoidal) and non-local (compact closed) regimes.

The same morphisms do appear across nuclear spectroscopy, topological quantum computing, loop quantum gravity, financial XVA, protein folding and the ribosome — because all of these are modelled in some monoidal category, and that is what monoidal categories are for.

What this does and does not buy. Naming the same morphism in two fields is real but thin: it transfers structure, not content. Direct tests in this corpus have repeatedly found that a shared categorical name does not imply a shared prediction — see x730a (Grassmannian path geometry reproduced the energy gradient rather than supplementing it), x731a (redox-ladder geometry lost to plain d-electron count), and x732a (Yutsis diagram topology failed to predict the good coupling tree, which is set by the exchange couplings). The earlier wording here — “they are not analogies, they are the same categorical morphisms running on different physical hardware” — overstated what the mapping delivers, and is withdrawn.

How precise are the ISA mappings across domains? The mappings range from exact algebraic theorems (Tier A: Fano commutation structure, Casimir identity, Wigner vertex theorem) to quantitative predictions verified by experiment (Tier B: MCMC optimal acceptance rates, GEV shape parameter, Shor mana = 0) to useful organisational language for hierarchies the field already knew were hierarchical (Tier C: Pearl’s causal ladder, fairness taxonomies). The programme does not claim all mappings are equally strong — see the full precision taxonomy.

The category theory behind the opcodes

This section explains why the ISA opcodes are rigorous mathematical objects and not just suggestive names — and why the same objects appear across physics, mathematics, and computing without any analogy or hand-waving.

The ladder of categories

The opcodes are generated by a strict hierarchy of categorical structures. Each level adds one new kind of morphism, and each addition corresponds to one H^k tier:

Category type	What it adds	Opcode unlocked	H^k tier
Monoidal category	Parallel composition (⊗); unit object 1	RESOLVE (fan-out sub-role) · PROJECT (projection + unit sub-roles)	H⁰
+ Symmetric	Swap morphism; wire crossing	RESOLVE (closed traces)	H⁰
+ Traced	Feedback loops (trace closing a wire on itself)	RESOLVE (full feedback)	H⁰
+ Frobenius	Comultiplication + counit satisfying Frobenius law	RESOLVE ↔ PROJECT duality	H⁰
+ Compact closed	Dual objects; cups and caps	FLIP (Born rule / fermionisation sub-role)	H⁰/H¹
+ Dagger	Anti-involution $(-)^\dagger$ reversing all arrows	FLIP (time-reversal sub-role)	H¹
+ Ribbon	Ribbon element $\theta_V$ (topological spin / twist)	TWIST (Berry phase)	H¹
+ Magmoidal	Non-trivial associator $\alpha_{A,B,C} \neq \mathrm{id}$	FUSE (entanglement)	H²

A monoidal category is any mathematical structure where operations can compose both sequentially (one after another, written ∘) and in parallel (side by side, written ⊗), with a unit object 1 for “doing nothing.” This covers essentially all of mathematical physics: quantum circuits, Feynman diagrams, tensor networks, representations of groups, the Langlands correspondence.

Each row in the table is a property that a monoidal category may or may not have. The opcodes are the canonical generators of each property — the minimal new morphism you must add to express it. This is why the opcodes are not arbitrary: they are forced by the categorical structure.

Why FUSE is special: magmoidal categories

Every category in the ladder above (monoidal through ribbon) satisfies the pentagon axiom: the associator is coherent, meaning all ways of re-bracketing a tensor product $(A \otimes B) \otimes C \cong A \otimes (B \otimes C)$ are consistent. In such categories, the associator is effectively invisible — you can ignore brackets.

A magmoidal category is one where the pentagon axiom fails: the associator $\alpha_{A,B,C}$ is genuinely non-trivial and cannot be set to the identity. This is the categorical home of:

Octonions — the non-associative normed division algebra; $e_i(e_j e_k) \neq (e_i e_j)e_k$
Non-Abelian anyons — fusion categories with non-trivial $F$-matrices (the $F$-matrix is the associator)
$G_2$ symmetry — the automorphism group of the octonions; the exceptional Lie group whose root system is the Fano plane

FUSE is the single opcode that requires magmoidal extension. Every opcode except FUSE lives in a ribbon category (associative, pentagon holds). FUSE is the morphism that encodes the associator itself — which is why it requires genuine multi-body correlation (H²) that no H⁰/H¹ approximation can reproduce.

The Frobenius algebra: why RESOLVE and PROJECT are dual

RESOLVE and PROJECT are not independent. Together with FLIP they form a Frobenius algebra $(A, \mu, \eta, \Delta, \varepsilon)$:

$\Delta: A \to A \otimes A$ — RESOLVE fan-out sub-role (comultiplication)
$\mu: A \otimes A \to A$ — PROJECT projection sub-role (multiplication)
$\eta: \mathbf{1} \to A$ — PROJECT unit sub-role
$\varepsilon: A \to \mathbf{1}$ — FLIP Born-rule sub-role (counit)

The Frobenius axiom $(\mu \otimes \mathrm{id}) \circ (\mathrm{id} \otimes \Delta) = \Delta \circ \mu = (\mathrm{id} \otimes \mu) \circ (\Delta \otimes \mathrm{id})$ is the algebraic statement that “fan-out then project = identity.” This is simultaneously:

The Pentagon identity in angular momentum theory (Biedenharn-Elliott)
The no-arbitrage condition in finance
The topological invariance of Ponzano-Regge amplitudes
The Reidemeister moves for knot diagrams

These are not analogies. They are the same equation, in the same Frobenius algebra, evaluated in different semirings over different physical hardware.

Arity: RESOLVE is 1 → 2, not 1 → 4 (settled 2026-08-03)

Three different arities for RESOLVE/SPLIT have appeared across the corpus. They are not alternatives; two of them are errors of type.

form	what it actually is
$\Delta: A \to A \otimes A$, 1 → 2	the generator. Frobenius comultiplication. This is RESOLVE.
$A \to A^{\otimes 4}$, 1 → 4	the iterated coproduct $(\mathrm{id}\otimes\Delta)\circ\Delta$ — a derived term built from two applications of the generator, not a primitive
$1 \to 4$ in the table above	the Pachner move, a different column and a different kind of object entirely — a move on a triangulation, not a morphism arity

The confusion arose because papers interpreting wires in Rep($G$) wrote the iterated form directly and read its arity off the diagram. A generator is whatever cannot be decomposed into other generators; the 4-legged form decomposes, so it is not one.

This is recorded here rather than in a paper because a definitional reference must be able to follow renames. Paper 591 (Categorical Foundations of the Origami ISA, 10.5281/zenodo.21300689, deleted 2026-08-03) reached the same conclusion on the same grounds, and was then superseded fifteen days after upload when ORBIT→RESOLVE, LABEL→PROJECT, BIND→FUSE were finalised on 2026-07-25. A PDF cannot track that; this file can.

What 591 claimed and what remains unverified. Its structural claim was that the seven associative opcodes are precisely the generating structure maps of a ribbon pivotal category with duals, and that FUSE (then BIND) is the associator — the obstruction to pentagon coherence. The second half is true by construction (it is how FUSE is defined here, see “Why FUSE is special”), so it carries no independent content. The first half — that these seven generate, and that no eighth associative generator is needed — was never checked. It is the claim worth proving, and it is name-independent, so it survives any future rename. Prior art to engage before attempting it: Kuperberg’s spiders (CMP 1996) for rank-2 groups, and Paper 572, whose x572a pentagon-relation check did not match the textbook prediction and was reinterpreted post hoc. Resolve 572 first.

Bonding opcodes: why Origami is still universal

Before addressing the bonding extension, a question must be answered directly: if Origami is universal, why does bonding require new opcodes?

The answer is that Origami is universal for single-object computation — morphisms within one Frobenius algebra (one atom, one orbital, one site). JOIN and CLEAVE are not missing from Origami; they are not the right type to exist in it. They are morphisms between two different Frobenius algebras. You cannot write them down until you have two objects.

This is not a patch. It is the standard categorical tower:

Level	Structure	ISA	Objects
Level 1	Single Frobenius algebra on A	Origami ISA	One orbital / site (an atom is a composite — see below)
Level 2	PROP of two Frobenius algebras A, B	bonding layer	Two atoms bonding

Origami is universal at Level 1 exactly as group theory is universal for symmetry — and group theory does not describe homomorphisms between groups until you have two groups. That is not a failure of group theory; it is a statement about categorical level. The bonding layer is the forced Level 2 extension, derived from Origami by the standard PROP construction.

The full argument — including why PAT-q is the canonical faithful representation of both levels simultaneously, and what a “hoperator” is — is developed in Paper 708 (The Universal Hoperator). This section records the conclusions; see that paper for proofs.

The CT↔chemistry bridge. An electron in a hydrogen 1s orbital traces a Hopf fibre on $S^3$ — topologically a $(1,1)$ torus knot $T(1,1)$. More generally, an electron in an orbital with quantum numbers $(n, \ell)$ traces a torus knot $T(n, \ell)$ on $S^3$. A covalent bond between atoms A and B combines their orbitals: the molecular orbital has knot type

\[T(n_A, \ell_A) \mathbin{\sharp} T(n_B, \ell_B)\]

the connected sum — additive genus $g_A + g_B$. This connected-sum operation is JOIN. CLEAVE is its Frobenius dual: the bond-breaking operation that recovers the two component torus knots. The Frobenius condition on (JOIN, CLEAVE) is not imposed on chemistry — it is the algebraic form of detailed balance, which reversible chemistry already satisfies.

Level 1 — Single-site Frobenius algebra (Origami ISA opcodes, unchanged):

CT symbol	CT name	Opcode	Type	Frobenius dual	Chemical meaning
$\Delta$	comultiplication	RESOLVE	$A \to A \otimes A$	PROJECT	spectral decomposition; diagonalise orbital
$\mu$	multiplication	PROJECT	$A \otimes A \to A$	RESOLVE	evaluation; project onto quantum number
$\eta$	unit	PROJECT (sub-role)	$\mathbf{1} \to A$	FLIP	orbital state preparation
$\varepsilon$	counit	FLIP (CUP sub-role)	$A \to \mathbf{1}$	PROJECT	measurement; orbital annihilation
$\theta$	ribbon twist	TWIST	$A \to A$	TWIST (self-dual)	Berry phase; spin-orbit coupling

Level 2 — Inter-site Frobenius algebra (bonding opcodes only):

CT symbol	CT name	Opcode	Type	Frobenius dual	Chemical meaning
$\mu$	multiplication	JOIN	$A \otimes B \to A\mathbin{\sharp}B$	CLEAVE	covalent bond: connected sum of torus knots
$\delta$	comultiplication	CLEAVE	$A\mathbin{\sharp}B \to A \otimes B$	JOIN	bond breaking: split connected-sum orbital
$\tau$	braiding	LINK	$A \otimes B \to A \otimes B$	LINK (self-dual)	coordinate/dative bond: Hopf linking, bond order = linking number

2-cell (between levels — neither Level 1 nor Level 2):

CT level	Opcode	Type	Inverse	Physical meaning
2-cell between PROPs	SNAP↑	$\mathcal{F} \to \mathcal{F}’$	SNAP↓	β* tier promotion; phase transition upward
2-cell between PROPs	SNAP↓	$\mathcal{F}’ \to \mathcal{F}$	SNAP↑	β* tier demotion; phase transition downward

SNAP is primitive by categorical level: no composition of Level 1 or Level 2 morphisms can produce a 2-cell. JOIN is a chemical reaction (within one thermodynamic phase). SNAP is a phase transition (between phases). These are not in the same layer.

Why JOIN and CLEAVE are forced (not chosen): any Frobenius algebra at Level 2 requires both a multiplication $\mu: A \otimes B \to A\mathbin{\sharp}B$ and a comultiplication $\delta: A\mathbin{\sharp}B \to A \otimes B$ satisfying the Frobenius condition. There is no choice about whether to include them — a Frobenius algebra without both is not a Frobenius algebra. Hybridisation (sp³, sp²) is not a new opcode: it is TWIST applied within Level 1 to the orbital colour label.

Why LINK is forced: the braided monoidal structure of the Level 2 PROP requires a braiding morphism. For orbital knots the natural braiding is Hopf linking — two torus knots $T(p_A, q_A)$ and $T(p_B, q_B)$ linked with linking number $\nu$ (bond order) without changing their topological types. This is the coordinate/dative bond.

The Frobenius condition as microscopic reversibility: $(\mathrm{id} \otimes \delta) \circ \mu = (\mu \otimes \mathrm{id}) \circ (\mathrm{id} \otimes \delta)$ Bond formation followed by bond breaking in either order gives the same result. Every reversible chemical reaction satisfies this. Irreversible reactions live outside the Frobenius sector.

The missing rung: is the atom itself Level 1, or a composite? (open, 2026-07-27)

The table above writes the Level 1 object as “one atom / orbital / site” — treating “atom” and “orbital” as interchangeable. They are not, except in the single-electron case (hydrogen-like ions), and the difference exposes a genuine gap in the tower.

The true Level-0 object is a single orbital $(n,\ell)$: one Frobenius algebra, acted on by RESOLVE/PROJECT/FLIP/TWIST exactly as the Level 1 table describes. A multi-electron atom is already composite — assembled from several Level-0 objects, not a single one. The Level 1 table’s “atom” label silently assumes this composite already exists; it does not name the assembly.

Why this is not the same problem Level 2 solved. JOIN/CLEAVE/LINK (Level 2) combine two distinct Frobenius algebras $A, B$ on two different nuclei — a PROP of two objects. The within-atom problem combines several copies of the same kind of object (orbitals) on one nucleus, under two hard constraints that a plain tensor product does not respect:

Pauli exclusion (kinematic — genuinely forces structure). At most one fermion per spin-orbital $(n,\ell,m_\ell,m_s)$; the chemist’s “two per orbital” is two spin-orbitals per spatial orbital. Free combination (an ordinary tensor product, unlimited copies) is the wrong structure. Pauli exclusion forces the shift from the tensor algebra to the exterior algebra $\Lambda(V)$ over $V=\bigoplus_i A_i$ (the spin-orbital modes), and this shift has a clean forcing axiom — the universal property of the exterior algebra: given any linear map $f: V \to B$ into a unital algebra with $f(v)^2 = 0$ for all $v$, there is a unique algebra homomorphism $\Lambda(V) \to B$. The condition $f(v)^2=0$ is Pauli exclusion ($v \wedge v = 0$). This is exactly parallel to how the Frobenius condition forces JOIN/CLEAVE — a real existence-of-morphism constraint, not a convenience. The forced morphism is the creation / wedge morphism $c^\dagger : V \otimes \Lambda^p(V) \to \Lambda^{p+1}(V)$.
Aufbau filling (dynamic — forces NO morphism). Orbitals fill by increasing $w=n+\ell+1$ (Paper 709). This does not force a new morphism: it is an order-theoretic selection of a state — “choose the minimal-$w$ antisymmetric $N$-vector in $\Lambda^N(V)$” — not a generator the algebra requires into existence. Nothing in the exterior-algebra axioms is violated by filling out of order; you simply get an excited state. Aufbau is a filling rule, not an operator. (Hund’s rule further constrains spin pairing within a degenerate $w$-shell — again a selection rule, not a morphism.)

Corrected statement (Opus + Gemini adversarial review, 2026-07-27). The honest result is a type correction plus a functorial embedding, not a new opcode:

The Origami ISA’s single-object Frobenius algebra describes one spin-orbital, not one atom. An atom is the image of a family of spin-orbitals under the fermionic Fock functor $\Lambda$ — standard second quantization. This functor sits between Level 1 (single orbital) and Level 2 (inter-atom bonding); call it Level 1.5. It maps single-orbital Frobenius algebras into an exterior algebra, a different kind of object from the plain tensor products at Levels 1 and 2, so Level 2 (JOIN/CLEAVE/LINK) properly acts on filtered exterior algebras, not on bare Frobenius algebras.
The creation/wedge morphism $c^\dagger$ is the Fock functor’s own canonical morphism, supplied for free by second quantization — not a framework-specific new primitive on par with JOIN. It should be named and used, but described as “the creation morphism $\Lambda$ already supplies,” never as “a new opcode we discovered.” Provisional label WEDGE (a.k.a. OCCUPY / $c^\dagger$); it is a structural morphism, not a fifth generator.
The $w$-ordering is an energy / Madelung filtration — a filtration by an integer-valued function on the modes. It is not a weight filtration in the representation-theoretic sense: $w=n+\ell+1$ additively mixes a radial and an angular quantum number and is not the eigenvalue of any Cartan element, and its well-known approximateness (Cr, Cu, Pd, La/Lu anomalies) is direct evidence it is not a genuine weight (which would be exact). Do not call it a weight or Koszul filtration — that would silently re-import the still-open SO(4,2) dynamical- symmetry claims, exactly the unearned Lie-theoretic upgrade that sank Papers 643/644/318.

This is category-theory / representation-theory research, not experimental — no x-series numerical validation is implied. Racah’s $G_2$ use for $f$-electron term structure is a related but distinct phenomenon: it governs electron-electron repulsion within a fixed configuration, not the assembly of the configuration.

Open problem (Gemini, 2026-07-27): how does JOIN (connected sum of orbital knots, Level 2 bonding) interact with the graded, antisymmetric structure that WEDGE builds at Level 1.5? Bond formation between two atoms must respect the fermionic sign rule across both atoms’ Fock spaces; the JOIN of two filtered exterior algebras is not obviously the JOIN of two bare Frobenius algebras. This consistency condition on the tower is the genuine research direction that came out of the review.

Status: open. See Paper 720 (in preparation).

Opcode duality across both levels:

Level 1 — intra-site Frobenius pair:

(RESOLVE, PROJECT) — $\Delta^\dagger = \mu$; spider identity PROJECT∘RESOLVE = id

Level 2 — inter-site Frobenius pair:

(JOIN, CLEAVE) — $\mu^\dagger = \delta$; bond formation/breaking

Compact closed pair (Level 1):

(FLIP, CUP) — dagger anti-involution and its counit sub-role; zigzag identity

TWIST and LINK are self-dual at their respective levels. FUSE has no Frobenius dual — it lives in the magmoidal extension beyond the Frobenius sector entirely.

Frobenius duality is not inversion: CLEAVE is not the inverse of JOIN (they are adjoint under the Frobenius condition, not composites yielding identity).

Why SNAP is primitive — and not derivable from JOIN + CLEAVE + FLIP: JOIN, CLEAVE, LINK, and all the other opcodes are 1-cell morphisms within a PROP — operations that act inside a fixed computational tier (a fixed free-energy basin). SNAP is a 2-cell morphism between PROPs — it marks the $\beta^*$ threshold where the system transitions between ISA tiers (H⁰ $\leftrightarrow$ H¹ $\leftrightarrow$ H²). No composition of 1-cells can produce a 2-cell: they live at different categorical levels. SNAP is therefore primitive by categorical level, not by type.

SNAP has no Frobenius dual but does have an inverse: as a 2-cell, SNAP↑ (tier promotion, $\mathcal{F} \to \mathcal{F}’$) and SNAP↓ (tier demotion, $\mathcal{F}’ \to \mathcal{F}$) are mutual inverses — together they form a 2-cell isomorphism. This corresponds physically to a reversible phase transition: increasing $\beta$ past $\beta^$ fires SNAP↑; decreasing $\beta$ back fires SNAP↓. Spin-crossover hysteresis is the pair (SNAP↑, SNAP↓) following different paths through the $\beta$-plane (nonzero imaginary part during the loop). LIESST is SNAP↑ firing while SNAP↓ is kinetically blocked at low temperature. The irreversibility of a *particular SNAP event is kinetic, not categorical: SNAP↑ and SNAP↓ are both valid 2-cells.

The physical analogue of SNAP’s categorical level: JOIN is a chemical reaction (reversible, within one thermodynamic phase, satisfies the Frobenius condition). SNAP is a phase transition (crosses between phases). No sequence of chemical reactions produces a phase transition. This is why SNAP cannot be written as FLIP∘JOIN — the two operations are not in the same categorical layer.

Consequence for PT-symmetric computing (Paper 664): the exceptional point (EP) crossing in a PT-symmetric material is a physical realisation of the SNAP 2-cell. The EP is simultaneously the $\beta^*$ snap threshold (MGE), the orbital knot crossing-change transition (torus curve changes type), and the tier boundary (ORBIT → TWIST → BIND regime). Existing PT-symmetric laser arrays are already executing RESOLVE-tier computation (tropical mode competition); adding laser driving at orbital-commensurate frequencies implements TWIST-tier gates; crossing the EP executes SNAP. See Papers 662 and 664.

The traced symmetric monoidal category (TSMC)

Combining symmetric monoidal (swap wires) with traced (close feedback loops) gives the traced symmetric monoidal category (TSMC) — the minimal structure needed to write programmes that have both parallel composition and feedback. The TSMC is:

The categorical foundation of dataflow computing (Abramsky & Duncan 2004)
The setting for Girard’s geometry of interaction (proof theory / linear logic)
The home of ZX calculus (Coecke & Duncan 2008) for qubit quantum mechanics

All opcodes except FUSE live in the free TSMC + Frobenius. FUSE requires the magmoidal extension beyond TSMC. The containment is strict:

\[\text{free TSMC + Frobenius} \;\subset\; \text{free magmoidal TSMC + Frobenius}\]

The left side generates all H⁰ and H¹ computation. The right side adds H².

What the categorical foundation does and does not buy

Stated carefully, because the easy version of each of these is circular.

The opcodes are defined, not discovered, to be general. Each opcode is a structure map of a monoidal category, so any system modelled in such a category has them — necessarily, and for the same reason every group has an identity element. This is a statement about the definition, not a finding about the world. What it buys is a vocabulary that ports: the value is in the naming and the cross-domain dictionary, not in an empirical claim.
The tier assignments follow from the definitions. The H^k tier of each opcode is read off from which level of the categorical hierarchy it requires. That makes the assignment principled rather than arbitrary — but it is bookkeeping internal to the framework, not an independent theorem about physics.
The cross-domain appearances are genuine identities. This one is real and is the strongest item on the list. When the Frobenius axiom appears in angular momentum theory, in finance and in knot theory, it is the same equation. Naming it makes a real structural coincidence explicit. Note the limit, established by direct test: this transfers structure and proof technique, not predictions — see the glossary.
Clifford simulability is a theorem; DFT’s limits are not. The Clifford case is genuine: Gottesman–Knill is a real theorem, and the H¹/H² boundary tracks it. DFT is a different matter and was previously overstated here. DFT’s failures are approximation-quality failures of exchange-correlation functionals — an active empirical question in computational chemistry, with no derivation from this framework. The two should not be presented as one consequence of the categorical structure.

The ISA is semiring-polymorphic

The Origami ISA is not tied to a specific number system. Every opcode has a semiring-polymorphic definition: the same programme computes different things depending on the semiring in which it is evaluated. The semiring is the runtime; the ISA is the programme.

Semiring	Runtime name	Hardware
$(\mathbb{R}\cup{-\infty}, \max, +)$	Tropical limit (Origami at β→∞)	CPU
$(\mathbb{R}_{>0}, +, \times)$	Gibbs / Forge ISA	GPU / TPU
$(\mathbb{C}, +, \times)$	Meld ISA	Quantum processor
$(\mathbb{Z}_p, +, \times)$	p-adic / U-MGE	PPU
$(\mathbb{A}_\mathbb{Q}, +, \times)$	Adèlic / A-MGE	PPU array + quantum

Semiring	RESOLVE computes	TWIST computes
Tropical	argmax fan-out	phase = sign flip
Gibbs	Boltzmann fan-out	Berry phase weight
Meld	amplitude fan-out	ribbon / Berry phase
p-adic	modular fan-out	Gauss sum $\tau_p$
Adèlic	adèlic fan-out	product of Gauss sums

This is why the same ISA vocabulary describes so many domains without modification: nuclear spectroscopy, quantum information, financial risk and protein folding are all well described by the same opcodes evaluated over different semirings. Described by, not executing — a folding protein is not running an instruction set, and the semiring table below is a modelling correspondence rather than a claim about mechanism. The Clifford group is the ISA’s Clifford sector evaluated in $(\mathbb{C},+,\times)$; tropical optimisation is the same ISA evaluated in $(\mathbb{R}\cup{-\infty},\max,+)$. The Gottesman-Knill theorem says the Clifford sector admits efficient classical simulation — equivalently, that the $(\mathbb{C},+,\times)$ ISA collapses to the $(\mathbb{R}\cup{-\infty},\max,+)$ ISA for Clifford-only programmes. Magic states are the programmes that do not collapse.

The semiring-programmable Origami processor is the long-term hardware vision: a single chip that accepts an ISA programme and a semiring specification at programme-load time, and routes to the appropriate arithmetic units — floating-point for the Forge ISA, NTT/Montgomery chain for the p-adic ISA, complex FMA for the Meld ISA. See forge-meld.md for the β-plane geometry that relates the semirings to each other.

The three theorems

Everything above is a dictionary. Three theorems give it teeth.

Theorem 1 — FUSE = Non-Abelian (Paper 447): A gapped topological phase has non-Abelian anyonic order if and only if its minimal ISA programme contains FUSE. This is a syntactic test: inspect the Hamiltonian for three-body terms that cannot be factored into products of two-body operators.

Theorem 2 — Universal Phase Boundary (Paper 447): For any model in the $\mathrm{SU}(2)_{q}$ family at $q = e^{i\pi\beta}$, the quantum phase transition is a TWIST failure at $\beta = \tfrac{1}{2}$, where the quantum dimension $d_{1/2}(\beta) = 2\cos(\pi\beta) = 0$ exactly.

Theorem 3 — The ISA Chain Complex (Papers 357, 571, 572): The H^k tiers are not merely a grading of computational levels. They are the homology groups of a well-defined chain complex

\[0 \;\longrightarrow\; C^0 \;\xrightarrow{\partial^0}\; C^1 \;\xrightarrow{\partial^1}\; C^2 \;\longrightarrow\; 0\]

where $C^k = \bigoplus_{|v|=k} A^{\otimes c(v)}$, $A = \mathbb{Z}[x]/(x^2)$ is the Frobenius algebra of RESOLVE/PROJECT opcodes, and $v$ ranges over the cube of resolutions of an ISA programme. The boundary map $\partial$ satisfies $\partial^2 = 0$ as a consequence of the Frobenius algebra axioms — which are exactly the pentagon identity and Frobenius condition proved in Paper 357.

The RESOLVE count is the Euler characteristic of this complex: $\chi = \sum_k (-1)^k \mathrm{rank}(H^k) = \mathrm{RESOLVE}(P)$. The Poincaré polynomial $\sum_k t^k \mathrm{rank}(H^k)$ is a strictly stronger invariant, categorifying the RESOLVE count in the same way Khovanov homology categorifies the Jones polynomial. At H²: the differential $\partial^1$ is given by the FUSE vertex — the trivalent generator of the Kuperberg $G_2$ spider (CMP 1996), whose completeness theorem provides a full diagrammatic axiomatisation of the H² tier.

Why this matters: earlier presentations of the ISA described H⁰, H¹, H² as three separate computational levels with no map between them — a graded direct sum, not a cohomology theory. Theorem 3 supplies the missing differential and confirms that the tiers are genuine homology groups. The RESOLVE count was always correct; it now has a proof that it equals an Euler characteristic, not just a heuristic count.

String diagrams

Every opcode has a string diagram — the graphical calculus of monoidal categories, popularised in quantum information by Coecke and Abramsky (2004) and in topological field theory by Reshetikhin and Turaev (1991). In string diagrams:

Wires (lines) represent objects (vector spaces, representations, anyons)
Boxes (nodes) represent morphisms (linear maps, operations)
Composition is vertical stacking (sequential)
Tensor product is horizontal juxtaposition (parallel)
Orientation of a wire matters: upward = the object, downward = its dual

The diagrams below are described in text; the LaTeX figures appear in Paper 258 (Origami Calculus) and Paper 349.

The ISA trilogy and the Baez threefold way

The three ISAs in the trilogy differ only in which number system their opcodes run over, and in the value of the inverse-temperature parameter $\beta$:

ISA	$\beta$	Arithmetic	Dyson $\beta_D$	Random matrix	AZ classes
Origami	$\beta \to \infty$	Tropical $(\max,+)$	$1$ (GOE)	Time-reversal symmetric	AI, BDI, D, CI, DIII
Forge	$0 < \beta < \infty$	Real Gibbs	$2$ (GUE)	No time reversal	A, AIII
Meld	$\beta = it$	Complex amplitudes	$4$ (GSE)	Kramers-degenerate	AII, CII, C, CI

The opcodes are the same in all three; only the number system and $\beta$ change.

As $\beta \to \infty$ the Gibbs softmax collapses to a tropical argmax — discrete, classical computation. At finite $\beta$ it is a smooth Gibbs distribution — the Forge ISA. The Wick rotation $\beta \to it$ turns real Boltzmann weights into complex amplitudes — quantum mechanics, the Meld ISA.

Behind all three sits the Ambient — the smooth $\beta \to 0$ limit in which the Gibbs measure is uniform, every path equally weighted, no decisions made. The Ambient is not an ISA; it is the smooth containing manifold from which the three ISAs are carved: the Origami is the tropical crystal precipitated from it as $\beta \to \infty$, the Forge is the thermodynamic engine between the Ambient and the crystal, and the Meld is a Wick slice through it.

This is Baez’s threefold way (2013): exactly three associative normed division algebras (Hurwitz’s theorem), exactly three consistent quantum-mechanical inner-product structures, exactly three Dyson $\beta_D$ values, exactly three ISA columns.

For a full treatment of $\beta$, the snap threshold, the Wick rotation, and the Ambient: see The Forge and Meld ISAs.

The 731-ISA extends beyond all three to the $\mathbb{O}$ (octonion) rung, adding FUSE and SPIN. See The Non-Associative Frontier.

The named ISAs

The opcode set is fixed. What varies is the regime — which β-value, which physical domain, which specialisation of the abstract opcodes is in play. Each named ISA is a specific point (or arc) in the β-plane, with a characteristic physical content, a set of patron thinkers, and a distinct informal name chosen for what it feels like to work in that regime.

Informal name	Formal name	Entry-point paper	β location	H^k reach	Patron(s)	Hook	IMAGINE count	Algebra	Graphical calculus
Origami	Origami ISA	Paper 631	all β (umbrella)	H⁰–H²	Weyl, Racah	Five-opcode open standard; tropical at β→∞, quantum at β=it	1	ℂ	ZX (spiders, undirected)
Forge	Forge ISA	Paper 419	0 < β < ∞ (real Gibbs)	H⁰–H²	Boltzmann, Gibbs	Free-energy routing; MGE soft threshold; snap at β*	1	ℂ	ZX (weighted)
Meld	Meld ISA	Paper 454	β = it/ℏ (quantum)	H⁰–H²	Shor, Grover	Complex-MGE quantum algorithm discovery; T-gate as octonion obstruction	1	ℂ	ZX (weighted)
Frog	Frog ISA	Paper 207 / 281 (623 withdrawn)	β → ∞ (exceptional)	H⁰–H³	Kauffman (731 Calculus)	Seven imaginary marks; Fano multiplication; non-associative	7	𝕆 (Moufang loop)	731 Frog Calculus (4-legged tetrahedra + ribbon-legs)

Reading the table:

The IMAGINE-count column follows the Hurwitz tower: the only normed division algebras are ℝ (0 imaginary units), ℂ (1), ℍ (3), and 𝕆 (7). Origami, Forge and Meld all live at the ℂ level — one imaginary direction, associative. The Frog ISA extends to 𝕆 (seven imaginary marks, O-calculus, non-associative Moufang loop, G₂ exceptional geometry). No ISA beyond Frog is possible: the Hurwitz theorem ends at 𝕆.

The graphical calculus column tracks the ZX hierarchy (from Paper 207): standard ZX (undirected spiders) works at ℂ; Directed ZX (oriented wires, still spiders) is needed at ℍ because quaternion multiplication is non-commutative — the Spider Theorem still holds but only if wire direction is respected; the 731 Frog Calculus is needed at 𝕆 because the Spider Theorem fails entirely (associativity is gone). Frogs replace spiders: each frog has exactly 4 legs (the 4 faces of a tetrahedron), each leg is a ribbon-leg carrying 3 Fano colours (the 3 vertices of that triangular face). The Fano colour triple specifies which octonion product eₐ · eᵦ = ±eᵧ fires at each face-weld. No leg may carry more than 4 connections because a 5-valent spider would silently invoke associativity.

Relationship to other graphical calculi

The ISA opcodes did not emerge from nowhere. Two graphical calculi were the direct predecessors, and a third recent one is worth comparing against because of the standard it sets.

ZX calculus (Coecke and Duncan, 2008) is a complete graphical language for qubit quantum mechanics built from two spider generators (Z and X) obeying the Frobenius equations. It covers RESOLVE (as SPLIT), PROJECT (as SPLAT), FLIP, and the sector sub-role of PROJECT fully, and handles TWIST partially (phase gates exist in ZX but the full ribbon/topological twist — Berry phase, anyonic spin, BKT transition — is not expressible). CUP is partially present as the compact structure (cups and caps) but the fermion-statistics interpretation ($C^2 = \pm 1$) is outside ZX’s scope. FUSE is entirely absent: ZX is strictly associative.

The 731 Frog Calculus extends ZX to the non-associative regime by adding the frog vertex — a trivalent node with a non-trivial associator, realised physically as $G_2$/octonion symmetry. The frog vertex is exactly the FUSE opcode. The two foundational papers are:

The 731 Frog Calculus, Part 1 (Paper 207) — three-dimensional spin foams, magmoidal category theory, and non-associative topology
The 731 Frog Calculus, Part 2 (Paper 281) — two-dimensional frog diagrams, ribbon-leg syntax, and $G_2$ spin foam rewriting rules

Neural circuit diagrams (Abbott & Zardini, MIT, arXiv:2604.07242, TMLR 2026) formalise deep learning architectures in a monoidal category over BorelStoch — standard Borel spaces and Markov kernels — with novel axis-stride and array-broadcasted categories handling parallel broadcasting.

It is a narrower project than the ISA and deliberately so: one domain, no cross-domain claim, no non-associativity. It is included here because of what it does that this page does not. Their framework compiles: a categorical term is turned into a runnable PyTorch module, and the compiled attention block is checked against a hand-written one (torch.equal → True). That is an executable semantics with a passing equivalence test, which is the standard a notation earns credibility by meeting, and the Origami ISA does not yet meet it.

Two of their findings bear directly on material here. Their copying morphism is not natural in BorelStoch — copying the output of a random process differs from copying the input and running it twice — which is the same obstruction that appears whenever a comonoid lacks a compatible monoid. And they keep two notions of equality apart: expressions that are mathematically equal but algorithmically distinct, because a redundant rearrangement is free in the mathematics and expensive on hardware. Any β-parameterised instruction set meant to run on more than one substrate needs that distinction.

The containment is strict:

\[\text{ZX calculus} \;\subset\; \text{731 Frog Calculus} \;\subset\; \text{Origami ISA}\]

ZX lives at H¹ (Clifford/stabiliser regime, $\mathbb{C}$-rung of the division algebra ladder). The Frog Calculus adds the H² FUSE opcode ($\mathbb{O}$-rung). The full Origami ISA extends both to all physical domains — spectroscopy, molecular computing, financial risk, climate economics — running the same categorical morphisms on different hardware.