STATUS (2026-08-04). This page was titled “The H^k stratification is not an analogy”. The title asserted more than the evidence supports and has been changed; the body, which already distinguished Tier A/B/C precision and listed what it does not claim, was more careful than its own headline.
Three of the four papers this page cited as support have been withdrawn — 533 (ISA survey), 557 (MCMC ladder) and 558 (EVT ladder). 420 survives but its routing algorithm does not; see the caveat below. The Tier B evidence line citing “Papers 420–421” should be read with that in mind, and 421 has no DOI on record at all.
What direct testing found. Four experiments this session asked whether a shared categorical structure yields a shared prediction, and none did: the Grassmannian path metric reproduced the energy gradient (ρ = 0.95), redox-ladder curvature lost to plain d-electron count, Yutsis diagram topology failed to predict the good spin-coupling tree (which is set by the exchange couplings), and the Weyl coordinate turned out to be a monotone function of the occupation number. A fifth test confirmed a transfer — spin-coupling multiplicities are tree-independent, exactly as the Frobenius spider normal form requires — but that fact is Wigner-era standard, so the framework predicted something the target field established in the 1930s.
The honest reading: the stratification is a real and useful organising principle, and the semiring-polymorphic claim below is a genuine theorem. Shared structure transfers vocabulary and proof technique; it does not, on the evidence so far, transfer new predictions.
The H^k stratification: what it is and is not
The same three-tier structure — fixed points, local phase corrections, global topological obstructions — appears in MCMC sampling, extreme value theory, causal inference, quantum algorithms, molecular chemistry, and a dozen other fields. Is this a coincidence? An analogy? Over-selling? This page gives the honest answer.
CAVEAT on Paper 420’s routing algorithm (added 2026-08-04). 420 is live and its framing — that hardness is graded rather than binary — is worth keeping. But its stated “key practical contribution”, computing the Euler characteristic χ = |V| − |E| + |F| in O(n+m) and inferring the Betti numbers to route an instance, does not work. χ = b₀ − b₁ + b₂ − … is one equation in several unknowns. A circle has (b₀,b₁,b₂) = (1,1,0) and a torus (1,2,1); both have χ = 0, but the first has H² = 0 (rung 1, polynomial) and the second H² ≠ 0 (rung 2, conjectured NP-hard). Same χ, opposite routing decision. χ cannot determine the rung, so the pre-diagnostic as described is unsound. Treat the ladder as a classification scheme, not as a polynomial-time oracle for hardness.
The claim
The H⁰/H¹/H² stratification is a universal organisational principle, not a metaphor. The ISA is one precise instantiation of it. Many fields have independently discovered the same skeleton, because any mature field studying transformations on spaces eventually needs to distinguish: fixed points (H⁰), local phase corrections (H¹), and global topological obstructions (H²).
The mappings from field to field range in precision. Some are exact algebraic identities. Some make quantitative predictions that experiments verify. Some are useful taxonomic language. The three cases are genuinely different, and it matters which is which.
Why the stratification keeps appearing
The three tiers are not arbitrary. They follow from the structure of cohomology itself:
-
H⁰ measures connected components — the coarsest invariant, the fixed points of the action. Classical equilibria, tropical optima, stationary distributions, and ground states all live here. An H⁰ object is one that the action cannot move.
-
H¹ measures 1-cycles that are not boundaries — the failure of local consistency to imply global consistency. Berry phases, monodromy, MCMC proposal corrections, option convexity, causal interventions, and tail-index power laws all live here. An H¹ object is one where going around a loop leaves a trace.
-
H² measures 2-cycles that are not boundaries — global topological obstructions that cannot be removed by local surgery. Non-Abelian holonomy, FUSE operations, compact-support extremes (Weibull), counterfactual twin-world loops, and topological quantum phases all live here. An H² object is one where a topological invariant forbids a continuous deformation.
Any field that studies transformations on spaces will eventually encounter all three. Algebraic topology named them first. The ISA gives them opcodes (RESOLVE / TWIST / FUSE). Physics calls them vacuum / perturbative / non-perturbative. The convergence is not a coincidence — it is the same mathematical fact, re-derived independently in each field.
The boundaries between tiers are as important as the tiers themselves
The transitions H⁰ → H¹ and H¹ → H² are not smooth crossovers — they are snap events at critical thresholds β, where a Gibbs distribution crystallises from a warm exploration into a hard commitment. These boundaries are computable: in chemistry, the β snap at Grassmannian angle θ_G ≈ 20° marks the point where single-reference MO theory fails and multi-reference CASSCF becomes mandatory; in MCMC, β* is the inverse temperature at which the acceptance rate transitions between the 0.234 (H⁰) and 0.574 (H¹) optima; in finance, the H¹/H² boundary is the threshold at which interbank cycle topology becomes globally inconsistent — a systemic crisis rather than a local stress event.
The snap events are themselves a universal feature. Knowing which tier a system sits in is the coarse classification; knowing where it sits relative to β* is the fine-grained prediction. The ISA makes both computable from the same underlying object (the β-deformed partition function), which is why the same snap threshold appears in such different physical systems.
A precision taxonomy
Not all ISA mappings are the same kind of claim. We distinguish three tiers:
Tier A — Exact algebraic identities
The ISA prediction is a theorem. The mapping is an isomorphism or near-isomorphism, and experiments verify it to numerical precision.
| Field | ISA prediction | Status |
|---|---|---|
| Quantum information (Papers 469–473) | TV = 1 iff stabiliser state; 9 SWAP-classes; Casimir c₂(proj) = 2 universally for C_{2k+1} | All experiments pass; Fano exceptionality k = 3 proved |
| Molecular chemistry (Papers 488–491) | Aufbau/Hund/Taube rules = RESOLVE/TWIST/FUSE theorems; tropical DFT 20/20 on SCO benchmark | x491a–d: 100% on SCO; Wigner vertex theorem proved |
| Protein proofreading | H⁰ × H¹ × H² gives 10⁹/10⁶/10⁴ fidelity for Pol III/RNAP/ribosome | Structural argument from known biochemistry |
| Proton stability (Paper 545) | Colour singlet = closed Fano RESOLVE; ΔE_Fano = 0.9–1.3 GeV | x545a: Routes A/C agree × 1.9 |
At Tier A, the ISA language is not introducing a new perspective on the field — it is proving theorems that did not have proofs before, or making predictions with specific numerical values that can be checked.
Tier B — Precise claims, quantitative predictions
The ISA framing makes specific quantitative predictions that can be verified independently of the framework. Individual claims are exact; the meta-narrative (that the field “implements” the ISA) is a chosen framing rather than a derived result.
| Field | ISA prediction | Status |
|---|---|---|
| MCMC | Optimal accept rates 0.234 (H⁰) < 0.574 (H¹) < 0.651 (H²); monotone across tiers | Roberts-Rosenthal / Sherlock-Roberts theorems independently proved these; ISA explains the monotonicity |
| Extreme value theory | GEV shape ξ = β-deformation parameter; Gumbel = tropical fixed point (exact: log(-log Λ) = -x) | Gumbel-as-tropical is exact; ξ-as-β is a structural parallel, not yet derived |
| Information geometry | α-connection = TWIST parameter; Uhlmann holonomy = FUSE; EM algorithm = RESOLVE/PROJECT cycle | Amari’s formalism independently derives the same three tiers; ISA names them |
| Quantum algorithms (Papers 420–421) | Shor = H¹ (mana = 0); Grover intermediate states Clifford-simulable | x472a–c, x473b pass; predictions made before experiments ran |
At Tier B, the ISA is a useful lens that organises existing results and sometimes generates new predictions. The predictions that have been checked have passed. The framing does not introduce errors but should not be confused with derivation.
Tier C — Taxonomic / organisational language
The H^k labelling provides useful scaffolding for an existing hierarchy that the field already knew was hierarchical. No quantitative prediction is added beyond what the field already knew; the ISA provides a cross-domain translation layer.
| Field | ISA framing | What it adds |
|---|---|---|
| Causal inference (Paper 559) | Pearl’s ladder (seeing/doing/imagining) = H⁰/H¹/H² | Names the tiers; clarifies why H² (counterfactual) is strictly harder than H¹ (interventional); suggests fairness hierarchy |
| Ergodicity economics (Paper 549) | GBM = multiplicative RESOLVE; Kelly = β* snap | Connects to ISA β-ladder; no new finance predictions beyond Kelly |
| Incentive geometry | Tragedy of commons = H¹ RESOLVE open; cap-and-trade = H¹ closure; climate clubs = H² FUSE | Provides language; the policy conclusions were already known |
At Tier C, the ISA framing is legitimate science — taxonomy papers are real contributions — but claims should be written as “the H^k framework provides useful organisational language” rather than “we have shown that causal inference is an instance of the ISA.”
What we do not claim
-
We do not claim that all three-tier structures are the same thing. The fact that Pearl’s causal hierarchy has three rungs and H^k cohomology has three tiers does not make them identical. The shared skeleton (H⁰ = coarse invariant, H¹ = local correction, H² = global obstruction) is a genuine structural parallel, not an isomorphism. The flesh is different in each field.
-
We do not claim the ISA originated the stratification. Algebraic topology named H⁰/H¹/H² long before the ISA. What the ISA adds is: (1) opcodes that make the tiers computable, (2) the β-plane connecting them as a continuous deformation parameter, and (3) a cross-domain dictionary that lets results transfer between fields.
-
We do not claim all mappings are equally precise. See the Tier A/B/C taxonomy above. The Fano commutation structure (Tier A) and the MCMC acceptance rates (Tier B) are different kinds of claim from the causal inference labelling (Tier C).
-
We do not claim the ISA is complete or final. The H^k tower is in principle infinite. H³ would require a quantum gravity computer. The β-plane may have structure we have not yet mapped. The survey paper is a progress report, not a closed theory.
The legitimate version of the strong claim
The opcodes page used to say: “They are not analogies. They are the same categorical morphisms, running on different physical hardware.” That sentence was withdrawn on 2026-08-04 and should not be quoted as support.
What survives of it is narrower and genuinely true: the semiring-polymorphic claim. The same programme really does compute different things over different semirings — shortest paths over (min,+), partition functions over (+,×), amplitudes over ℂ — and that is a theorem, not a metaphor (see The ISA is semiring-polymorphic). That is a statement about one programme run over different arithmetics, which is much stronger evidence than two fields having a morphism with the same name.
The weaker but broader claim is: the H^k stratification is a universal organisational principle, and the ISA is the first systematic attempt to give it a unified computational syntax. Each field rediscovered the three tiers independently because the tiers are forced by the mathematics. The ISA’s contribution is to make the cross-field translation explicit, to give the tiers names that travel across domains, and — in the Tier A cases — to use the unified language to prove results that field-specific language had not reached.
That is a significant contribution. It does not require claiming more than it is.
Open questions
-
Can the Tier C mappings be promoted to Tier B? For causal inference, this would require a quantitative prediction — e.g., a new bound on the sample complexity of counterfactual estimation derived from the H² FUSE structure — that can be checked independently. If such a prediction can be made and tested, Paper 559 moves from taxonomic to predictive.
-
Is there a meta-theorem? Specifically: is there a theorem that says “any field whose objects form a traced monoidal category will exhibit a three-tier H^k stratification”? If yes, the universality of the stratification is a theorem of category theory, not an empirical observation. This would be the cleanest possible justification for the programme.
-
Where does the stratification fail? The strongest evidence that a classification is real is knowing where it breaks. Are there fields where the H⁰/H¹/H² decomposition does not apply, or applies with more than three tiers? Identifying these would sharpen the claim considerably.
See also: The ISA Opcodes · β is a coordinate · Quantum speedup has a cohomological address · Paper 533 — ISA Survey