The One-Axiom Pattern
Handout · brainstorm register · nothing sealed · ΔM = 0 What the Hodge and Navier-Stokes pairs have in common, whether the Riemann Hypothesis and P versus NP admit the same shape, what the three terminal tokens turn out to be labelling, and how that bears on the census conjecture.
Date. 2026-09-11. Everything below is at brainstorm grade unless a grade is printed beside it. Two findings are refusals and are marked as such.
1 · The pattern, as the two seated pairs exhibit it
Two papers now stand in the same shape and neither was built to match the other.
Hodge. Axiom W asserts that every Hodge class on every smooth projective variety carries a semiregular witness: a local complete intersection subscheme, or a coherent sheaf, whose semiregularity map is injective. Grant it and the deformation argument carries the class along its locus and the conditional theorem delivers the conjecture. Decline it and the block stands at rigid classes, where the locus is a point and the relocation family is empty.
Navier-Stokes. EAD asserts a data-uniform modulus dominating the alignment defect over the admissible class. Grant it and the continuation theorem at fixed exponents forbids finite-time breakdown. Decline it and the block stands at the scale-critical locus, where the relocation family is a full group and is inert.
The pattern in one sentence. One axiom, accepted, proves the conjecture; the same axiom, examined, is exactly what the formal-alone register cannot supply. Two papers, not one: a conditional proof and a termination result, and the second is the mirror of the first.
2 · Does it extend? The first answer is no, and the no is informative
The naive extension fails, and testing it is what locates the real one.
| object | token | halts in | what is missing |
|---|---|---|---|
| the Riemann formal string | [Ξ₀] | the reader | a bit, the truth-sign no readout of the class can carry |
| the P versus NP formal string | [Ø₀] | the terrain | a ground, a fixed locus for any separating instrument |
| Hodge at rigid classes | [Ψ] | the witness | a cycle, a constructed object |
| Navier-Stokes at the critical locus | [Ψ] | the witness | a modulus, a constructed object |
Only the last two are missing a constructed object, and that is precisely what an axiom of the Hodge or EAD kind supplies: such an axiom asserts the existence of the missing construction, so granting it discharges the gap by decree.
The Riemann string is missing a bit, and an axiom asserting the bit is the conjecture rather than a premise for it. The P versus NP string is missing a ground, and an axiom asserting a ground on a fixed-point-free terrain asserts what the terrain refutes. The pattern does not transfer to the strings.
3 · Where it does extend: the constructive route is a different object
The strings are not the only objects these conjectures carry. Each also carries a constructive route: an axiom paired with a short conditional theorem. Ask of each conjecture whether a single existence claim about a constructed object yields it in a few lines, and the answer is yes in all four cases.
Riemann · Hilbert–Pólya. There exists a self-adjoint operator on a Hilbert space whose eigenvalues are exactly the numbers t such that ½ + it is a nontrivial zero. Grant it and the conjecture follows in two lines, since a self-adjoint operator has real spectrum and every zero therefore has real part one half. A century old, never supplied.
P versus NP · GCT obstruction existence. For every n there exists a representation-theoretic obstruction separating the permanent orbit closure from the determinant orbit closure at the stated size. Grant it and the separation follows, since the obstruction is by construction a witness that no small determinant expresses the permanent. Mulmuley and Sohoni's program, never supplied.
The four-property test.
| axiom | universally quantified existence | construction-supplied, reading-barred | grant yields the conjecture | strictly stronger than the conjecture |
|---|---|---|---|---|
| Axiom W | yes | yes | yes | yes |
| EAD | yes | yes | yes | yes |
| Hilbert–Pólya | yes | yes | yes | yes |
| GCT obstructions | yes | yes | yes | yes |
| Weil positivity | no | no | yes | no |
Weil positivity is the instructive failure. It is equivalent to the Riemann Hypothesis, so granting it is granting the conjecture. That is the name-strip gate firing, and it is why Hilbert–Pólya and not Weil is the right candidate: the operator is strictly more than the zero locations, and a premise must be strictly stronger than what it buys or it is a restatement wearing a premise's clothes.
4 · What the tokens were actually labelling
This reorganizes the table rather than extending it.
| problem | the formal string, read directly | the constructive route | its axiom |
|---|---|---|---|
| Riemann | [Ξ₀] reader: the sign unreadable by theorem | [Ψ] witness: the operator unsupplied | Hilbert–Pólya |
| P versus NP | [Ø₀] terrain: rootless, walled by three courts | [Ψ] witness: the obstructions unsupplied | GCT obstructions |
| Hodge | not separately typed in the register | [Ψ] witness: the cycle unsupplied | Axiom W |
| Navier-Stokes | not a truth-string of that kind | [Ψ] witness: the modulus unsupplied | EAD |
Every constructive route blocks at [Ψ], and this is forced rather than observed. A constructive route's missing item is by definition a constructed object, and the formal-alone register never authors one. So [Ψ] is not a per-problem label at all: it is what a constructive route always blocks at.
And [Ξ₀] and [Ø₀] label the direct reading only. Three tokens, two columns, and the columns were never distinguished before.
A consequence worth stating plainly. A problem can carry two blocks at once, of different species, on different objects. The Riemann Hypothesis is reader-blocked on its string and witness-blocked on its Hilbert–Pólya route, and neither block is the other's restatement. Reporting one and calling it the block on the problem is a category error the two-column table forecloses.
5 · The census conjecture, cross-tabulated
The ICH question was whether resistance to a principal method class implies a defeater under which that class is even. The tokens bear on it, and not in the direction one expects.
| object | token | ICH cell | the family |
|---|---|---|---|
| Riemann string | [Ξ₀] | A, family exists and readout even | the reflection s ↦ 1−s |
| P versus NP string | [Ø₀] | A, family exists and technique constant | the oracle family |
| Navier-Stokes unforced | none, provable | C, no family | every candidate map tested and failed |
| social choice | none, provable | B, family exists and class odd | transposition of alternatives |
The Riemann string and the P versus NP string sit in the same ICH cell and carry different tokens. So the two taxonomies are independent axes rather than one classification in two vocabularies. ICH asks whether a defeater exists and whether the class is even, which is a fact about a method relative to a symmetry. The token asks where the halt sits, which is a fact about what is missing. A problem has both coordinates and neither determines the other.
That is what the tokens were telling us about ICH, and it is neither support nor refutation: the census conjecture is a claim about one axis, and the other axis can neither confirm nor refute it. Anyone hoping the token family would settle the census should stop, and anyone fearing it would sink it can stop too.
One thing the cross-tabulation does buy. It gives the census a clean statement of its own scope: ICH ranges over the method axis and says nothing about supply. A problem may have no defeater and still be blocked, if what it lacks is a construction rather than a readout. That is exactly the Hodge and Navier-Stokes situation, and it means the census must be stated over method-blocked problems and not over hard problems in general.
6 · Two refusals, recorded because a brainstorm that refuses nothing has tested nothing
Refused: that RH's string admits a Hodge-style axiom. It is missing a bit, not an object, and an axiom asserting the bit is the conjecture. The extension found in §3 is on a different object and the difference is not cosmetic.
Refused: that the token family confirms the census. Section 5 shows the axes independent, and a coordinate cannot confirm a claim about a different coordinate. This is the second time in two days the census has been offered a confirmation that dissolved on testing, and the pattern in that is worth noticing: the census is more isolated than it looks, and its only real evidence would be the natural-proofs subsumption question already carried as R-007.
7 · What follows, if anything, and at what grade
A candidate general statement, at brainstorm grade and offered to be attacked. Every major open conjecture carries at least one constructive route, its route blocks at [Ψ], and the route's axiom is the most informative single object about the conjecture's difficulty, because it names exactly what nobody can build. The statement is unfalsified over four cases and four cases is not evidence for a universal.
What would test it. Find a major conjecture with no constructive route, or with a route whose axiom fails the four-property test in a way Weil positivity does not. The Poincaré conjecture before Perelman is the obvious place to look, since it is now closed and its route is inspectable with hindsight.
What would sink it. A constructive route whose block is not at the witness. The token table's own indexing predicts this cannot happen, so exhibiting one would break the indexing principle rather than merely the generalization.
8 · Preparation for the register
Nothing below is seated and the seating is the architect's act. What follows is what a seating would carry, so the decision can be made on a printed plan.
Candidate one, the two-column token table. The present table indexes tokens by halting locus with one row per problem. The corrected form indexes by locus and distinguishes the formal string from the constructive route, so a problem occupies two cells. This is a correction to a resident table and not a new coordinate; it would ride as an amendment to the token family's own card.
Candidate two, the one-axiom pattern as a named structure. Four instances, two seated as paper pairs and two identified here from the literature. A coordinate would carry: the pattern statement, the four-property test, the Weil-positivity negative control, and the two refusals of §6. Grade: structural on the pattern, corroboration on the four instances, and the universal claim of §7 explicitly not carried.
Candidate three, the independence of the two axes. The cross-tabulation of §5 is the sharpest single result here and it is a scope correction to the census conjecture: ICH ranges over the method axis and is silent on supply. This belongs beside the census wherever the census is seated, since it changes how the conjecture must be stated.
What must not be seated. The Hilbert–Pólya and GCT identifications are readings of other people's programs, not results; they are cited, not claimed. The universal statement of §7 is a candidate and fails the object-level gate anyway, being meta-work. And nothing here touches whether any of the four conjectures is true.
Anti-duplication note. Before any of the three is forged, the register is resolved current from git and searched for a resident coordinate on the token family and on the census. Candidate one in particular is likely to be an extension of an existing card rather than a new seat, and the Prime Rule governs.
9 · Open items this handout leaves
The natural-proofs subsumption question, carried as R-007 and unchanged by anything here, is still the census's best available witness. The census completeness question, R-001, is now better scoped and no closer to settled. The third-ceiling reading and the Tongue exemption, R-006, are untouched: this handout is about supply and that one is about readout, which §5 shows are different axes.
And one item this handout adds: whether the Poincaré conjecture's pre-Perelman constructive route fits the pattern, which is the cheapest available test of §7 and needs a literature reading rather than a computation.