What did axiom-explorer actually find? A conjectural cardinality envelope across three branches of mathematics

A few days ago I published the methodology paper that describes axiom-explorer, the LLM-assisted cross-search workflow I built. Today the companion paper is out: the actual case-study output, the mathematical thing the workflow surfaced. It is now on Zenodo under DOI 10.5281/zenodo.20184660.

If you only have time for one sentence: the conjecture says that something Haine and his collaborators recently computed for one geometric example looks like an instance of a much wider cardinality pattern that also appears in model theory and in set theory, and the synthesis seems not to be written down anywhere as such.

That is the whole claim. Let me unpack what each piece means without requiring you to be a working mathematician.

What is the object we are bounding?

In late 2025, a six-author group — Peter J. Haine, Tim Holzschuh, Marcin Lara, Catrin Mair, Louis Martini, and Sebastian Wolf — posted a 103-page paper titled The condensed homotopy type of a scheme. They define a new way of attaching a “fundamental group” to a geometric object (a scheme, in algebraic-geometry terms). The construction uses condensed mathematics, a recent (2019) re-foundation of topology by Clausen and Scholze that is currently very active.

Calling it a “fundamental group” is the standard analogy: this is the algebraic gadget that records loops you cannot continuously deform away in the space. The classical version (the étale fundamental group) has been studied for decades. The new condensed version is finer; it sees distinctions the classical one misses.

The paper does a careful calculation for two simple examples:

For the projective line over the complex numbers, the new fundamental group has exactly $2^c$ elements (where $c$ is the size of the continuum, i.e. the number of real numbers).
For the projective line over the rational numbers, the same group has at most $c$ elements.

In both cases the size depends only on the size of the base field. Doubling that number once gets you to the upper bound.

That “double once” is the pattern the conjecture proposes is universal.

The conjecture in one line

Let $\kappa_{\mathcal X}$ measure the size of a geometric object $\mathcal X$ in a careful sense (this is a technical detail of the paper). Then the conjecture says:

\bigl|\,\text{condensed }\pi_1(\mathcal X)\,\bigr| \;\le\; 2^{\kappa_{\mathcal X}}

The bound: “you double the size of the object once, and that is as big as the fundamental group can get”. The Haine et al. attestations match exactly. The first attestation attains the bound; the second sits inside it.

That is the geometric side of the conjecture.

Why this might be more than a one-off

In a second 2026 paper, Haine alone announced something striking: that the same construction, applied to different inputs, recovers two apparently unrelated objects:

On one hand, the pro-étale fundamental group of a scheme — the geometric side above.
On the other hand, the Lascar group of a complete first-order theory — an object from model theory, a branch that has nothing to do with geometry on the face of it.

The Lascar group is the model-theoretic analogue of an absolute Galois group. It measures how many automorphisms of a “monster” model exist up to a certain notion of finiteness. Classical model theory has known for decades that the Lascar group of a countable theory has at most $2^{|T|}$ elements where $|T|$ is the size of the theory. That is exactly the same cardinality envelope as the geometric one.

The conjecture promotes that observation: the same cardinality envelope $2^{\kappa}$ governs both sides, not by coincidence, but because they are both instances of one construction.

A third, more analogical strand: in set theory and condensed mathematics, several recent papers (Clausen-Scholze on Whitehead’s problem in condensed abelian groups, Bergfalk-Lambie-Hanson-Šaroch on its forcing-theoretic proof, Bannister-Basak on a geometric morphism from the Solovay-model topos to pyknotic sets) describe situations where the cardinality of the real line controls the regime in essentially the same way. This is not a direct attestation of the conjecture — the paper is careful to distinguish related cardinal phenomena from direct attestations — but it is the third leg that makes the synthesis feel structural rather than coincidental.

What the paper claims, exactly

The paper carries a four-level confidence ladder that I described in the methodology paper post:

L0 — verified mechanically.
L1 — strong: standard cited result.
L2 — plausible: argued, not directly verified.
L3 — speculative: a guess.

The main conjecture sits at L2. The structural-argument sketch is explicitly not a proof in the paper; §5 says so directly. The saturation half of the conjecture (when does the upper bound become an equality?) sits at L3 and is framed as an open question, not as a claim.

Two of the four motivating attestations are direct geometric calculations from the Haine et al. paper, with full bibliographic backing. The other two — the model-theoretic and set-theoretic ones — are placed in their own section as analogical evidence, with an explicit caveat that they are not direct instances and may be wrong as attestations of the conjecture even if the conjecture itself is right.

This framing is the whole point. The paper is a registered conjectural observation, not a theorem. The role of registration is to make the question existable in a citable form so that specialists can confirm it as folklore, refute it with a counterexample, or refine it. Any of those three is a useful outcome.

What would falsify it

Section 8 of the paper lists three concrete falsifiers. Roughly:

A smooth quasi-projective scheme whose condensed fundamental group exceeds $2^{|k|}$ where $k$ is the base field.
A complete first-order theory whose Lascar group exceeds the $\kappa$ -bound it predicts on the model-theoretic side.
A condensed-extension cardinality exceeding $2^{|\mathbb{R}|}$ in any of the set-theoretic settings the related cardinal phenomena section cites.

I could not construct any of these. If a reader can, the conjecture is wrong, and I would be glad to know it.

What this paper is not

I have to be repetitive about this because the methodology paper got this part right and the conjecture paper has to match:

It is not a proof. It is a falsifiable conjecture with explicit evidence.
It is not a new construction. The classifying-anima unification on which it sits is announced by Haine in his 2026 paper. The cardinality calculation it builds on is due to Haine et al. in 2025. The Lascar-group identification it uses is due to Campion, Cousins, and Ye in 2021. The condensed-Whitehead context is due to Clausen-Scholze and follow-ups.
The synthesis contribution is narrow. It is the formulation of a uniform cardinality envelope across those previously-separate constructions, as a single line that is, as far as I could search, not stated as such in any single source.

That is a smaller claim than it could sound like. It is also, plausibly, the right size of claim for the workflow that produced it.

The trajectory of the manuscript

Five rounds of AI peer review refined this paper before it received its DOI:

v1 → v1.1: substantive math fixes (the size invariant $\kappa$ , the cardinality bound on free profinite groups, the Lascar group of $\mathrm{ACF}_0$ , DLO, the falsifiability formulation).
v1.1 → v1.2: more substantive math fixes ( $\kappa$ refined to use topological weight; the Campion-Cousins-Ye citation corrected, since v1.1 referenced a paper on turbulence instead of the right paper; the $\mathrm{ACF}_0$ attainment claim corrected).
v1.2 → v1.2.1: precision (removed a potential circularity in the definition of $\kappa$ , softened a claim about profinite weight, corrected example numbers in citations, version-stamp residues cleaned).
v1.2.1 → v1.2.2: DOI embedding and explicit citation of the companion methodology paper.

The full delta is in the Acknowledgments section of the manuscript. The methodology paper post explained how this loop works.

Why I am genuinely curious about the outcome

There are three things I would be happy to learn:

The conjecture is folklore — known to specialists in condensed mathematics, in model theory, or in pyknotic-flavoured set theory. In that case, I would update the preprint with the canonical reference and the registration becomes a useful pointer.
The conjecture is wrong — and there is a specific counterexample I missed. In that case, the paper is corrected and the case study becomes a clean negative result, which is fine.
The conjecture is genuinely new — in which case there is a specialist somewhere who can either prove it or argue why it is open, and the registration becomes the first citation for whatever follows.

All three of those are interesting outcomes. The whole point of registered conjectural observations is to make the question publicly askable.

If you are a specialist reading this

I would be genuinely grateful if you read the preprint with a critical eye and let me know what you find. The Falsifiability section, in particular, is explicit about what would refute the conjecture. The companion methodology paper describes the workflow that produced it, including the four hard stop rules and the AI peer-review loop.

If you find the conjecture wrong, or already known, or just not very interesting, that is the most useful thing you can tell me. I will update the preprint accordingly, with credit.

The conjecture paper is open access (CC-BY-4.0) at DOI 10.5281/zenodo.20184660. The companion methodology paper is at DOI 10.5281/zenodo.20184068. The repository is at github.com/Dredok/axiom-explorer.

What is next

If the conjecture survives expert contact (in any of the three forms), the next experiment in my queue is a second case study with a deliberately different seed quadruple. The point is to see whether the workflow surfaces a different shape of finding from a different starting point, or only the same shape. If it is only the same shape, the workflow is narrower than I hoped. If it is a different shape, that itself is informative.

Either way, I will write about it here.