# We checked RoundSquare’s zero-axiom Lean proof against the Clay Yang–Mills statement

Abe, part of the Honesty Tools team | Honesty.Tools | 5 November 2026

We checked [RoundSquare](https://github.com/Dr-Logvinovich/round-square), the artifact behind [this claim](https://x.com/dr_logvinovich/status/2118380634343105007) of a zero-axiom Lean 4 negative resolution of the Clay Yang–Mills mass-gap problem on non-compact \(\mathbb{R}^4\).

Here’s what held: the Lean kernel seal held. The repository builds, and **114** `#print axioms` checks report no axiom dependencies. Here’s what didn’t: the claim as a negative resolution of Clay Yang–Mills. The theorem lives inside an author-defined `Arena` / `CanonicalPhysics` ontology; it is not a faithful formalization of the existence-and-mass-gap statement in **Jaffe–Witten §4**.

That distinction is the finding. A kernel can certify a proof without certifying that its statement means what the headline says.

![Dark two-panel card. Left (held): 114 #print axioms all zero-axiom; lake build ok; sorry/admit/native_decide/axiom decls 0. Right (did not): Arena/CanonicalPhysics ≠ Jaffe–Witten §4 — no gauge group G, flag not QFT on R4, OS as five flags, class-relative no-go.](/media/blog/roundsquare-clay-ym-reading-note/c10_card_16x9.webp "Kernel seal held. Clay faithfulness did not. Honesty.tools reading note · RoundSquare @ 70128f3.")

## What the repository claims

The README and file header describe a:

> “World's First Zero-Axiom Machine-Verified Negative-Result Resolution (Impossibility Theorem) of the Clay Millennium Yang–Mills Problem on Pure Non-Compact \(\mathbb{R}^4\).”

The core displayed implication is:

\[
\mathrm{CanonicalPhysics}(\mathbb{R}^4) \land \mathrm{OS} \land (\Delta > 0) \Longrightarrow \bot.
\]

A companion claim says the mass gap is “automatic” on the compact manifold \(S^4 \times T^3\). That story is mostly in `YM_V5.pdf`; the Lean development’s Section 30 instead derives `¬ nonCompact` under assumed Strocchi-style laws and `ClayPair`.

## What we ran

We audited commit [`70128f365357ba6ab661c0e8a822c05d1f5c2190`](https://github.com/Dr-Logvinovich/round-square/tree/70128f365357ba6ab661c0e8a822c05d1f5c2190), using the repository’s `leanprover/lean4:v4.11.0` toolchain.

| Check | Result |
|---|---|
| Local `lake build` | Success, exit 0, approximately 8 seconds |
| Local `lean RoundSquare.lean` | Success, exit 0 |
| `#print axioms` directives | 114 |
| Local “does not depend on any axioms” results | 114 |
| GitHub Actions run `42075796526` at commit `70128f3` | Success: “Lean 4 Zero-Axiom Kernel Verification” |
| `axiom` declarations in code | 0 |
| `sorry` / `admit` proof terms | 0; the words occur only in comments/documentation |
| `native_decide` | 0 |
| `unchecked` / `trusted` / `opaque` / `unsafe` | 0 |
| Mathlib imports or dependency | 0; standalone Lean file |
| `Classical.choice`, `propext`, `Quot.sound` dependencies in audited theorems | None, according to `#print axioms` |

This is a real, locally reproducible, Mathlib-free Lean artifact. The mechanical zero-axiom claim held at this commit. It is not a claim we checked only by reading the README: we built the code and ran Lean’s axiom reports.

## Faithfulness: RoundSquare versus Jaffe–Witten §4

The [official Clay statement](https://www.claymath.org/wp-content/uploads/2022/06/yangmills.pdf), by Arthur Jaffe and Edward Witten, asks in §4:

> **Yang–Mills Existence and Mass Gap.** Prove that for any compact simple gauge group \(G\), a non-trivial quantum Yang–Mills theory exists on \(\mathbb{R}^4\) and has a mass gap \(\Delta > 0\). Existence includes establishing axiomatic properties at least as strong as those cited in [45, 35] (Streater–Wightman; Osterwalder–Schrader).

The comparison is about the statement given to Lean, not about whether Lean checked that statement correctly.

| Clay requirement | What RoundSquare.lean contains |
|---|---|
| A nontrivial quantum Yang–Mills theory for **every compact simple gauge group \(G\)** | No \(G\) parameter; no connection, curvature, Wilson loop, or continuum measure |
| Existence on **non-compact \(\mathbb{R}^4\)** | `nonCompact : Prop`, a flag in the author’s `Arena` structure |
| A mass gap \(\Delta > 0\) in the spectrum of the Hamiltonian | A `strictGap` proposition and toy spectral constructions, not the spectrum of a self-adjoint Hamiltonian on a QFT Hilbert space |
| Axioms at least as strong as **Osterwalder–Schrader / Wightman** | `OS` is five flags; stub correlators stand in for OS axiom schemas over distributions |
| The Clay existence-and-mass-gap problem | A proof that `OS ∧ Gap` is incompatible with a custom `CanonicalPhysics` / `InCompleteR4GaugeClass` class |
| Compact manifolds as an intermediate problem, not a replacement for \(\mathbb{R}^4\) | Lean derives `¬ nonCompact` under assumed laws; the \(S^4 \times T^3\) “automatic gap” promotion is mostly PDF-side |

`Arena` is a structure of proposition fields. `Gap A` is `A.strictGap`, and `ClayPair A` is `OS A ∧ Gap A`. The structures `MethodI_Physics` through `MethodIV_Physics` include the physics implications themselves as fields—for example, an arrow from gap and Gribov-copy assumptions through restriction and complex poles to negated reflection positivity. `CanonicalPhysics` packages those four methods. The no-go theorem then chains the packaged arrows.

That is valid Lean. But it establishes the incompatibility **relative to the encoding**. It does not formalize Singer, Strocchi, or Gribov–Zwanziger from first principles, and it does not negatively resolve Jaffe–Witten’s QFT existence problem.

The compact/non-compact distinction is named correctly in the headline. The faithfulness miss is not that the repository forgot non-compactness; it is that the formal statement does not encode Clay’s QFT object.

## The useful part: `ArenaDefect` partially documents the bridge gap

The file contains an unusually candid internal check. Sections 25–28, including `ArenaDefect_maximal_honest_statement` and `ArenaDefect_v4_status`, machine-verify that:

- class-relative exclusion holds inside `CanonicalPhysics` / `InCompleteR4GaugeClass`;
- bridges from “every physically admissible YM” or “every YM axiom package” into that class are **not logical truths** in the current ontology (`ArenaDefect_v4_physical_bridge_is_not_a_logical_truth` and `ArenaDefect_v4_abstract_bridge_is_not_a_logical_truth`);
- an artificial `UnconstrainedCompatibleModel` satisfies the bare conjunction outside that class.

Section 29 then defines `PhysicallyAdmissibleR4YangMills` essentially as the infrared dichotomy that supplies the bridge, proves `PhysicalAdmissibilityBridge` for that predicate, and obtains `unconditional_prize_level_physical_no_go` relative to it.

So “unconditional” here means unconditional after admissibility has been identified with the obstructing class’s closure properties—not unconditional over every construction Clay would accept. The author’s `ArenaDefect` sections partially document this gap before Section 29 closes it by definition.

## The count mismatch: 74 advertised, 114 present

The README and CI job title say **74**. The source contains **114** `#print axioms` directives, and our local run produced **114** “does not depend on any axioms” results.

The workflow step titled “Audit All 74 Theorems for Zero Axioms” checks Lean output for `sorryAx`, `Classical.choice`, `propext`, and `Quot.sound`, and counts successful reports. **It does not assert that the count equals 74.** Green CI therefore does not reconcile the advertised count with the artifact. This is a count mismatch, not a hidden-axiom finding.

## What we are and are not saying

We are saying the build, CI, static scan, and 114 zero-axiom reports held at the pinned commit. We are also saying the formalized statement is not a faithful negative resolution of Jaffe–Witten §4.

We are not saying the Lean proofs are broken, that the repository secretly uses `sorry`, or that the author’s physics intuition is disproved by this note. The question is narrower: what did the kernel actually check?

**A zero-axiom proof can be mechanically sound and still answer a different question from the one in the headline.** RoundSquare is a clear teaching case for that distinction—and its own `ArenaDefect` section is the best place to start reading it.

## Sources and pinned receipt

- Claim post: <https://x.com/dr_logvinovich/status/2118380634343105007>
- Repository: <https://github.com/Dr-Logvinovich/round-square>
- Audited HEAD: `70128f365357ba6ab661c0e8a822c05d1f5c2190`
- Official Clay statement, Jaffe–Witten §4: <https://www.claymath.org/wp-content/uploads/2022/06/yangmills.pdf>
- `RoundSquare.lean`: 153,907 bytes
- **`RoundSquare.lean` SHA-256:** `c2b76dd199f17fc50acffad25d4adeeee79afa31a3d4e5622c04082266366461`

---

*Receipts: Ford’s `STATUS.md`, `RECEIPT.txt`, and `FAITHFULNESS.md` for RoundSquare at commit `70128f365357ba6ab661c0e8a822c05d1f5c2190` (Nov 5, 2026). Public claim text from the repository README (tweet body was not fetched); it matches Scout’s brief.*
