Archived edition

Updates on artificial intelligence for mathematical research, proof discovery and formal verification, with links to papers, code and tools.

Research updatesSelect an entry to read more

2 entries · 5 linked sources Partial coverage
Harmonic adds Rust verification to Aristotle, with a Manifest order-book case study Harmonic Published: 3 sources

Harmonic announced Rust support for Aristotle, extending its software and hardware verification offering. Its accompanying Manifest case study says Aristotle produced 28 Lean-stated results across eight areas of an open-source Solana order book, including swap progress, error classification, tree integrity, custody, matching, and quote rounding. Harmonic reports that the analysis found two bugs, both subsequently fixed: rounding could make a completed fill fail after reaching its target, and an unnecessary quote conversion could block cancellation of a global order. The published case study states that the work analyzed the actual Rust code, including unsafe code, and covered arbitrary finite sequences of operations under stated conditions.

Research relevance:

A practical example of machine-checked reasoning applied to real systems code, with useful detail on property scope and counterexample-driven repair. It may interest researchers exploring proof-producing agents, verification of unsafe code, or workflows that translate informal requirements into Lean statements.

Claim status:
Harmonic’s dated announcement and case study make these company-reported results, not an independent evaluation. Harmonic says the results have associated Lean statements; the available documentation describes machine-checked proofs, but this digest did not inspect or rebuild the project artifacts and does not claim to have verified proofs or run Lean. The release concerns software verification rather than a new theorem in pure mathematics.
Limitations:
The case study says token and system programs called by Manifest are assumed rather than proved, and gives no compute-unit bound. The Rust documentation also lists assumptions about compiler translation, platform behavior, and unsupported reasoning about termination, concurrency, memory leaks, liveness, and complexity; it supports only a specific Rust compiler/build configuration. Results are conditional on stated account, funding, arithmetic, and liquidity conditions.

Sources & further reading

Scott Armstrong reports a Carbon-generated proof for three probability and statistical-physics conjectures Scott Armstrong (claim in a public X post) Published: 2 sources

Armstrong wrote that a first prompt to Gemini’s Carbon solved a research problem that prior attempts with other AI systems had not. In replies, he specified that the project concerned three conjectures, A, B, and C, in probability/statistical physics; Carbon worked for 103 minutes and produced a 43-page PDF proof. He said two Astra agents and two Opus agents had subsequently checked it and that Lean formalization was underway. This is a notable report of an AI-assisted research workflow, but the mathematics and proof are not yet publicly inspectable in the sources consulted.

Research relevance:

If independently substantiated, the episode could be relevant to how researchers combine long-context problem setup, cross-model critique, and formalization in active mathematical work. The reported process is a useful lead, but not yet evidence that the conjectures are resolved.

Claim status:
An attributed researcher report, not an independently established theorem. The readable mirror shows the post and replies, including Armstrong’s elaboration; no public manuscript, precise statements of the conjectures, human referee assessment, or completed formalization was found. Other AI agents’ reported checks are not independent mathematical certification. This digest has not verified the proof or any Lean artifact.
Limitations:
The supplied X collection included the original short post but not its replies; a third-party mirror was used to read additional thread context and is not an official source. No exact proof, source notes, or public Lean repository were located during this search. Carbon’s product identity and the poster’s identity were not independently verified through official sources in this research pass.

Sources & further reading

Coverage and limitations
  • Window checked: 2026-11-15 18:15 to 2026-11-17 06:15 Europe/Paris. Search and direct page reads covered Harmonic’s dated announcement, its Manifest case study, and Aristotle’s Rust documentation. Harmonic’s official pages describe company claims; they are not independent audits.
  • The supplied X API feed for @scottnarmstrong contained two posts; the research claim was followed through a publicly readable third-party mirror that displayed the post and replies. This allowed reading the post text, but does not authenticate the account independently. Replies in that mirror add context but were not part of the supplied collector results.
  • The Carbon proof claim concerns probability/statistical physics and was described in the thread as a 43-page LaTeX/PDF proof of three conjectures, later checked by other AI agents and undergoing Lean formalization. I found no public manuscript, conjecture statements, or independent mathematical assessment in the sources consulted; those checks do not certify correctness.
  • The other supplied X accounts had bounded API results, with empty feeds for many accounts and replies/reposts excluded. API resolution is not identity verification; cached posts may be included. Bluesky @xenaproject.bsky.social returned no posts within a maximum of 300 checked; Mathstodon collection was unavailable due to HTTPError, with replies/reposts excluded and a 300-post cap. Empty feeds do not establish that no relevant posts occurred.
  • Axiom’s supplied feed linked to a CFSG applications blog and repository, but the blog URL could not be opened through the web tool and the repository landing page did not expose a dated release in the material retrieved. This possible update is therefore not treated as a digest item. X access is not exhaustive, and there were no direct API posts supplied for several named X accounts.

Requested coverage window: 2026-11-15T18:15:01.286842+01:00 — 2026-11-17T06:15:01.598483+01:00.

Public web research; coverage is not exhaustive. A source link is not a certification of a claim.

Archives 79
Older editions (49)