Statement identity
Freeze the exact theorem, notation, domain, quantifiers, dependencies and version. Separate the claimed endpoint from commentary and consequences.
Mathematical proof verification
Test the theorem, not just the prose around it.
A Haivian proof audit fixes one exact statement and examines whether the definitions, assumptions and reasoning support it. Formal execution can strengthen the record when a suitable artifact exists; it does not replace mathematical judgment.
What the audit can examine
A claim-bound mathematical review.
The accepted plan selects only the checks relevant to the theorem and available evidence.
Reasoning and counterexamples
Inspect key proof transitions, hidden assumptions, boundary cases and plausible counterexamples. Record any step that remains unsupported or ambiguous.
Formal artifacts
When Lean or another agreed formal artifact is supplied, inspect the declared target, build path, axiom dependencies and kernel acceptance within the available toolchain.
Evidence returned
A result that states exactly what passed.
A positive result attaches to the reviewed claim and version. It does not certify every nearby theorem, implementation or novelty claim.
Frozen scope
The report identifies the exact manuscript, theorem, formal targets, supplied files, exclusions and trust assumptions.
Reproducible checks
Where permitted, it records commands, module counts, inspected closures, hashes and outputs needed to understand the execution.
Bounded finding
Findings may be supported, contradicted, inconclusive or revision-required. Material dissent and failed checks stay in the record.
Human accountability
The final report identifies the responsible human decision and the role of AI-assisted analysis. It does not present same-provider review as external peer review.
What this service does not promise
Formal checking has a defined boundary.
A proof audit does not guarantee novelty, publication, downstream software correctness or universal truth outside the accepted assumptions. A compiled formal target remains subject to the identified theorem statement, dependency set, axioms and toolchain.
If no suitable formal artifact exists, the audit can still test the mathematical argument and specify what further formalization would be needed. Any additional formalization work is separately scoped.
Inputs and fit
What makes a proof audit reviewable.
The strongest submission gives the reviewer one stable object to inspect rather than a moving collection of claims.
Supply the exact version
Provide the manuscript or preprint, the primary theorem statement, definitions, references and any formal repository or archive at a named commit. Identify which consequences matter to the intended decision.
State the review question
Say whether you need a direct mathematical challenge, a formal-build check, an axiom review, a counterexample search or a combination. Haivian will not silently treat one check as all of them.
Declare relationships
Disclose authorship, commissioning, material AI use, funding and known conflicts. Public wording distinguishes process separation from external human or institutional independence.
Existing public examples. The Validation Journal records bounded audits of a Boolean-function degree theorem, Seymour's second-neighborhood conjecture, a Borsuk counterexample and a finite-host Sidorenko counterexample. Those records show methods and limits for their exact targets; they are not templates that guarantee the same outcome for another submission.