A Public Standard for This Work

The research program

This is an independent, curiosity-driven program in experimental mathematics. I choose the questions, direct the work, decide what deserves attention, and take responsibility for what is published under my name. AI is a powerful collaborator in the process, but it is not an author, a referee, or a source of authority.

I expect the form of the work to vary. Some results may be conventional papers. Others may be shorter notes, exact constructions, counterexamples, computational classifications, certificates, or repositories that combine prose, code, data, and an audit trail. The admission standard concerns the clarity and support of the claim, not the length or prestige of the format.

What every result link will contain

Each research link from the main page will lead to a uniform result page. That page will serve as a compact guide to the complete artifact and will contain the following elements.

1. Claim A short, precise statement of the principal result.
2. Context What was known before, which earlier authors are most relevant, and the specific contribution claimed here.
3. Evidence The proof, computation, certificate, tests, or other checks supporting the claim.
4. Artifacts Links to the paper or note, living repository, code, data, and verification instructions.
5. Notes, when needed Material assumptions, caveats, unresolved points, or novelty qualifications that are not already clear from the claim, context, or evidence.
6. Citation and version A preferred citation, publication date, version, permanent archive or DOI when available, and correction history.

Admission checklist

A research result may be linked from the main collection only after every applicable item below is satisfied. If an item does not apply, the artifact should make that clear rather than silently omit it.

Claim and scope

• The principal claim can be stated precisely and without promotional language.
• The main result is not stated more strongly than its supporting argument or computation allows.
• Any condition that materially changes how the principal claim should be read is visible near it.
• The abstract, repository summary, and paper state materially the same result.

Prior work and credit

• The closest known prior work has been sought and discussed.
• Original sources are cited where they can reasonably be located and checked.
• The artifact distinguishes a new existence result from a new proof, method, formula, implementation, interpretation, or example.
• When novelty remains uncertain, the claim says “apparently new” or otherwise states the boundary of the search.

Evidence and reproducibility

• The proof or evidence actually supports the advertised claim at the advertised level.
• Critical computations use exact arithmetic or independently checkable certificates when practical.
• Code, inputs, dependencies, and commands needed to reproduce material computational claims are public when they can be shared.
• Recorded outputs are tied to the version of the code and data that produced them.
• Sampling, numerical tolerance, search bounds, and other limits are stated explicitly.

Adversarial checking

• Important claims have been challenged through counterexample searches, alternate derivations, dependency tracing, or fresh-context audits appropriate to the result.
• Material findings from those checks are resolved, accepted as risks, or displayed as open.
• Model agreement is not described as independent peer review.
• Any external human review that is mentioned is described accurately and only with permission.

Release and stewardship

• The public artifact contains no placeholders, private material, unverifiable references, or known stale claims.
• The paper, repository, and result page carry consistent titles, author information, dates, and principal claims.
• A stable release is tagged and citable; substantial releases receive a permanent archive or DOI when appropriate.
• The living repository identifies the archived release while remaining available for corrections and further work.
• A correction, withdrawal, or supersession policy is visible and prior public versions remain identifiable.

The role of AI

Unless an item says otherwise, readers should assume that AI built the work under my direction. That may include developing candidate mechanisms, drafting arguments, writing and testing code, searching for prior art, finding counterexamples, conducting audits, and producing the exposition. Repeating that provenance on every result page would add little; it is the default for this research program.

Much of the supporting text is also written for AI agents. The repositories are intended to be ingested, questioned, traced, and extended by systems that can move between claims, proofs, code, data, dependencies, rejected approaches, and audit records. This may produce artifacts that are more structured, extensive, or operational than a conventional paper.

Humans remain part of the intended audience. A motivated subject-matter expert should be able to identify the principal claim, understand the argument and evidence, locate the relevant artifacts, and evaluate the important qualifications without having to trust an AI summary. Agent-readability is an additional interface, not a substitute for human comprehensibility.

AI output receives no presumption of correctness. Different agents may share training, methods, and blind spots, so agreement among them is evidence about a process, not independent validation. The purpose of multi-agent or fresh-context checking is to expose errors and sharpen questions. It does not recreate expert peer review.

I remain responsible for choosing the questions, directing the work, setting the admission standard, deciding what to publish, and correcting or withdrawing it when necessary.

Credit and priority

Mathematics advances by building on other people’s work. A result is not improved by making its ancestry hard to see. Each artifact should identify the nearest relevant results, credit the authors whose ideas or constructions it uses, and state the difference between those results and the present contribution.

Literature searches can miss things, especially across older terminology or adjacent fields. If earlier work is later found, the public record will be corrected promptly. A narrowed novelty claim is a successful correction, not a reason to obscure the prior art.

Corrections are part of the record

Errors are possible, including serious ones. Corrections should be visible, dated, and linked to the affected claim. Material changes should create a new version rather than silently replacing the citable object. Withdrawn claims should remain findable with a clear explanation so that readers do not unknowingly rely on them.

The standard itself may also change. When experience shows that a check is weak, a category is misleading, or a better preservation practice exists, this page will be revised and its revision history recorded.