Bring a finding.
Give it a future.
From a first result to a careful replication, make your contribution readable, inspectable, and ready for others to build upon.
The paper and its context.
Supply a title, abstract, manuscript, fields, and the author names and affiliations you want on the byline. Choose an explicit manuscript license and declare code and data licenses where applicable. Add references, keywords, and any funding or competing-interest statements you wish to provide.
No identity verification or AI-use declaration is required. ORCID identifiers, affiliations, and contributor roles are optional and self-reported. Your private project credential controls access; it does not determine the byline.
State the claim. Show the boundary.
The science manifest identifies the paper’s primary claims, manuscript anchors, assumptions, inputs, environments, and commands. Link each requested check to the exact claim it addresses. Mathematical, computational, empirical, and interpretive claims can coexist in one paper.
Any language can use the common execution contract, subject to available runtime capabilities. Do not include credentials, access tokens, or private secrets in a manuscript, manifest, or source upload.
Begin in a private workspace.
Save an immutable private version and upload its source archive. Request source inspection, then choose a supported execution or claim check. Each job has a quote and a maximum charge that you accept before it starts. Revisions create new versions and retain earlier evidence.
The pilot supports bounded Python and compiled-language execution, exact finite-statistic checks, restricted natural-number proofs, and scientific AI review. Check results show their scope and limitations. Public release still requires editorial acceptance and journal launch operations; a saved draft remains private.
Let the review reach the sources.
Link each citation to the claims it supports. Review considers relevance, necessity, actual support, and unnecessary citation inflation. The pilot accepts permitted PDF copies for every source needed in a full review, including works behind a paywall.
These copies and their extracted pages stay confidential. They are never bundled with the public article. The current review limit is 12 pages per reference; unavailable or incomplete full text leaves an explicit coverage gap instead of a claim that it was reviewed.
A useful result need not be a new headline.
Optionally register a protocol before the work, record declared prior access, and keep amendments and outcomes in its history. Stage-one acceptance concerns the method and reporting commitment; it does not establish a completed result.
Short replication, counterexample, and boundary reports identify an exact target and explain what was learned. Confirmation, contradiction, and inconclusive attempts have distinct meanings. Optional contributor roles describe work such as validation, software, and data curation.
Correct, respond, and carry the evidence forward.
Reviews and reasoned author responses remain attached to their versions. Authors can record appeals, submit corrections, and respond to contributions. Declared scientific dependencies receive scoped correction notices through the private inbox.
Public correction, concern, and retraction notices stay linked to their exact version. Readers see them alongside the article and in its citation exports. New versions preserve earlier citations, evidence, and notices.
A notice or challenge does not automatically refute a paper. The experimental aura indicator follows separately adjudicated claim standing, with reasons and appeal history. Citation totals, spending, and repeated checks do not earn automatic credit.