claim first, status on separate axes
The unit is the claim, not the paper. A source can establish what an author reported without establishing that Bathymark reproduced it, that it survives costs, or that it still holds now.
The first release is intentionally manual and version controlled. It stores curated bibliographic facts, Bathymark paraphrases, source locators, structured reported numbers, limits, and status history. It stores no source abstracts or full text, makes no public provider request, and uses no model to publish claim records.
| Axis | States used now | What the axis answers |
|---|---|---|
| Literature | Source reviewed, Conditional, Mixed evidence, Disputed, Method guardrail, Working paper | What the reviewed source and linked research support, and how publication state changes the weight of that record. |
| Bathymark reproduction | Not started, Test outline ready, Reproduced, Failed replication, Method record | Whether Bathymark has only specified the test, rerun the same computation, or recorded a failed reproduction. |
| Live validity | Not monitored, Collecting evidence, Currently supported, Currently conditional, Decayed, Withheld | Whether a reproduced claim is collecting prospective evidence, currently supported, conditional, decayed, or withheld. |
Obtain a consistent result with the same data, code, and computational steps. A reproduced computation can still contain a conceptual or data error.
Test the same scientific question with newly obtained data or an independently rebuilt artifact. This is stronger than rerunning author materials.
Test whether the relationship extends to another venue, asset, period, chain, implementation, or market regime.
Crossref bibliographic facts and registered update relations can enter the bounded owner-run metadata audit through the source-use gate. Crossref abstracts, publisher full text, OpenAlex content files, excerpts, claim extraction, and model processing are separate rights questions. None are enabled by the first release.
FAIR guides machine-actionable stewardship. PROV-O supplies interoperable provenance. RO-Crate 1.3 is the target package for a future completed replication. SWHID can identify exact source code content. None replaces a license, complete runtime, preserved data, or evidence review.
Research objects should be findable, accessible, interoperable, and reusable for people and machines.
Portable provenance relationships for entities, activities, agents, derivation, and revision.
JSON-LD packaging for research data, code, workflows, people, licenses, and provenance.
Content-derived identifiers for exact source files, trees, commits, and other software artifacts.
Bibliographic identity, DOI versions, update relations, and Retraction Watch links, excluding abstract and full-text ingestion.
Method reviewed 2026-07-13. This is an evidence and research workflow, not a certification or financial advice.