Everything you need to reproduce, verify, or build on our results
Every tab in the top bar, one line each — so this page can be about everything that is not a tab. Counts on this page are measured at build time from the files they describe, never typed.
The ChatGPT lane’s Track-2 v1.5 claim manifest set against the sealed suite, one row per claim, with the Registrar records that cite it and the suite label beside the lane label. Built because the Registrar has printed Track-2 claim ids since S305 and nothing on the site resolved them.
The lane’s manifest as shipped, JSON and CSV — claim id, wording, status, attachment, evidence state, caution tags, source and release location. Shipped unedited, and zero of its claims carry an author-acknowledged erratum: the one on the v1.3 wording of F-RESP-PREM-01 is discharged by the v1.5 reissue, and the Track 1 / Track 2 join says so beside the claim.
The sealed suite as sha-pinned, page-preserving text in four parts, so any quote on this site can be checked against the page it names without opening the PDF.
Every printed comparator resolves to a row here: value, uncertainty, source and edition.
Coalition packages: charter, math kernel, theory text, cold-start bundles. Copy, paste into any AI, reproduce any result.
The one canonical file every Registrar page is generated from. One record per line, schema-checked against the canonical contract before every cut.
The archive, and its own front door — it is deliberately not a tab. The old Grok library (v1, frozen), the March Proof Library index that used to stand at the Library tab, retired claims with the decision that retired each one, and every frozen revision of the suite.
The Live tab folded into the Home page at S313: the channel plays on the theatre screen there. This page is kept as the record of the tab that stood here.
The March 2026 archive as data. Historical: entries carry their original March status, not a current disposition. The sealed suite wins wherever they differ.
What this site is, for a machine reader — generated from the corpus, with the indexing and review state measured from the built bytes rather than asserted.
What this material is and is not: a pre-print programme, not peer-reviewed, open for scrutiny.
Step-by-step instructions for reproducing any result. Not written yet — said plainly rather than as “coming soon”. The kernel and the cut-and-paste bundles above are what exists today.