CONTRACT
Verification is a ladder, and the rung is printed on the claim.
Every review on Pith carries a mark, T0 through T4, stating exactly how much of the paper was mechanically checked: the depth of the method, never the truth of the paper and never the standing of its authors. Most of the record sits at T0 and says so; that honesty is what makes the higher rungs mean something.
-
T4
Kernel-checked proof reserved · 0 artifacts as of 2026-07-09
The claim is proved in Lean 4 and the kernel accepts the proof, with the axiom closure printed so you can see exactly which foundations the result rests on. Nothing about the authors, their institution, or their history enters into it.
how to earn it Submit the paper with its Lean sources. The review reproduces the build from a pinned toolchain, checks zero-sorry status, and audits the axiom set (ticket grades V3/V5; V4 adds T2). If you have a Lean-proofed result and the pipeline fights you, email review@pith.science.
Held unfilled until the first external kernel-checked artifact earns it. Gold marks this rung only and appears nowhere else on Pith.
-
T3
Formal statement
The paper's central claims are stated in Lean 4 and compile against a pinned toolchain. A formal statement is precise and citable before any proof exists: it pins down exactly what is being claimed.
how to earn it Link Lean theorem statements for your central claims with your submission (ticket grade V2).
-
T2
Reproducible compute
The paper's own computational artifacts, code, data, notebooks, are re-executed and the stated outputs reproduce. The honest ceiling for most empirical science.
how to earn it Supply runnable artifacts with your submission (ticket grade V4 on the review).
-
T1
Deterministic integrity audit
Reference resolution, citation-quote validity, and artifact checks run reproducibly on the raw files, with no model judgment in the loop. Anyone can re-run them and get the same answer.
how it happens The integrity layer runs continuously; see /integrity and the protocol. A free standalone audit for any bibliography: /tools/reference-audit.
-
T0
Structured rubric review default · most of the record
A machine referee reads the full paper and writes a structured report against a public rubric: core claim, load-bearing premise, what would settle it, and a referee-style assessment.
how it happens Automatic for every paper on the record. The rubric and the passes are public at /methodology.