Skip to the content.

0001: The paper’s statement in coverage.yaml

Status: accepted. Author: drafted by the AI operations agent at Le Chen’s request. Created: 2026-10-17. Decided: accepted 2026-10-17 by Le Chen (see Decision below). Implemented in: (release).

Summary

Add one optional key to coverage/1: rows[].paper.statement on a paper row and rows[].statement on a library row, the result in mathematical words (TeX). lean-pkg page shows it beside the Lean statement, so a reader can compare the two by eye without opening the paper.

Motivation

The verification page now opens with “Lean has checked every step of the proofs listed here. It cannot check that a Lean statement says what the paper says: that comparison is yours”, and shows each result’s Lean statement copied from the source (design, pull request #205). The other half of that comparison, the paper’s statement, is not in any file lean-pkg reads. Today the page can only say “Theorem 1 of the paper: read its statement there”. Palomar’s entry pages lead with the informal account before the formal statement; a lean-pkg page should do the same for each result.

Design

rows:
  - paper:
      number: "Theorem 1.2"
      title: "Main estimate"
      statement: |
        For every $n > 0$,
        \[ \sum_{k \le n} a_k \le C\, n^{1/2} . \]
    class: same-assumptions
    claims: [my-companion/main-estimate]

Compatibility and migration

Trust model

No receipt, gate or trust level changes. The statement is the authors’ transcription: nothing checks it against the manuscript, and the page does not say it does. The risk is a transcription that agrees with a weakened Lean statement but not with the paper. The page already tells the reader that fidelity is theirs to check, and the paper reference stays beside it. A later proposal could compare the text with the labelled environment of the anchored e-print (formalization anchor records its SHA-256; paper.label names the environment).

Alternatives

Open questions

Decision

Accepted on 2026-10-17 by Le Chen, the maintainer of record, as designed above. Typeset mathematics stays out for now: the page shows the statement as TeX source only, so no script and no markup renderer touches the authors’ text. The open questions were not decided: the first implementation uses the 2000-character limit and does not warn about a row without a statement.

AI assistance

Drafted by Claude (Anthropic) from the page redesign; the implementation and its tests exist on a branch and wait for this proposal.