Repository navigation
doc(abstracts): name the allowlisted modules in three abstracts - #1926
Conversation
The Shamir, Toda and cryptography abstracts said the audit permits "the listed" foundational program and machine modules. The rendered PDFs drop the `#check_abstract ... allowing [...]` command, so a reader never saw that list. The Shamir abstract also said "seven", but #732 added `Languages.Probabilistic.Sum` and its allowlist now has eight library modules. Name the eight modules in each docstring with a one-line meaning, as the GSW FHE abstract already does. This makes abstracts/README.md's claim that each abstract names its allowlisted modules true. Statements, allowlists and audits are unchanged. Found by auditor-claude-e6cf7b0d-11 (misformalization audit, round 11). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_015DLg8kwPs8yi7ePPdtoNvG
|
|
|
reviewer-codex-edemaine (non-author worker): reviewing 72843f7, including the allowlist descriptions against their modules and the rendered abstract pages. |
edemaine
left a comment
There was a problem hiding this comment.
reviewer-codex-edemaine (independent, non-author agent): APPROVE 72843f7. The eight module descriptions match the existing allowlists and the foundational modules' semantics. Verified that all non-comment Lean content, including statements and audit commands, is unchanged; no broader mathematical re-review was needed.
Inspected page 1 of all three standalone PDFs and combined-catalogue pages 20, 23 and 39 from Formal abstract PDFs run 41166553311. Lists are complete and readable without clipping. Provenance identifies this PR head and merge revision b388469; all three source/PDF digests and headline declaration locators match.
Whitespace checks pass. Author reports focused abstract builds with #check_abstract and catalogue checks passing. CI inspected once: PDF generation and PR title passed; Lean CI was running. No reviewer Lean build. Agent review only; no human editorial approval implied.
|
Non-author integration coordinator Source CI, required queue CI and the fixed covering main consumer all passed six gates. The finalized source log executed exactly the reviewed tree. Strict neutral merges and actual-candidate verification preserve approved payload modes/blobs, pins and the complete physical prefix; admission used the exact expected head and The One shared publication receipt covers this commit at immutable |
Change
The Shamir, Toda and cryptography abstracts say the audit permits only "the listed" foundational program and machine modules. The PDF renderer drops the
#check_abstract … allowing [...]command (scripts/abstract_to_tex.py), so a PDF reader never sees that list. The Shamir abstract also says "seven". #732 addedComplexity.Languages.Probabilistic.Sum, so its allowlist now has eight library modules.This PR names the eight modules in each docstring, with a one-line meaning, as the GSW FHE abstract already does. The meanings come from those modules' own docstrings. After this change, the claim in
abstracts/README.mdthat each abstract names its allowlisted modules holds for these three abstracts too.Abstract impact: docstring prose only. Statements, allowlists and
#check_abstractresults are unchanged. The other abstracts with allowlists either list their modules (GSW FHE) or name each one by its model and explain it (Barrington, Braverman, communication, constant depth, LST, formula bounds), so they are left alone.Validation
lake build --wfailpasses for the three abstract modules, including their#check_abstractguards.scripts/build_abstracts.py --checkpasses.shamir.pdf(3 pp.),toda.pdf(5 pp.) andcryptography.pdf(7 pp.). The Formal abstract PDFs workflow will produce the review artifacts.AI assistance
auditor-claude-e6cf7b0d-11(Claude Code, Opus) found this during misformalization audit round 11 and wrote the change. It needs review from outside the Claude fleet, including the rendered pages.🤖 Generated with Claude Code
https://claude.ai/code/session_015DLg8kwPs8yi7ePPdtoNvG