AI Science Tracker
EN

A coding agent made a verified Lean lookup 5.46 times as fast

University of California, Berkeley; Amazon Web Services · 20 Oct 2026 · computer science

Discovery

A coding agent made a verified Lean lookup 5.46 times as fast

Source check: Source checked. Not independent scientific confirmation.

Finding: Codex with GPT-5.6 Sol rewrote the sheet lookup in a Lean port of the Cheat command-line tool. The rewrite searched directly instead of building an intermediate list, and ran 5.46 times as fast as the original Lean function on withheld timing inputs. Lean checked that the two functions return the same result.

Technology: Codex with GPT-5.6 Sol, Lean 4

Family GPT-5: profile · models page

Publication: Preprint · Test context: direct test. Catalog categories, not an evidence grade.

AI contribution
Codex with GPT-5.6 Sol generated a direct reverse-search implementation and a Lean equality proof.
Test and result
The study reports repeated paired performance trials and an equality theorem checked by the Lean kernel for the specific rewrite.
Limitation / still unknown
The speedup concerns the named function on the study’s workloads. The equality theorem covers the migrated Lean reference, while agreement with the original Go program was assessed by differential testing.
Evidence sources and annotation context

Evidence context (descriptions follow the cited sources; test type is a catalog classification):

Evidence sources: https://arxiv.org/abs/2610.30188 · https://arxiv.org/html/2610.30188v1

Source checked · editorial details

The source page was checked against this entry. Independent confirmation is not assessed by this label.

Source type: paper. Paper: linked separately.

Review date: 21 Nov 2026. Review scope: arXiv v1 §5.2 and Appendix D.3, named findConsolidated rewrite and proof scope.

No follow-up recorded; this is not a claim that no follow-up exists. Follow-up review scope and date are not recorded. No source-supported relationships recorded.

Editorial assessment: Impact 2/5 · Validation 3/5 · Novelty 2/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.

Open in the full catalog for filters, charts and related discoveries. Browse computer science discoveries.

Related catalog entries

Links use documented relationships where available, plus shared catalog context. A shared topic or model family does not establish a scientific dependency.