Ganguly, Debargha, Raghunathan, Divya, Jovanović, Dejan, Singh, Vikash, Kong, Soonho, Tomb, Aaron, Goel, Aman, Erata, Ferhat, Bayless, Sam · Oct 28, 2026
Scaling Lean Theorem Proving for Real-World Code Verification
Markdown · JSON · Original source
The paper’s authors present EasyProver, a proving system built from recursive agents, and Aegis, a designed to test verification on real implementations. They argue that familiar benchmarks are close to saturation, so Aegis turns competitive programming problems into harder verification tasks involving fixed-width arithmetic, explicit failure, generated loops, casts, and functionalized mutation. The authors report EasyProver’s results across existing benchmarks. On a stratified sample of 2,000 Aegis statements, it proves 1,054 within a 30-minute budget per statement. The pipeline creates Rust solutions, translates them into Lean, generates specifications, and labels statements with kernel-checked proofs or refutations. The paper presents both a proving system and a source of more demanding verification tasks. The authors say those tasks could also benchmark other provers or provide rewards for . The abstract reports results on the specified benchmarks and sample; it does not establish performance on implementations beyond those tasks.
Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims.
Original source
cs.AI · Article
Source details
Original source date: Oct 28, 2026 (2026-10-28).
Date basis: Date basis not provided.
Added to Escape Meridian: 2026-11-11T03:54:04.048Z. First observed: 2026-10-30T06:42:32.442Z. Retrieved: 2026-10-31T04:02:12.272Z.
Referenced by
Recorded citations and public file attachments. A reference alone does not establish agreement or influence.
No public references recorded here yet.
Limits
The reported Aegis result is for a stratified sample of 2,000 statements, with a 30-minute budget per statement; the abstract does not report performance beyond the specified and sample.
Open question
How will EasyProver perform on implementations and specifications beyond Aegis’s generated Rust- pairs and its reported sample?
Source-stated
The authors introduce EasyProver, a Lean proving system built from recursive agents and other search and elaboration components.
We present EasyProver, a Lean proving system built from recursive agents with isolated contexts, asynchronous search, parallel tactic probing, and an elastic elaboration service.
Source-stated
The authors introduce Aegis because they consider existing benchmarks close to saturation; it creates harder verification tasks from competitive programming problems and real implementations.
Because these benchmarks are close to saturation, we also introduce Aegis, which converts competitive programming problems into much harder verification tasks over real implementations.
Source-stated
Aegis produces Rust-Lean program pairs and candidate specifications, with each frozen statement labeled with a kernel-checked proof or refutation.
It produces 2,045 differentially validated Rust-Lean program pairs and 11,533 candidate specifications. Unlike hand-written benchmarks, these tasks carry implementation-level obligations
Source-stated
On a stratified sample of 2,000 Aegis statements, EasyProver proved 1,054 within a 30-minute budget per statement.
On a stratified sample of 2,000 statements, EasyProver proves 1,054 (52.7%) within a 30-minute budget per statement.
Source and contributor context
Ganguly, Debargha, Raghunathan, Divya, Jovanović, Dejan, Singh, Vikash, Kong, Soonho, Tomb, Aaron, Goel, Aman, Erata, Ferhat, Bayless, Sam · Article author
These are credits stated by the source; they do not establish a profile match or authorship of every passage.
Source media
No suitable media was found by the supported page extraction.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
Benchmark
A defined set of tests used to compare systems. Results depend on the tasks and measures selected.
Reinforcement learning
Training by trial and error, in which a system is rewarded for the outcomes its designers want.
Lean
Lean is a programming language and proof assistant used to express mathematical statements and check formal proofs.
arXiv
arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Editorial provenance
Version 1. Publication ID: 687e053b-b7b5-4f42-8d69-daba20689bc5.
Prepared by gpt-6-luna; grounding review by gpt-6-luna.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-11T03:54:04.048Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: material_development.
Observation: 2a4f190b3a2c7fbc08d1415ec15cc2ee06bb0be25ec0161d9e7bd532ed520bf2. Source text hash: 5e6e85009f56fe7e36004813fc2aa5b4672c3ca49b49b91da34cdb39ee40564f.
Revision and source-check history
Version 1 · 2026-11-11T03:54:04.048Z
Scaling Lean Theorem Proving for Real-World Code Verification
The paper’s authors present EasyProver, a proving system built from recursive agents, and Aegis, a designed to test verification on real implementations. They argue that familiar benchmarks are close to saturation, so Aegis turns competitive programming problems into harder verification tasks involving fixed-width arithmetic, explicit failure, generated loops, casts, and functionalized mutation. The authors report EasyProver’s results across existing benchmarks. On a stratified sample of 2,000 Aegis statements, it proves 1,054 within a 30-minute budget per statement. The pipeline creates Rust solutions, translates them into Lean, generates specifications, and labels statements with kernel-checked proofs or refutations. The paper presents both a proving system and a source of more demanding verification tasks. The authors say those tasks could also benchmark other provers or provide rewards for . The abstract reports results on the specified benchmarks and sample; it does not establish performance on implementations beyond those tasks.
Original source
cs.AI · Article
Source details
Original source date: Oct 28, 2026 (2026-10-28).
Date basis: Date basis not provided.
Added to Escape Meridian: 2026-11-11T03:54:04.048Z. First observed: 2026-10-30T06:42:32.442Z. Retrieved: 2026-10-31T04:02:12.272Z.
Limits
The reported Aegis result is for a stratified sample of 2,000 statements, with a 30-minute budget per statement; the abstract does not report performance beyond the specified and sample.
Open question
How will EasyProver perform on implementations and specifications beyond Aegis’s generated Rust- pairs and its reported sample?
Source-stated
The authors introduce EasyProver, a Lean proving system built from recursive agents and other search and elaboration components.
We present EasyProver, a Lean proving system built from recursive agents with isolated contexts, asynchronous search, parallel tactic probing, and an elastic elaboration service.
Source-stated
The authors introduce Aegis because they consider existing benchmarks close to saturation; it creates harder verification tasks from competitive programming problems and real implementations.
Because these benchmarks are close to saturation, we also introduce Aegis, which converts competitive programming problems into much harder verification tasks over real implementations.
Source-stated
Aegis produces Rust-Lean program pairs and candidate specifications, with each frozen statement labeled with a kernel-checked proof or refutation.
It produces 2,045 differentially validated Rust-Lean program pairs and 11,533 candidate specifications. Unlike hand-written benchmarks, these tasks carry implementation-level obligations
Source-stated
On a stratified sample of 2,000 Aegis statements, EasyProver proved 1,054 within a 30-minute budget per statement.
On a stratified sample of 2,000 statements, EasyProver proves 1,054 (52.7%) within a 30-minute budget per statement.
Source and contributor context
Ganguly, Debargha, Raghunathan, Divya, Jovanović, Dejan, Singh, Vikash, Kong, Soonho, Tomb, Aaron, Goel, Aman, Erata, Ferhat, Bayless, Sam · Article author
These are credits stated by the source; they do not establish a profile match or authorship of every passage.
Source media
No suitable media was found by the supported page extraction.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
Benchmark
A defined set of tests used to compare systems. Results depend on the tasks and measures selected.
Reinforcement learning
Training by trial and error, in which a system is rewarded for the outcomes its designers want.
Lean
Lean is a programming language and proof assistant used to express mathematical statements and check formal proofs.
arXiv
arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Editorial provenance
Version 1. Publication ID: 687e053b-b7b5-4f42-8d69-daba20689bc5.
Prepared by gpt-6-luna; grounding review by gpt-6-luna.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-11T03:54:04.048Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: material_development.
Observation: 2a4f190b3a2c7fbc08d1415ec15cc2ee06bb0be25ec0161d9e7bd532ed520bf2. Source text hash: 5e6e85009f56fe7e36004813fc2aa5b4672c3ca49b49b91da34cdb39ee40564f.
uncertain · 2026-11-11T03:54:04.048Z · anthropic/claude-opus-5.5
Publication: 687e053b-b7b5-4f42-8d69-daba20689bc5
attribution — pass: The reading attributes EasyProver, Aegis, the saturation argument and all results to the paper's authors. The authors are listed on this arXiv page, and the reading presents their results as reported claims rather than established facts.
Authors:Debargha Ganguly, Divya Raghunathan, Dejan Jovanović, Vikash Singh, Soonho Kong, Aaron Tomb, Aman Goel, Ferhat Erata, Sam Bayless
summary — uncertain: The reading is faithful but hard for a non-specialist to follow. Central terms like Lean, theorem proving, code verification, kernel-checked proof, specification, fixed-width arithmetic and functionalized mutation are never explained or tooltipped. Smallest repair: add plain-language glosses or term-library tooltips. The reading also omits a key element of the authors' framing: EasyProver uses an off-the-shelf frontier model with no additional training. It says only that results were reported on existing benchmarks, without the figures. Restoring both details would preserve why the authors call those benchmarks near saturation.
Using an off-the-shelf frontier model with no additional training, it proves 669/672 PutnamBench problems and all six IMO 2026 formalizations
date — pass: The originalPublished date of 2026-10-28 matches the v1 submission date shown on the arXiv page. The feed timestamp of 2026-10-31 is later and is correctly not used.
[Submitted on 28 Oct 2026]
support — pass: All claims are stated in the abstract. The 1,054-of-2,000 result is correctly scoped to the sample and the 30-minute budget. The limitation and the question stay within the evidence. A minor issue: claim 3's excerpt does not contain the kernel-checked labeling language, but that point appears elsewhere in the abstract. The reinforcement-learning use is correctly framed as something the tasks could be used for.
labels each frozen statement with a kernel-checked proof or refutation
media — pass: The selected media is the verified direct link to the arXiv abstract page, which is the original source. No better source media was found.
Challenge or correct this reading
Send a correction or an opposing source with this reading’s URL. Disagreement and uncertainty can remain unresolved.
Lean is a programming language and proof assistant used to express mathematical statements and check formal proofs.
Explore LeanA defined set of tests used to compare systems. Results depend on the tasks and measures selected.
Definition referenceTraining by trial and error, in which a system is rewarded for the outcomes its designers want.
Definition referencearXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Explore arXivA defined set of tests used to compare systems. Results depend on the tasks and measures selected.
Definition referenceLean is a programming language and proof assistant used to express mathematical statements and check formal proofs.
Explore LeanLean is a programming language and proof assistant used to express mathematical statements and check formal proofs.
Explore LeanA defined set of tests used to compare systems. Results depend on the tasks and measures selected.
Definition referenceTraining by trial and error, in which a system is rewarded for the outcomes its designers want.
Definition referencearXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Explore arXivA defined set of tests used to compare systems. Results depend on the tasks and measures selected.
Definition referenceLean is a programming language and proof assistant used to express mathematical statements and check formal proofs.
Explore Lean