AI Paper Atlas

Research map / Large language models / Agents, code and math

SL-VC: A Benchmark and Automated Framework for Separation Logic Verification Condition Proving

Hanyang Wang, Xiwei Wu, Qinxiang Cao

ICML 2026 · Large language models › Agents, code and math · approach: Retrieval-augmented formal theorem proving

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add LLM-friendly tactic library to single-stage proving (ablation)Human-Friendly Tactics · SL-VC (207-VC core subset)pass rate @ k=10: 45.9% -> 61.8%Improved
add LLM-friendly tactic library without repair (ablation)Human-Friendly Tactics · SL-VC (207-VC core subset)pass rate @ k=0: 3.9% -> 10.1%Improved
add two-stage spatial/pure workflow (ablation)LLM-Friendly Tactics · SL-VC (207-VC core subset)pass rate @ k=10: 61.8% -> 73.4%Improved
add two-stage spatial/pure workflow without repair (ablation)LLM-Friendly Tactics · SL-VC (207-VC core subset)pass rate @ k=0: 10.1% -> 17.4%Improved
replace human-friendly tactics baseline with full SPLITHuman-Friendly Tactics · SL-VC (288 VCs, full set)pass rate @ k=10: 43.4% -> 66.0%Improved
replace human-friendly tactics baseline with full SPLIT without repairHuman-Friendly Tactics · SL-VC (288 VCs, full set)pass rate @ k=0: 4.5% -> 14.2%Improved
use SPLIT instead of PALMPALM · SL-VC (207-VC core subset)pass rate @ k=10: 14.0% -> 73.4%Improved
use SPLIT instead of RangoRango · SL-VC (207-VC core subset)pass rate @ k=10: 7.7% -> 73.4%Improved
increase repair budget from k=0 to k=10 (ablation)SPLIT · SL-VC (207-VC core subset)pass rate: 17.4% -> 73.4%Improved
use Qwen3-235B-A22B instead of DeepSeek-V3.2 as LLM backendSPLIT · SL-VC (207-VC core subset)pass rate @ k=10: 73.4% -> 51.2%Hurt
use Kimi-K2-Thinking instead of DeepSeek-V3.2 as LLM backendSPLIT · SL-VC (207-VC core subset)pass rate @ k=10: 73.4% -> 51.7%Hurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy