LongWoF-Bench: Evaluating EvoMap Genes for Verifiable Long-Workflow Tasks
2026-08-24 • Computation and Language
Computation and Language
AI summaryⓘ
The authors explore how large language models can get better at completing long, complex tasks by saving and reusing past successful attempts instead of starting from scratch each time. They created EvoMap, a way to store verified step-by-step solutions, and tested it on many tasks involving coding, reasoning, and following rules. Their results show that using these stored solutions helps models perform tasks more accurately and efficiently. This improvement depends on having verified experiences, not just compact summaries. Overall, their work suggests models can learn from past verified successes to handle complex workflows better.
large language modelsworkflow executionverificationEvoMapLong-Workflow Benchmarkcode generationmathematical reasoningagent-environment synthesisexperience reusemodel efficiency
Authors
Xiao Zhang, Qumeng Sun, Jihao Li, Yiming Ren, Xiang Liu, Haoyang Zhang, Junjie Wang
Abstract
Large language models are increasingly expected to execute complex workflows whose success depends on maintaining interdependent constraints and producing artifacts that satisfy strict end-to-end verification. Yet successful execution experience is typically lost after a single run, forcing subsequent models to rediscover strategies and failure modes from scratch. We study whether such experience can instead be externalized and reused through EvoMap, where verifier-confirmed execution trajectories are consolidated into structured Gene. To evaluate this setting, we introduce the Long-Workflow Benchmark (LongWoF-Bench), comprising 778 machine-verifiable tasks across code generation, agent-environment synthesis, mathematical reasoning, and rule following. On the 252 tasks with verifier-confirmed Opus trajectories, evolved EvoMap Gene outperform Skill across all seven evaluated models by 8.7-15.5 percentage points, with the gains extending to consumer models from different model families. In contrast, reference-distilled Gene do not exhibit the same advantage, indicating that compact representation alone is insufficient and that Gene utility is closely associated with verified experience provenance. For Claude Opus, Gene reuse also completes 39 more tasks than Skill while reducing solve-time token consumption by 9.9%. Together, these results show that verified execution experience can be retained and shared as a reusable external resource, enabling models to improve long-workflow completion without repeatedly paying the full cost of experience discovery.