Search papers, labs, and topics across Lattice.
ReproAgent is a novel four-stage pipeline designed to convert research papers into executable code repositories while preserving the original methodologies and artifacts. The approach addresses the challenge of lost explicit and implicit details during the reproduction process by utilizing a persistent implementation contract with two distinct channels: one for implementation requirements and another for reference evidence. On the PaperBench Code-Dev benchmark, ReproAgent outperformed existing scaffolds, achieving the highest mean score with both Claude-Sonnet-4.5 and Gemini-3-Flash models, demonstrating its effectiveness in enhancing paper-to-code reproduction.
ReproAgent achieves unprecedented accuracy in translating research papers into executable code, outperforming existing methods by leveraging a dual-channel contract system.
Paper-to-code reproduction asks scientific AI agents to turn research papers into executable repositories that preserve the paper's method, protocol and artifacts. This is difficult because the specification is split: explicit paper content such as algorithms, metrics and artifacts is often lost across long agent trajectories, while implicit details such as framework defaults and conventions inherited from related work are absent from the paper. We introduce ReproAgent, a four-stage Prepare--Plan--Generate--Repair pipeline built around a persistent implementation contract with two channels: an implementation-requirement channel that turns paper snippets into code obligations, and a reference-evidence channel that retrieves content and structure evidence from related repositories. Both are bound to work packages, projected into file-level contracts, and consumed across generation and repair. On PaperBench Code-Dev, ReproAgent reaches the highest mean score among same-backbone scaffolds under both Claude-Sonnet-4.5 and Gemini-3-Flash. End-to-end channel ablations and per-paper cases support the contribution of both channels. Code and experimental artifacts are publicly available.
