DataEvolver is the engineering companion repository for DataRSI. DataRSI builds on the synthetic-data construction infrastructure developed in DataEvolver and formalizes failure-driven dataset evolution as a bounded revision process.
- DataEvolver maintains reusable engineering infrastructure: asset generation, 3D reconstruction, scene rendering, VLM review and repair, multimodal exports, and runtime tooling.
- DataRSI hosts the research framework and experimental artifacts for DataRSI: A 3D Data Harness for Failure-to-Data Evolution, including registered protocols, failure-to-data requests, sample admission, and regression-aware model admission.
The paper connects two loops through a Failure-to-Data Feedback interface:
- Data Self-Evolution Loop: inspect candidate supervision, apply bounded repairs, and admit samples under a deterministic sample contract.
- Model Admission Loop: evaluate a candidate update against the frozen champion, check target gains and protected-region regression, and return
PROMOTEorROLLBACK.
The research protocol freezes the budget, evaluators, trainer, and admission rules for each bounded revision. General-purpose engineering defaults in DataEvolver are not a substitute for the paper's registered experimental settings.
| Goal | Entry point |
|---|---|
| Configure and run the engineering pipeline | DataEvolver Quick Start |
| Understand the DataRSI method and experimental scope | DataRSI README |
| Reproduce or inspect paper results | DataRSI reproduction guide |
| Check the relationship from the research repository | DataRSI's relationship document |
| Cite the DataRSI work | DataRSI citation metadata |
| Cite the earlier DataEvolver work | DataEvolver citation |
DataEvolver retains its repository name, dataevolver package namespace, CLI names, and existing deployment paths. The two repositories have distinct release histories and scopes. Existing DataEvolver publication links and recognition refer to the earlier DataEvolver work; the DataRSI research repository provides the corresponding paper materials.
DataEvolver 是 DataRSI 的配套工程仓库。 DataEvolver 提供长期维护的合成数据构建基础设施；DataRSI 在此基础上研究由模型失败驱动的有界监督数据修订过程。
本工程仓库主要用于资产生成、3D 重建、场景渲染、VLM 审查与修复、多模态数据导出和运行环境配置。DataRSI 论文仓库主要用于方法说明、固定协议、实验材料和结果复现。
DataRSI 通过失败到数据反馈接口连接数据自进化闭环与模型准入闭环：前者依据确定性样本合同接纳修复后的数据，后者依据目标收益和受保护区域的退化检查，决定晋升或回滚候选模型。论文复现应使用 DataRSI 注册的预算、评估器、训练配置和准入规则。
工程仓库继续使用 DataEvolver 名称及 dataevolver 包名，现有安装命令与路径保持兼容。引用 DataRSI 方法及实验时，请使用论文仓库的引用信息；引用既有 DataEvolver 工作时，请使用本仓库原有的引用信息。