CIRCA-RESS Computational Evidence Archive

Byte-preserved computational evidence for auditable partial identification of counterfactual risk reduction under runtime intervention. Version 2.0 retains the complete V6 PyBullet confirmation, failure-boundary, internal clean-room reconstruction, and public-case screening records, and adds the outcome-locked CIRCA-RESS-V9-SCI-S3 Gazebo diversity evidence: 7,424/7,424 rollouts, lossless trace arrays, frozen analysis, contracts, source lock, seed manifest, and terminal receipts. The pooled paired endpoint passed (estimate 0.1391; preregistered one-sided 95% lower endpoint 0.1032 against a strict 0.10 threshold), while the full route failed mandatory family and witness-validation gates. This mixed result documents target-family heterogeneity and does not establish family-robust effectiveness, hardware safety, deployment validity, or certification. The internal reproduction is not third-party replication, and no screened public case met the frozen identification criteria. Funding support was provided by 江西省教育厅科技项目 (DA201904140) and 博士启动金项目 (EA201404005); the funders had no role in study design, data analysis, manuscript writing, or the decision to submit.