← Về thư mục
📄 / / profiles / r-and-d / raw_evidence / OpenAI4S / workflows / environment-provenance / README.md

workflows/environment-provenance/

An artifact's environment provenance — An artifact must record the kernel generation that actually produced it, never the daemon's own interpreter. A snapshot taken without a generation to measure has to say it is assumed — provenance that is wrong is worse than provenance that is missing, because it is believed.

Steps: open_session, register_kernel_generation, capture_environment Permissions: kernel:python Declared artifacts: —

File Purpose
workflow.json The versioned manifest: steps, permissions, declared artifacts, failure conditions, and the cases below. Version 1.0.0. JSON rather than YAML for the same reason the core is, and versioned because a benchmark whose cases can change silently measures nothing across time.

Cases

Case Declared outcome What it pins
environment-provenance/measured provenance With a kernel generation, the snapshot is marked verified
environment-provenance/assumed provenance With no kernel generation, it must say so rather than assume

Failure conditions the manifest declares