rseng-provenance

# Software and data provenance Provenance answers the question every published number eventually faces: where did this come from, exactly? The complete answer names the input data (which version), the code (which commit), the parameters, the environment, who or what ran it, and when - for every derivation step from raw data to figure. This pack already records provenance in layers (seeds and configs in rseng-reproducibility, environment lockfiles, AI contributions in rseng-ai-declaration, file-level origin attributes in rseng-scientific-file-formats); this skill makes the capture systematic, standard and packaged. ## Capture at run time, not from memory Provenance reconstructed after the fact is testimony; provenance captured at run time is evidence. Instrument the pipeline so every run writes its own record: - Per run, record automatically: input paths WITH checksums or dataset versions (rseng-data-management), the code identity (commit hash, dirty-tree flag), the resolved configuration (the actual parameters, not the defaults file), the environment (lockfile hash or container digest), seeds, start/ end times, and the executing agent - human, scheduler or AI (rseng-ai-declaration's runtime counterpart). For the code side, lean on git's own metadata rather than restating it: authorship is the commit author/committer fields and trailers, and a SIGNED release tag is a cryptographic provenance anchor (rseng-version-control-review) - reference the tag, do ...