| name | dataset-provenance-manifest-governance |
| description | Keep local training dataset provenance, merged outputs, and manifest truth aligned across dataset-building repos. Use when teacher distillation, enrichment exports, or study manifests change together. |
Dataset Provenance Manifest Governance
Use this skill when the Penny training-data chain changes across Qwen Training and Media Workbench surfaces.
Workflow
- Identify the source manifests, merged outputs, and derived dataset artifacts involved.
- Confirm provenance rules before changing batch assembly or export paths.
- Keep teacher, student, and enrichment stages legible in the manifest chain.
- Re-run the relevant validation or export proof after updates.
- Record whether the dataset remains local-only and reproducible.
Required Evidence
- source manifest or config touched
- derived dataset artifact or export touched
- provenance rule or weighting rule preserved
- validation or export proof for the changed path
Guardrails
- Do not collapse teacher, enrichment, and student stages into one opaque artifact.
- Keep local-only provenance explicit.
- Do not silently rewrite upstream source media or manifests.
- Fail closed if the manifest chain no longer explains the output.
Best-Fit Companion Skills
$qwen-training-dataset-factory$qwen-training-campaign-ops$media-workbench-indexing-governance$docs-alignment-lock
Scope Boundary
Use this skill only for dataset provenance, manifest lineage, and derived training-artifact governance.
For checkpoint evaluation or fine-tune execution, route through the more specific training skills.
