Learning outcomes
- Audit denominator and missing-data rules
- Check association, alignment, and path-length policies
- Connect Docker reproducibility to public table eligibility
A row is a claim
A numeric row claims that a specific system, configuration, dataset sequence, trajectory, timing file, EPA output, and resource context belong together.
If any part is missing, the public value should stay TBD, N/A, Partial, or Unresolved.
Aggregation is conditional
Averages require a visible denominator and a rule for which rows are included.
Do not mix rows with different output scopes, dataset conversions, timing policies, or resource platforms.
Checklist
- Denominator is visible
- Missing rows are not silently averaged
- EPA policy is attached
- GitHub/Docker reproduction path exists
Advanced notes
OpenVINS compatibility outputs are useful for migration and sanity checks. EPA remains the active public evaluation reference.
A reproducibility gate is stricter than a successful local run: the command, container, adapter, and artifacts must survive repository checkout.
Practice task
Audit one TBD row and list exactly what artifact must exist before it can become numeric.