Context
The transcript review noted that the saved-H5 scoring path accepts aggregate-quality datasets (reject_aggregate=False / minimum quality aggregate) while the main dataset path rejects aggregate-quality inputs. That may be deliberate for approximate years, but it must be explicit in docs, manifests, and release gates.
Work
- Document the difference between main dataset resolution and saved-H5 scoring quality thresholds.
- Add manifest fields for calibration quality and whether aggregate-quality scoring was accepted.
- Require explicit CLI flags/metadata for approximate or aggregate-quality years.
- Ensure the dashboard/paper caveat approximate late years where applicable.
Acceptance Criteria
- No saved-H5 score can silently bypass stricter main-pipeline calibration quality without release metadata.
- Approximate years, if any, are visible in artifacts and QA reports.
- The 75-year release distinguishes exact/approximate dataset quality by year.
Related: #84, #86, #97, #103.
Context
The transcript review noted that the saved-H5 scoring path accepts aggregate-quality datasets (
reject_aggregate=False/ minimum qualityaggregate) while the main dataset path rejects aggregate-quality inputs. That may be deliberate for approximate years, but it must be explicit in docs, manifests, and release gates.Work
Acceptance Criteria
Related: #84, #86, #97, #103.