| name | elfquake-data-refresh |
| description | Use when running ELFQuake prospective real-data refresh scripts, including relabeling matured VLF rows and rebuilding real model input artifacts. |
ELFQuake Data Refresh
Run from the repository root. Keep data Italy-scoped, preserve raw records, normalize timestamps to UTC, and do not claim prediction value from refreshed data alone.
Usual Order
- Refresh real captures and prospective labels:
./scripts/refresh-prospective-labels.sh
- Rebuild real model inputs:
./scripts/prepare-real-model-inputs.sh
- Check readiness JSON before training real models:
rg -n "status|labeled_row_count|positive_count|negative_count" data/derived/models/*real_vlf_aligned_windows.readiness.json data/derived/multimodal/*labeled.readiness.json
Only run real training when readiness shows both positive and negative examples. If class variation is absent, report the blocker and keep collecting data.
Useful Follow-ups
Use these only when relevant:
./scripts/materialize-real-vlf-sequence.sh
./scripts/train-real-tabular-model.sh
After meaningful count or status changes, update docs/report.md and docs/next-actions.md concisely.