Your data in Heimdall
How Heimdall stages data before ML or Forecast training.
One page to rule them all
Everything lives under Data in the app (/data) — bronze, silver, and gold in one catalog.
Three stages
| Stage | In the app | Meaning |
|---|---|---|
| Bronze | Bronze tab | Raw files as uploaded (or from Forge) |
| Silver | Silver tab | Optional: join, filter, drop columns |
| Gold | Gold tab | Validated, ready for ML or Forecast |
Common paths
CSV → ML
Bronze upload → Gold (ML) → ML wizard picks gold
Two tables → forecast
Bronze ×2 → Silver join → Gold (Forecast) → Forecast wizard
Zip of images
Forge → Bronze table → Silver/Gold if needed
Forge vs CSV
- CSV / Excel → bronze directly
- Zip (images, text folders) → The Forge first
Next steps
- Add your first table
- Publish for modeling
- No data yet? → Sample quick start
FAQ
Old “upload in ML wizard”?
Gone for production. Train from gold only; samples are the exception.
Do not skip gold for production
Training without a published gold artifact bypasses validation. Bad columns or dtypes can fail mid-train.