Arc Research Arc Research
Research
Stories Chats Research Agents Source Documents Feeds Reports Desk Knowledge Explorer
Commodities
View all commodities
Blog Agents Chat Archie Desk Quote Board Docs
Sign up Log in

Alpha

Build and train

End-to-end loop for Commodity ML forecast models. Requires Arc Research Starter plus the Commodity ML alpha flag.

  1. Export training data for a commodity (scheduled jobs or admin export). Each export pins a feature schema and CSV split.
  2. Create or open a model under Commodity ML. Alpha users work on their own models; admins see all.
  3. Start an experiment with a parameter grid (JSON). The default sweeps reg_param and standardize_features.
  4. Wait for the sweep job. Each trial trains a Rumale ridge model and scores walk-forward vs naive.
  5. Publish (alpha/owner): set visibility to subscribers/published for your model — never becomes platform core.
  6. Promote to core (admin only): marks the trial’s model as platform core so PublishForecastsJob prefers it for public/paid forecast pages.
Need access? Subscribe to Arc Research Starter, then email [email protected] or ask Archie to request Commodity ML alpha access.

Agents can run the same flow via MCP — see the Forecast modeling (alpha) tool group.