Training Jobs

Quant training is launched from the Jobs surface, not from the old PyPScript training wizard.

The mental model is:

• Research is where you author
• Jobs is where you build
• Artifacts is where you inspect outputs
• Releases is where you package validated outputs

How quant training works

1. Save strategy.py and quant.config.json
2. Open /quant/jobs
3. Launch a job with the selected preset
4. PyP creates a QUANT_JOB
5. GitHub Actions runs the training workflow
6. The compiler uploads the artifact and metrics back into PyP
7. The completed output appears in Artifacts

What the training environment does

The training workflow:

• writes the strategy and config to a temporary job workspace
• downloads supported market data
• imports the strategy module
• calls train(data, config)
• emits an artifact bundle plus metrics

This is why the strategy contract and config shape matter so much. The training system is structured, not ad hoc.

Job statuses

Typical statuses:

What to expect after completion

A successful job should give you:

• metrics
• artifact records
• versionable output for release creation

It is better to think of a quant job as an artifact build step than as a one-off notebook run.

Best practice

• keep presets narrow and intentional
• compare jobs by metrics and artifact target
• use releases to promote outputs, not raw jobs

The job is an execution event. The artifact is what downstream runtime actually cares about.