v0.3.1: System-One / Jev-style portfolio decisions for quant pipelines

[initial-d]

ml-quant-trading v0.3.1 adds a preview System-One / Jev-style portfolio decision surface.

The core framing is:

pipeline state + typed portfolio questions -> typed portfolio-intent answers

This is different from putting an external model inside the strategy pipeline. The pipeline itself can now expose factor state, masks, model scores, current weights, target weights, tradability, liquidity, risk estimates, and cost assumptions as a structured decision state.

The adapter can answer typed portfolio questions such as:

• asset_action: increase / hold / decrease / exclude
• target_weight: bounded research target weight
• trade_allowed: whether tradability, liquidity, and risk checks allow the asset
• risk_bucket: low / medium / high / blocked
• max_participation_rate: research participation constraint

This is intentionally scoped as research portfolio intent. It does not add broker execution, account access, order routing, live services, official Jev integration, Jev certification, or calibrated probabilities.

Try it after running the demo:

mlquant systemone build-state \
  --artifacts-dir artifacts/small \
  --as-of-index -2 \
  --output artifacts/systemone/decision_state.json

mlquant systemone decide \
  --state artifacts/systemone/decision_state.json \
  --output artifacts/systemone/portfolio_decision.json

For offline diagnostics, rebuild the state with --include-realized and run mlquant systemone evaluate.

Links:

• Release: https://github.com/initial-d/ml-quant-trading/releases/tag/v0.3.1
• Adapter docs: https://github.com/initial-d/ml-quant-trading/blob/main/docs/systemone_adapter.md
• Chinese article: https://github.com/initial-d/ml-quant-trading/blob/main/docs/article_zh_quant_jev_portfolio_decision.md
• Project facts: https://github.com/initial-d/ml-quant-trading/blob/main/docs/answer_index.md

Useful feedback would be:

• Does this typed decision surface match how you would expose a quant pipeline to agents or evaluation systems?
• Which additional portfolio questions should be included before this becomes a stronger reference interface?
• Can you run the adapter on your own research artifacts and share a redacted evaluation note?