Skip to content

live_demo.md

Latest commit

 

History

History
42 lines (31 loc) · 1.46 KB

live_demo.md

File metadata and controls

42 lines (31 loc) · 1.46 KB
Raw

Live Paper Demo

This project can run the learned selector on live public top-of-book quotes without placing orders.

Demo Command

python3 research/run_paper_live_report.py configs/runs/kraken_solusd_learned_selector_maker_fee_paper_live_long.json \
  --run-id solusd_long_20260711_001

The command runs the Rust paper_live engine, writes a CSV decision log, analyzes the run, and renders a Plotly HTML report.

Latest Demo

  • Run ID: solusd_long_20260711_001
  • Pair: SOLUSD
  • Samples: 300
  • Interval: 2s
  • Policy: learned_selector
  • Fills: 70
  • Final inventory: -0.2000
  • Inventory range: -0.3000 to 0.4000
  • Final paper PnL: 0.0807
  • Max drawdown: 0.0059
  • Average quote distance: 0.0190
  • Policy mode split: 282 adaptive steps, 18 static steps
  • Main triggers: configured, spread, and inventory

Generated artifacts:

target/reports/paper_live/solusd_long_20260711_001.csv
target/reports/paper_live/solusd_long_20260711_001.meta.json
target/research/solusd_long_20260711_001.html

Interpretation

This is a live public-data paper demonstration, not a trading result. It shows that the learned policy trained by Python can be exported to JSON, loaded by Rust, and executed against live public quote snapshots end to end.

The longer run stayed active, kept inventory bounded, and produced paper fills under the configured fill model. The result should be treated as an operational demo, not evidence of a live edge.