This repository is a trading/research project built around Kalshi event markets, with a newer streamlined pipeline for data collection, model training, evaluation, and live execution, plus a legacy archive of earlier strategy experiments.

If you are browsing this as a portfolio project, scroll down to the "Sample Output Gallery" section to view example PNG outputs from archived experiments (illustrative only).

• A refactored "active" pipeline for:
  • high-resolution market snapshot collection
  • label alignment from settled markets
  • RNN-based edge model training
  • strategy evaluation (fixed + walk-forward)
  • dry-run / live execution
• A legacy archive of prior scripts, backtests, and visual outputs kept for reference (legacy/)

• Unified snapshot schema (pipeline/schemas.py) used across training and runtime
• Reusable data + feature pipeline (pipeline/data.py)
• Torch LSTM classifier for market microstructure sequences (pipeline/model.py)
• Pluggable strategy runtime with built-in rnn_edge_v1 and plugin support (pipeline/strategy_runtime.py)
• CLI scripts and Makefile targets for end-to-end operation

• scripts/: entrypoints for download, labeling, training, evaluation, and live execution
• pipeline/: core training/runtime logic and schemas
• market_data/: Kalshi HTTP/WS clients and market-selection helpers
• config/: runtime config defaults (paths + endpoints)
• legacy/code/: archived strategy research stacks and backtest utilities
• legacy/artifacts/: archived charts and experiment outputs

1. Collect high-resolution snapshots (scripts/download_highres.py)
2. Build or sync settlement labels (scripts/build_labels.py, scripts/sync_labels_from_api.py)
3. Train an RNN edge model (scripts/train_edge_rnn.py)
4. Evaluate strategy performance (scripts/evaluate_strategy.py)
5. Run the strategy executor (scripts/run_live.py)

• Secrets are intentionally excluded from version control (secrets/, .env, *.pem).
• Runtime data and local training outputs (for example pipeline_data/) are excluded from the repo.
• Some legacy datasets / raw market-data caches are excluded, while legacy scripts and charts are retained.

• Install dependencies from requirements.txt
• Create local auth files (see config/runtime.json and .env.example)
• Use make help to see common commands

For the detailed operational guide for the active pipeline (commands, config fields, and workflow examples), see README.pipeline.md.

The PNGs below are sample output displays from archived experiments and diagnostics. They are illustrative only and are not intended to reflect accurate market data, validated signals, or real trading performance.

Sample output displays only (illustrative; not accurate market data).

Sample output displays only (illustrative; not accurate market data).

This is research/trading software and not financial advice. Use live trading functionality at your own risk.