stoic-kal/betting-model

Purpose and entry points

This is an MLB betting-model platform: it ingests baseball/odds data, builds features, trains and promotes moneyline/totals models, generates daily picks, tracks outcomes, and exposes results through a Python web application plus iOS and analytics-dashboard clients. Primary runtime entry points are app.py, run_betting_model.py, generate_game_card.py, and src/production_runner.py.

Application/API layer

The Python application is organized as a route-and-service system. Route modules present game, pick, result, analytics, diagnostics, and system endpoints; services own business workflows such as pick generation, model control, execution, reporting, and notifications.

• app.py

• web/routes/picks_routes.py

• web/routes/game_routes.py

• services/pick_service.py

• services/execution_service.py

• services/model_control_service.py

Data ingestion and feature stores

Historical and live workflows consume local baseball statistics, Statcast-derived tables, retrospective splits, and Vegas odds files. Pipeline ingestion/build stages normalize these sources into versioned game-feature datasets; live features are a boundary between scheduled games/current context and inference.

• pipeline/retrosplits_ingestion.py

• pipeline/vegas_ingestion.py

• pipeline/statcast_store.py

• pipeline/build_features_v2.py

• pipeline/live_features_v2.py

• data/games_v2_features_final.csv

Modeling, calibration, and promotion

Separate moneyline and totals training pipelines share common training/calibration machinery. Model artifacts are versioned in a filesystem registry, with latest.json pointers, metadata, promotion reports, and validation/audit outputs; this is the deployable-model contract rather than an external model service.

• pipeline/train_moneyline.py

• pipeline/train_totals.py

• pipeline/train_experimental_candidates.py

• pipeline/model_registry.py

• pipeline/retrain.py

• models/registry/moneyline/latest.json

• models/registry/totals/latest.json

Prediction and betting decisions

Production prediction combines trained artifacts with game context, odds/market signals, and prediction math to produce picks. Supporting services cover portfolios, shadow inference, opposite-model comparison, execution recording, and postgame/loss learning. Daily and historical pick files are durable handoff outputs.

• src/get_game_predictions_v2.py

• services/ml_model_v3.py

• services/totals_model_v3.py

• services/market_signals_service.py

• services/portfolio_risk_service.py

• results/daily_picks.csv

Operations, results, and external integrations

The service layer integrates odds feeds, Discord notifications, and Pikkit-related tracking, while scripts poll/fetch odds and import results. Deployment artifacts indicate Docker Compose and macOS launchd service operation. No cloud database or queue is evidenced in the tree; CSV/JSONL/JSON files are the visible persistence boundary.

• services/odds_feed_service.py

• services/notifications/engine.py

• services/pikkit_service.py

• scripts/odds/poll_odds.py

• docker-compose.yml

• deploy/com.jingleez.flask.plist

User-facing clients and analytics

The server-rendered web UI is under web/templates. A native SwiftUI iOS app calls an API client and presents Today, Results, Insights, Alerts, and Account views. projects/athena is a separate React/TypeScript/Vite diagnostics frontend with API/stream clients, state stores, and feature pages for calibration, CLV, drift/data health, edge, and prediction analysis.

• ios/JingleezMobile/JingleezMobile/APIClient.swift

• ios/JingleezMobile/JingleezMobile/JingleezMobileApp.swift

• projects/athena/src/services/api.ts

• projects/athena/src/app/router.tsx

• projects/athena/src/features/overview/OverviewPage.tsx

Research and governance boundary

Offline research supports leakage-aware historical replay, calibration/root-cause diagnostics, and reporting. These workflows consume data/model outputs and generate experiment artifacts; they should remain separate from live pick execution.

• research/historical_replay/cli.py

• research/historical_replay/engine.py

• research/diagnostics/totals/run_all.py

• pipeline/execution_audit.py

• reports/execution/complete_execution_audit.json