Shiva hunts Polymarket NBA and MLB game-day markets where the model's probability diverges from the market price by ≥5%. Moneyline, spread, over/under, and player props — anywhere there's real information edge against a crowd anchored to recency bias and narrative. 243 unit tests, 12 Playwright end-to-end tests, 92% coverage.
The 6-factor probability model pulls from ESPN's unofficial API (no key required) for NBA stats, injuries, team form, and schedules, plus the MLB Stats API for baseball data. Each estimate combines: home court/field advantage, rest days differential, recent form (last 5 games), injury impact, head-to-head history, and media sentiment from Perplexity. The edge calculator then computes EV, Kelly fraction, and signal strength before the risk gate decides whether to act.
Market types supported: MONEYLINE (game winner), SPREAD (point spread), TOTAL (over/under), PROP_POINTS, PROP_REBOUNDS, PROP_ASSISTS. The trader filters price 0.10–0.90 to skip near-resolved markets. ShivaClaimer runs in the background monitoring resolved markets and updating PnL in real time.
Named for the Destroyer. Markets that underestimate variance get destroyed. GAMMA API polling every 30 seconds. SQLite logging to shiva.db. Mission Control /shiva page with 4 tabs: Positions, Performance, Controls, Signals.