Skip to main content

Infrastructure-grade execution.

Misprice is a production system running on EC2 with a modular architecture designed for auditability and extension.

Execution Pipeline

1
Data Aggregation
Real-time prices from Polymarket, Kalshi, and public sources
2
Probability Engine
AI models (Vertex AI) estimate calibrated probabilities
3
Truth Gap Detection
Flag markets where model and price diverge by ≥3%
4
Gate Framework
Block low-confidence signals (stale data, timing, source health)
5
Portfolio Tracking
Log decisions, track P&L, measure calibration
6
Evidence Pipeline
Generate hash-verified artifacts for every prediction

Orchestrator

  • Python-based scheduler
  • Configurable cycle intervals
  • Health check integration
  • Graceful degradation modes

Adapters

  • Pluggable architecture
  • Standardized output schema
  • Source health tracking
  • Rate limiting + retry logic

Snapshot Store

  • SQLite (proof phase)
  • Time-indexed queries
  • Hash verification on read
  • PostgreSQL planned for prod

Truth Models

  • Modular model registry
  • Weighted synthesis
  • Confidence intervals
  • Calibration tracking

Edge Detector

  • Compare truth to live prices
  • Apply confidence gates
  • Liquidity thresholds
  • Surface or block signals

Evidence Log

  • Append-only structure
  • SHA-256 chaining
  • JSON-serialized entries
  • Export for audit

Deployment

EC2 (t3.micro)Python 3.11SQLiteSystemdCloudWatch

Running 24/7 during proof phase. ~$12/month infrastructure cost.