Momentum Shift Detector
The Momentum Shift Detector analyzes real-time match events to identify significant momentum changes. It uses a sliding window approach over event data — shots, possession changes, pressing intensity, and territorial control — to detect inflection points. When a momentum shift is detected, the agent generates a signal with confidence scoring based on the magnitude and consistency of the shift. Historical backtesting shows these shifts correlate with goal-scoring probability changes within the next 15 minutes.
Agent Logic & Documentation
Core Logic
Data Sources - Live event stream (goals, shots, fouls, corners, possession) - xG model output (rolling 10-minute windows) - Pressing intensity metrics - Territorial control zones
Algorithm 1. Calculate rolling event density per 5-minute window 2. Apply change-point detection (CUSUM algorithm) 3. Cross-reference with xG flow differential 4. Generate momentum score: -1.0 (away dominant) to +1.0 (home dominant) 5. Signal emitted when score changes by > 0.3 within 10 minutes
Confidence Scoring - Base confidence from change-point p-value - Boosted by xG alignment (+0.1 if xG flow confirms) - Reduced by low event density (-0.1 if < 5 events in window)
Known Limitations - Less reliable in low-event matches (0-0 tactical battles) - Early match signals (0-15 min) have lower accuracy - Weather conditions not yet factored
Community Feedback
3Really clean implementation of CUSUM for sports data. Have you considered adding a Bayesian changepoint detection as an alternative? Might handle the low-event problem better.
Been using this in my pipeline for 3 weeks. The xG alignment boost is a nice touch — catches a lot of false positives.
Great agent. We've noticed it pairs particularly well with the Set-Piece Agent for corner kick momentum cascades. Worth exploring.
Have feedback for this agent? Join the builder community.
Join as a Builder