AI Coach Intelligence

ENDURE uses Claude AI and machine learning models to provide coaching-quality insights at /intelligence.

Discoveries

The Intelligence page surfaces AI-generated discoveries about your training, organized by importance. Discoveries are computed from 12 months of activity data and include:

• Performance trends — Improvements or regressions in specific power durations
• Training pattern insights — How your actual training distribution compares to your methodology's targets
• Recovery patterns — Correlations between recovery behaviors and performance
• Compliance patterns — Trends in workout adherence

Discoveries are tier-gated: Free tier users see 3 discoveries; paid tiers see unlimited.

Anomaly Detection

The system runs daily anomaly detection looking for:

• HRV drops — Unusually low HRV relative to your baseline (possible overtraining or illness)
• Resting HR spikes — A jump of 10+ bpm above baseline
• Sleep deprivation — Multiple nights of insufficient sleep
• Performance drops — Unexpected power loss relative to recent capabilities
• Readiness cliff — Abrupt shift from GO to REST
• Injury risk — High training load combined with high fatigue indicators

Anomalies are viewable at /insights/anomalies with a timeline showing detection events, severity, and details. Critical anomalies also trigger a banner on the Today dashboard.

Correlation Matrix

ENDURE computes correlations between your training metrics (e.g., HRV vs. CTL, sleep quality vs. performance, stress vs. readiness). The correlation matrix helps identify which factors most influence your readiness and performance.

Pre-Workout Prediction

Before each workout, the prediction engine estimates how the session will feel based on your current fatigue state (CTL, ATL, TSB), recent training load, sleep data, and recovery metrics. The prediction uses a decision tree model trained on your historical data. It shows:

• A feeling prediction (from "Very Weak" to "Very Strong")
• Feature importance bars showing which inputs matter most
• Key factors influencing the prediction

What-If Simulator

Adjust your planned training load for the next 7 days and see how it affects your projected CTL, ATL, TSB, and readiness probability. Each day has an editable TSS input (0-500). The forecast updates with a 500ms debounce as you change values.

Default weekly template: Mon 80, Tue 60, Wed 40, Thu 80, Fri 30, Sat 100, Sun 0 TSS. Useful for planning: "If I add a rest day Wednesday, will I be fresh enough for Saturday's race?"

Machine Learning Infrastructure

ENDURE's ML pipeline includes:

• Model registry — Training, evaluation, and deployment of prediction models
• Decision tree and random forest models — For readiness and performance prediction
• Forecast models — Exponential smoothing and PMC-based projections
• Feature engineering pipeline — Automated feature extraction from raw training data
• Shadow mode — New models run in shadow mode before being promoted to production

---