2-Week Trajectory

Source: (inline in dashboard/page.tsx)

---

What You See

A card with three columns side by side: Maintain, Push, and Ease Off. Each column shows a projected finish time for 2 weeks from now. Below each time is a number showing how many seconds faster or slower that scenario is compared to today. A small confidence percentage may appear at the bottom of each column.

How the Numbers Work

The backend simulates three training scenarios for the next 14 days. 'Maintain' assumes you keep training at the same level. 'Push' assumes you increase training load. 'Ease Off' assumes you reduce volume. Each scenario runs through the projection engine to produce a predicted finish time. The delta is the difference between that scenario and your current projection.

Step by Step

Models & algorithms used: DTW, Dynamic Time Warping, Gradient Boosting, LSTM

PIRX tries two methods to predict your trajectory. The primary method uses Dynamic Time Warping (DTW): it divides your training history into 14-day blocks, creates a fingerprint for each block (distance, duration, pace, heart rate), z-normalizes the values, and finds the 3 most similar historical blocks using DTW distance. It then looks at what happened to your projection in the 14 days after each similar block. Maintain uses the mean outcome, Push uses the 75th percentile, and Ease Off uses the 25th percentile. If fewer than 4 historical blocks exist, PIRX falls back to a heuristic: it multiplies your current features by scenario factors -- Push scales volume by 1.05x and intensity by 1.15x; Ease Off scales volume by 0.80x and intensity by 0.90x; Maintain keeps everything at 1.0x. The adjusted features are run through the projection engine (or the Gradient Boosting / LSTM model if available) to produce each scenario's predicted time. Confidence levels are set at 0.85 for Maintain, 0.65 for Push, and 0.75 for Ease Off.