Mistakes & AI Coach
The technical logic behind automatic mistake quantification.
Routster's AI Coach doesn't just look for "time lost"—it categorizes why time was lost. By mathematically analyzing the geometry and velocity of your GPS track, it breaks mistakes down into actionable insights.
Private vs Public Analysis
Mistake Detection Heuristics
The engine runs your GPS track through a pipeline of geometric and velocity-based filters. Here are the exact thresholds used to classify mistakes:
| Parameter | Default | Description |
|---|---|---|
| Stops | > 12:00 min/km | Triggered when your pace drops below the walking threshold. Multiple consecutive slow points are aggregated into a single stop event. |
| Slowdowns | > 1.5x Median Pace | Detected dynamically. The engine calculates a 60-second rolling median pace for the leg. If your pace drops to 1.5x slower than this median, it is flagged as a hesitation or slowdown. |
| Off-Route | > 30 meters | Requires a Guessed Route. Calculates the perpendicular distance from your GPS track to the optimal drawn route. If the distance exceeds the threshold, time spent out there is flagged. |
| Backtracks | > 120° turn | Analyzes the angle between consecutive track vectors. If you reverse direction by more than 120 degrees, it calculates the time wasted running backwards and returning to the original point. |
The 60-Second Rolling Median
Why use a rolling median instead of a flat threshold for slowdowns? Because terrain changes. If you enter a steep, thick green marsh, your pace naturally drops. A flat threshold would flag this as a mistake. By using a 60-second rolling median, the engine learns what "normal running speed" is for that specific section of terrain, and only flags you if you suddenly drop well below that baseline (e.g., stopping to read the map or hesitating at a feature).
Integration with Per-Leg View
Mistakes aren't just listed; they are intrinsically linked to the map. Clicking on a mistake in the AI Coach sidebar immediately rotates the Per-Leg Navigator to the correct leg and flashes a pulsing beacon at the exact (x, y) pixel coordinate where the mistake occurred.