Time and course-split validation: methods, samples and limits

Methods and aggregate results from re-running the existing validation against current recorded data.

Time-prediction sample and method

Last validated 2026-11-05: 139 runners, 326 comparisons and 38 races. An index inferred from another race is used to estimate a target-race time. The target runner’s matching time and race-score reference is excluded.

Results and interpretation

Median absolute relative time error was 4.9%; 79% were within ±10% and 97% within ±20%. Comparisons include repeated runners, not 326 independent people.

Distinguish retrospective evaluation from future accuracy

This is retrospective evaluation. The other result is not always earlier than the target, and other finishers of the target race can remain in the references. It is not a prospective test that excludes all information about future editions, nor identical to entering your current reported index. Results are not a guarantee for every runner or race.

Course splits are evaluated separately

Terrain-based split allocation was evaluated on 74 races, 375 bands and 2205 segments; 88.8% were within ±20% of observed splits. This is not a measure of finish-probability accuracy. Updating these explanations does not automatically change scoring or prediction coefficients.

Sample limitations and privacy

The sample favors recorded races and runners with multiple results. It cannot explain every course change, surface, weather, fueling or health condition. This page publishes aggregates only, without runner names, IDs, activities or individual result rows.

Compiled from each race's official information. Details can change, so always check the race's official website for the latest. Spotted a mistake? Please tell us through the contact form.
TRAIL STATUS · FAQ · Privacy policy