Methodology
This page documents how the numbers on this site are produced, and — more usefully — where they are fragile. It is written so that any individual page can be checked against it.
What the data covers
The results used across the site come from publicly accessible information about HYROX competitions. That information is reprocessed and consolidated to produce the profiles, rankings and statistics shown here. Nothing is invented, estimated or filled in: a value that is not available is left empty rather than approximated.
Raw facts — a time, a rank, a date, a venue — are not our creation. What we build is the consolidation: linking a result to a person, a person to a career, and a career to a comparable population.
How a result is attached to an athlete
This is the hardest part of the job, and the one that produces most of the errors we correct.
Participation is not identity
Each race entry carries an identifier that is unique to a participation, not to a person. It is stable enough to deduplicate results, but it is shared by both members of a doubles team and it is never enough on its own to say that two entries belong to the same human being.
Namesakes
Two different people with the same name can end up merged into a single profile. When that happens we separate them through a persistent, reviewed override rather than a heuristic — because a heuristic that splits careers automatically would break more profiles than it fixes. The authority on who an athlete is is the athlete's own declaration, not a similarity score.
Duplicates
The same performance can appear more than once when an event is republished under a different identifier. A detector flags entries that share a participation identifier across events, and confirmed duplicates are removed.
How often it refreshes
Results are collected as events publish them, and aggregate statistics are recomputed on a daily schedule. Sitemaps are regenerated daily. Editorial pages carry a visible last-reviewed date derived from the actual change history of the page, not from the day you happen to load it — a date that moves every day is noise, not freshness.
How predictions are produced, and where they are weak
The prediction tools estimate a finish time from an athlete's own history. They are explicitly not a ranking, and they carry their own uncertainty rather than hiding it.
What makes a prediction reliable
- Recent solo races in the target category. Solo history is the only clean signal of individual performance.
- For a team, races the same pair has already run together — a known shared running pace and a settled split across stations is worth more than two strong individual profiles.
What makes it unreliable, and is flagged as such
- No history in the target category. Predicting a Pro race from Open history means guessing how an athlete absorbs heavier loads.
- Running splits from doubles races. A doubles pace is the slower partner's pace, so it says little about either athlete individually and is never used to estimate individual running.
- Runs 1 and 8 are not exactly one kilometre — venue layouts trade distance between them — which makes those two segments the least predictable of the race.
- Old data. A race more than a year or two back describes a different athlete.
Every prediction exposes a confidence value, a reliability level and the specific reasons it may be off. When there is not enough data, the tool says so instead of producing a confident-looking number.
What we do not do
- We do not publish a time, a rank or a statistic that we cannot trace back to a recorded result.
- We do not sell position. No club, coach or athlete can pay to rank higher, appear more often, or have a figure changed.
- We do not take a commission on anything you buy, and we do not let a commercial relationship change editorial content. Where a partner section appears inside an article, it is labelled and marked as sponsored, and the advice around it is written without reference to it.
- We do not present training or nutrition guidance as medical advice, and we say so on the pages concerned.
When something is wrong
Data at this scale is wrong somewhere, always. A missing race, a merged namesake, a closed club, a result attributed to the wrong person — these are reported to us regularly and fixed.
How to report an error