The archive · counted, not modelled
Every runner who has stopped in a race we hold is recorded at the checkpoint they last reached. That is 14,625 of them across the whole archive, going back to 1994. Put those next to everyone who kept going, and each aid station tells you the truth about itself.
Ten races have enough editions on record to say anything useful. Those are the ones below. The rest are in the archive but not yet deep enough to publish.
No race organiser publishes this. We are not guessing at it either: every number below is a count.
Your runner just came through
A deadline marker is not a cause of death. Some checkpoints carry a published cut-off and the clock is genuinely what stops people. Others have no deadline at all and still end more races than anything near them, because they are where the crews wait, the beds are, and the race is half done. UTMB’s Courmayeur ends more runners than almost any point on that course and mostly has no clock on it. Champex-Lac carries a cut-off in every edition and 93% of the runners who reach it still finish. The marker says only that a deadline exists there, taken from each race’s own published course file. Deadlines are available for UTMB and Lavaredo so far; the other families on this page publish none we can read, and we would rather show nothing than guess one.
The four columns are arrival time, not ability. Everyone ever recorded at a checkpoint is sorted by how long they took to get there and split into four equal groups, fastest quarter first. The number in each is the share of that group who went on to finish that same race. The clock range underneath is the actual elapsed time that group arrived in, so you can find your own runner in it.
A runner's last checkpoint is the last place we have a record of them. It is not necessarily where they withdrew. Somebody logged at mile 31 may have walked another eight miles before stopping. We count where the record ends, because that is what we know.
This is history, not a forecast. It says what happened to everyone who came through here before. It says nothing about whether one specific runner will finish today, and we will not print a number that pretends otherwise. Cutoffs, courses and weather all move between years, and pooling editions cannot separate those.
Thin data is left out rather than smoothed. A checkpoint needs at least 300 recorded runners across at least three editions, with at least 40 in each quarter, before it appears here at all. How we track, and how wrong it gets.