Seven editions of La Marmotte, and the 25,208 people who finished one

Seven editions, and no gap. That is worth saying because the sister event on the same mountain — the Triathlon de l’Alpe d’Huez, whose archive I built last week — simply has no 2020. La Marmotte did not cancel. It moved: 2020 and 2021 were both run in September rather than July, and 2020 went ahead with 1,827 finishers against the 5,005 of 2026. A quarter of the usual race, four months late, up the same mountain.

The field has otherwise grown steadily, and the women’s share with it — from 3.7% of finishers in 2020 to 5.2% in 2026. That is a real trend and also a small number: this remains a race that about ninety-five people in a hundred ride as men.

La Marmotte freezes the clock on its dangerous descents. Off the Glandon, and in later years through the La Grave traverse as well, your time is taken at the top, taken again at the bottom, and the difference handed back to you. The official classification runs on that adjusted net time. Your gross time — the one your own computer shows — is typically half an hour longer.

Both numbers get published. Granfondo Daily News prints net; Granfondo Guide prints gross. Neither says which. For the 2026 edition the two outlets printed different podiums for the same race, because the gross order is not the net order.

The 2020 edition settles it from the inside. That year’s results survive only as a sixty-three-page PDF, and the PDF carries both columns. Michiel Minnaert’s official time is 5:30:04. The figure the press reported for him is 5:55:19 — which is, to the second, the gross column of that same row. So the discrepancy is not an error anybody made. It is two outlets reading two different columns and neither of them labelling it.

Everything on this site is net, everywhere, because that is what the organiser ranks on: sorting the field by gross time disagrees with the published order thousands of times per edition.

The published winner of 2024 is bib 6058, at 4:50:09. That is an average of 36.6 km/h over 177 km and five thousand metres of climbing. Across all 25,208 classified rides in this archive, it is the only one above 33.5 km/h. Every other winner in seven editions sits between 31.6 and 34.6.

The detail document says what happened, and it is almost boringly mundane. His finish row carries a start clock of 07:00:22. His two mat rows carry a different one: 07:57:28, fifty-seven minutes later. His finish clock is 12:47:37. And 12:47:37 minus 07:57:28 is exactly 4:50:09. His winning time is his finish clock minus a start he did not use.

Two other things corroborate it. He has 2 mat reads for the whole day where the riders around him have eleven, and both of those reads are timestamped after he is supposed to have finished — one of them puts him at kilometre 163 two hours past his own finish, and in the wrong order relative to kilometre 164.

The rider who actually rode the fastest race that day is Giuseppe Orlando, published second at 5:23:16 — which is exactly the time the cycling press reports as the 2024 winning time. The gap between the artefact and the real ride is 33:07.

The explorer still shows bib 6058 in first place, because that is the classification the organiser published and it is not my race to re-rank. What it does not do is call him the winner without saying all of the above.

The neutralisation only works if the timekeeper sees you at both ends of it. Miss the mat at the top or the mat at the bottom and there is no interval to subtract, so no credit is applied — and nobody compensates. Your official time then contains a descent that everybody you are being ranked against had removed from theirs.

It is not rare. 285 classified riders across the seven editions were credited nothing at all, and in every single case the cause is the same: a missing mat read at one end of a frozen section. In 2024 that cohort is 74 riders deep and reaches the very top of the results.

There is a rarer and stranger version. Five riders in the archive carry a negative credit — the timekeeper derived it from a pair of mat reads that were in the wrong order, got a negative number, and subtracted it. Subtracting a negative adds. Those five were ranked on a time longer than the one they actually rode, which is a novel way to lose a bike race.

Roadworks made the Glandon descent unrideable in 2023, so the race went over the Col de la Croix de Fer instead and then, rather than dropping straight into the Maurienne, descended and re-climbed the Col du Mollard. That is 185.6 km against the usual 177, eleven timing mats instead of nine, and three neutralised descents instead of two.

The reported distance for that edition varies by source — I have seen 186 and 187 km in print. It is neither. The timekeeper’s own average-speed column, multiplied by each rider’s gun time, reproduces 185.6 km across 3,684 finishers with a standard deviation of three metres, and the sibling Ultrafondo event labels the same physical mat “185km Alpe”. So 185.6 it is, measured rather than quoted.

None of that makes 2023 a slow year. It makes it a different race, and no time comparison on this page crosses that line without saying so.