Open method, public results

How Trotedge works

A model built on every French trotting race since 2013, tested year by year on the past, and now tested live in public. This page explains what we do, what the test shows, and when a tip is worth acting on.

86,000French races since 2013 in the model
9 yearstested one by one at ZEturf's prices, trained only on earlier years
2 minutesbefore the start: when the real tip exists

1 · The data

Every PMU programme and result since 2013: 86,000 races and 1.7 million starts, with drivers, trainers, shoeing, post positions, morning and final odds, km times, dividends and the race readers' comments. LeTrot careers back to 1981 and Swedish ATG races complete the picture.

2 · What the model looks at

103 facts per horse that were known before the race: recent form, best and recent km times, days of rest, experience on the distance and the track, driver and trainer strike rates, class moves, shoeing, post position, how the market rated the horse before and whether it delivered, track bias, weather, signals from the comments (hampered, raced wide, finished strongly), the driver's own form over the last 7, 14 and 30 days and, since version 1.4 (October 2026), how the post position does over this distance at this track and the kind and recency of the horse's disqualifications.

3 · The model

Two models give each horse a chance: a conditional logit of the kind professional syndicates have used since the 1980s, and gradient boosting, with larger trees since version 1.3 (October 2026). Their chance is then blended with the betting market, roughly one quarter model and three quarters market, plus a term for money arriving late. The blend predicts better than either alone, in every year of the test.

4 · The test on the past

For each year from 2018 to 2026 the model is trained only on earlier years, then bets €1 on every horse whose chance times odds beats the price by at least 10%, paid at ZEturf's real dividend, the pool our members bet into. At the final odds that earns +18% over all French meetings, but the difference between meetings is large: +55% at regional and +37% at local meetings, against +3% at the national meetings (Vincennes, the Quinté+), where the market already knows almost everything. At the morning odds the regional and local meetings still earn about +12%, positive in every year from 2018 to 2026.

5 · Why timing is everything

The final odds are only known when the race starts. Two minutes before the off nearly the whole market, including the late money, is in the price: that is when we compute the value and, if it is there, publish a late tip. At regional and local meetings ZEturf's pool is small and its odds can run far above PMU's just before the start, then fall back, so we measure the value on the lower of the two prices.

6 · The live test

Since October 2026 every late tip is recorded two minutes before the start with the odds of that moment and settled with the official result. Late tips are given only at regional and local meetings. Pre-race and late tips are shown separately on the results page, and nothing is ever removed. A few hundred late tips will tell whether the edge survives on prices you can actually bet at.

7 · Other countries

The Netherlands: every Dutch result since 2012 is in our data and a Dutch model is live. In its test it beat the tote only from 20% value (+60% on 134 bets), so that is the threshold for Dutch tips. Sweden: our Swedish model does not yet beat ATG's market, so there are no Swedish tips.

What this is not

Not a guarantee, and not a way to get rich: tote pools are small, so even a working edge supports modest stakes before it moves the odds against itself. Bet only what you can afford to lose.

See the live results → Today's tips →