Styrian Grand Prix 2021

Each forecast (race before the sprint, race after the sprint, and sprint) is compared with the official result once the race has been run, only if there are at least 8 classified finishers. It measures: whether the predicted order resembles the real one (rank correlation, from 0 = no relationship to 1 = identical), the average error in positions for those who finish, the same for the grid as a reference, the “probability error” for the win and the podium (the average of the squared difference between the probability given and what happened: 0 is perfect, lower is better), what share of riders finished within their 80% interval, who the favourite was and whether they won, and what probability the actual winner had. Everything is walk-forward: each forecast was made only with data from before that race.

Keep in mind. A single race says little: the model should be judged over a whole season. To compare the position error fairly there is also a “corrected grid” (the grid adjusted using previous races, because the pole-sitter cannot gain places and the last rider cannot lose any); against this reference the improvement before the sprint is small, and after the sprint it is real.

Three paces per rider and GP (since 2018). Practice: estimated race pace from laps done on medium front and rear tyres that are already used (fresh rubber inflates the pace), not counting FP1 (unrepresentative time of day and track conditions); only laps no more than 0.20 s per kilometre above the same rider’s best lap in these conditions are kept (0.25 in Thailand and Malaysia), and the median is taken. Race and sprint: median of the actual clean laps, with a minimum of 4. The pace per km, the scatter, the sectors, the tyre age range of the laps used and, as context, the rider’s typical degradation on the medium in practice are also given.

Keep in mind. Practice pace is an estimate: we do not know how much fuel or which engine map each rider is running, which is why it is not used to forecast the race. Compare riders within the same GP; across circuits, use the pace per km with caution.

For each rider and race or sprint (since 2018) in which they did not pit, their clean laps are taken excluding the first 2 and the last 2 (where the fight at the start and the easing off at the end happen), with a minimum of 7. Since lap times change over the race (the tyre wears, the tank empties), first the straight line that best follows that trend is drawn and then how far the laps deviate from it is measured: the scatter around the line, as a % of the typical lap time. The lower, the more consistent. To compare circuits and years this % is divided by the median of all riders in that race: an index below 1 means more consistent than the field, above 1 less. For the season the median of the index across all their races is taken, and the ranking only includes riders with at least 5 races. Bike-swap races (flag-to-flag) do not count; races restarted after a red flag do, with the part that counts.

Keep in mind. It only measures consistency, not speed: a rider can be very consistent and slow. Riders who went into the pits are left out. The whole-career figure is the median index across all their races since 2018 (full races and sprints separately), using the same measure; the site only shows riders with 20 races or more, because with few races the figure depends heavily on luck.

Everything comes from the order of crossing the line on each lap (official lap-by-lap position sheet, since 2010). An overtake is a pair of riders who swap order between two consecutive laps (moving up because someone retires does not count). A battle is two riders running less than 1 second apart for at least 5 consecutive laps. Lead changes, different leaders, laps led by the winner, the closest finish (gap of 2nd at the line), the biggest comeback (from the official grid to the finishing position) and whether there was a red flag are also counted. Per rider: position on the first lap and at the finish, laps led, overtakes made and suffered, best and worst position.

Keep in mind. Only what happens at the finish line is seen: an overtake and a counter-overtake within the same lap do not show up. It is a measure of the race’s “movement”, not an official overtaking statistic.

In each race and sprint since 2018, a rider’s compound is the one that appears on most of their laps (front and rear). Per rider: the median of the clean laps, the % gap to the best pace of the session, the degradation (how many seconds per lap they slow down on average, calculated if they have at least 6 clean laps), the grid and the result; “minority” means their rear compound is not the one chosen by most of the field. Per compound: what % of the field chose it, the median pace and degradation, wins, podiums, average finishing position of those who finish and % of retirements (out of those who started on that compound). It helps answer “did the minority gamble pay off?”.

Keep in mind. Degradation mixes the tyre, the fuel being burned and traffic, so it only makes sense to compare riders within the same race, not across circuits. A compound chosen by 1 or 2 riders gives figures that depend heavily on who ran it.

For each session with lap data (since 2018) and each rider, two readings. Time attack: their best valid lap (neither cancelled nor a pit lap), with the tyre they were on, and the gap to the best in the session in seconds and in %. Race sim: the median of the clean laps from their long runs (at least 5 consecutive fast laps; in the race and sprint, the whole race), with a minimum of 4 laps, and the gap to the best long-run pace of the session. Top speed and the % of laps done on slicks (below 100 means the track was wet) are also added.

Keep in mind. If a rider has not done any long run, their race sim is left blank. Compare only within the same session: each team tests different things (fuel, tyres, maps) and a long run on old rubber is not comparable with one on fresh rubber.

Three paces per rider and GP (since 2018). Practice: estimated race pace from laps done on medium front and rear tyres that are already used (fresh rubber inflates the pace), not counting FP1 (unrepresentative time of day and track conditions); only laps no more than 0.20 s per kilometre above the same rider’s best lap in these conditions are kept (0.25 in Thailand and Malaysia), and the median is taken. Race and sprint: median of the actual clean laps, with a minimum of 4. The pace per km, the scatter, the sectors, the tyre age range of the laps used and, as context, the rider’s typical degradation on the medium in practice are also given.

Keep in mind. Practice pace is an estimate: we do not know how much fuel or which engine map each rider is running, which is why it is not used to forecast the race. Compare riders within the same GP; across circuits, use the pace per km with caution.

The same predictor as in Q1, applied to the 12 riders in Q2 (the top 10 from Practice and the 2 who come through from Q1): each rider’s average position among the Q2 riders in Practice and FP2, using valid laps; before 2023, the best lap across all of practice. The probabilities of pole, front row (top 3 in Q2) and top 6 come from history: of all riders predicted in that same position in previous Q2 sessions, how many took pole, the front row or a top 6 (with a minimum of 10 cases). It is published on Saturday after Q1 and, once Q2 ends, it stays next to the actual result.

Keep in mind. Since 2023, the favourite takes pole 49% of the time and the forecast captures 55% of the front row: qualifying is much more open than the race. The probability does not look at who the rider is, only at the position practice puts them in.

The starting point is the official grid, because it is the best predictor of the finishing order. From there the race is simulated 4,000 times, adding three sources of randomness drawn only from previous races (since 2015): how many places are gained or lost on the first lap, how much the order shuffles from the first lap to the flag, and the probability that each rider retires (their historical retirement rate, smoothed towards the field average as if we added 6 “average” races, and capped at 60%). The probabilities (win, podium, top 5, top 10, retirement) are the percentage of simulations in which it happens and already include the risk of retiring. The expected position and the 80% interval (the range of positions they finish in 8 times out of 10) are calculated only over the simulations in which the rider finishes. The post-sprint version (Saturday afternoon) replaces the starting point with a mix of half grid, half actual pace in the sprint (median of the laps excluding the first, cancelled laps and pit laps, with a minimum of 5 laps); the website always shows the most recent version.

Keep in mind. Before the sprint the model predicts the order just as well as the grid (over 213 races, 0.76 in both cases): its value is that the probabilities are honest (when it says 35%, it happens roughly 35% of the time); the post-sprint version is the only one that genuinely beats the grid. It is deliberately cautious: it never gives more than a 70% podium chance, and the 80% interval is right slightly more often than it claims (83%). It is not recalculated during the race.

Sector map

The lap is divided into the 4 official timing sectors. In each dry session (laps on slicks, since 2018), for each rider with at least 3 clean laps their best time in each sector is taken and compared with the best in the session for that sector: 0% is the fastest, 0.5% means half a percent slower. Then the median across all sessions is taken (at least 3 in total since 2018, at least 2 in a given season). To draw the sectors on the track layout, where each sector starts and ends is approximated from how the time is split: if sector 1 takes 25% of the lap, it covers the first 25% of the layout (based on clean race laps from the most recent season).

Keep in mind. A sector is a different physical stretch at each circuit: it tells you where a rider is fast, not whether they are good on the brakes or through corners. The drawn boundaries are approximate (the typical error is 5-10% of the lap) because they assume the same average speed in all sectors, until the official splits are loaded by hand.

Circuit layout

The track drawing comes from OpenStreetMap: of all the segments tagged as a race track, the closed circuit whose length is closest to the official one is taken, the direction of racing is deduced from the one-way tags (or from the number of right and left corners) and the finish line is placed on the pit lane. The elevation of each point comes from a smoothed digital terrain model (Copernicus, 90 m cells, via Open-Meteo), and enables the 3D relief. There are currently 25 circuits. © OpenStreetMap contributors (ODbL licence).

Keep in mind. The elevation is that of the terrain at a 90 m resolution, not measured on the asphalt: it is useful for seeing climbs and descents, not for giving exact gradients. The position of the sectors on the layout is approximate (see “Sector map”).

Everything comes from the order of crossing the line on each lap (official lap-by-lap position sheet, since 2010). An overtake is a pair of riders who swap order between two consecutive laps (moving up because someone retires does not count). A battle is two riders running less than 1 second apart for at least 5 consecutive laps. Lead changes, different leaders, laps led by the winner, the closest finish (gap of 2nd at the line), the biggest comeback (from the official grid to the finishing position) and whether there was a red flag are also counted. Per rider: position on the first lap and at the finish, laps led, overtakes made and suffered, best and worst position.

Keep in mind. Only what happens at the finish line is seen: an overtake and a counter-overtake within the same lap do not show up. It is a measure of the race’s “movement”, not an official overtaking statistic.

Three paces per rider and GP (since 2018). Practice: estimated race pace from laps done on medium front and rear tyres that are already used (fresh rubber inflates the pace), not counting FP1 (unrepresentative time of day and track conditions); only laps no more than 0.20 s per kilometre above the same rider’s best lap in these conditions are kept (0.25 in Thailand and Malaysia), and the median is taken. Race and sprint: median of the actual clean laps, with a minimum of 4. The pace per km, the scatter, the sectors, the tyre age range of the laps used and, as context, the rider’s typical degradation on the medium in practice are also given.

Keep in mind. Practice pace is an estimate: we do not know how much fuel or which engine map each rider is running, which is why it is not used to forecast the race. Compare riders within the same GP; across circuits, use the pace per km with caution.

In each race and sprint since 2018, a rider’s compound is the one that appears on most of their laps (front and rear). Per rider: the median of the clean laps, the % gap to the best pace of the session, the degradation (how many seconds per lap they slow down on average, calculated if they have at least 6 clean laps), the grid and the result; “minority” means their rear compound is not the one chosen by most of the field. Per compound: what % of the field chose it, the median pace and degradation, wins, podiums, average finishing position of those who finish and % of retirements (out of those who started on that compound). It helps answer “did the minority gamble pay off?”.

Keep in mind. Degradation mixes the tyre, the fuel being burned and traffic, so it only makes sense to compare riders within the same race, not across circuits. A compound chosen by 1 or 2 riders gives figures that depend heavily on who ran it.