Skip to content
Reference Lap

2026 performance vs model

Each driver's and team's results vs model expectation over the season's grands prix (sprints are not included). Race gain is the expected finish before the race minus the actual finish, averaged over races: positive means better than the model expected from the grid. Quali gain is how much qualifying improved the expected finish compared with the forecast made before qualifying.

A persistent gain can mean the driver outperformed, or that the model under-rates them: these numbers cannot tell the two apart on their own. The SE is the standard error of the mean.

20 rounds analysed: 20 rounds from the held-out backtest.Backtest · held-outReplayed after the fact on the held-out 2026 test set (rounds 1–20), using only data available before the session. Out of sample: never used for training or tuning.

Drivers

Drivers sorted by mean race gain (best first). Gains in places per race; positive = better than the model expected. Exp. pts: expected points before qualifying, summed over the races; Pts vs exp.: points minus that.
#DriverRace gain ± SEInterval chart (values in the race gain column)Quali gainRacesClassifiedDNFsPtsExp. ptsPts vs exp.
1HIRR. Hirakawa† fewer than 5 races5.62 places better than the model expected0.63 places worse than the model expected11—10.2+0.8
2LAWL. Lawson2.54 places better than the model expected ± 0.810.19 places better than the model expected20191 unk8232.8+49.2
3ALOF. Alonso1.87 places better than the model expected ± 0.900.02 places better than the model expected20137 unk70.5+6.5
4HADI. Hadjar1.19 places better than the model expected ± 1.260.05 places worse than the model expected17152 unk12285.5+36.5
5PERS. Pérez1.15 places better than the model expected ± 0.580.02 places better than the model expected20155 unk00.5−0.5
6OCOE. Ocon0.94 places better than the model expected ± 0.630.05 places worse than the model expected18171 unk76.2+0.8
7ANTA. Antonelli0.83 places better than the model expected ± 1.010.05 places better than the model expected2020—359312.8+46.2
8TSUY. Tsunoda† fewer than 5 races0.68 places better than the model expected ± 1.200.74 places worse than the model expected33—13.8−2.8
9SAIC. Sainz0.51 places better than the model expected ± 0.770.07 places worse than the model expected20182 unk74.8+2.2
10STRL. Stroll0.44 places better than the model expected ± 0.650.01 places worse than the model expected20614 unk00.4−0.4
11HAML. Hamilton0.38 places better than the model expected ± 0.950.25 places better than the model expected20191 unk234219.1+14.9
12LINA. Lindblad0.33 places better than the model expected ± 0.91as the model expected20181 dns · 1 unk3730.9+6.1
13COLF. Colapinto0.18 places better than the model expected ± 0.960.07 places better than the model expected20173 unk2713.2+13.8
14BOTV. Bottas0.05 places worse than the model expected ± 0.550.01 places better than the model expected20101 dns · 9 unk00.3−0.3
15GASP. Gasly0.09 places worse than the model expected ± 1.040.31 places better than the model expected20182 unk4031.7+8.3
16VERM. Verstappen0.24 places worse than the model expected ± 1.580.08 places worse than the model expected20173 unk259232.3+26.7
17BORG. Bortoleto0.41 places worse than the model expected ± 0.870.22 places worse than the model expected20181 dns · 1 unk1422.2−8.2
18BEAO. Bearman0.83 places worse than the model expected ± 1.010.04 places worse than the model expected20155 unk199.1+9.9
19HULN. Hülkenberg0.87 places worse than the model expected ± 0.990.09 places worse than the model expected20161 dns · 3 unk1227.5−15.5
20ALBA. Albon1.02 places worse than the model expected ± 0.82as the model expected20141 dns · 5 unk53.9+1.1
21LECC. Leclerc1.11 places worse than the model expected ± 1.260.15 places better than the model expected20182 unk215258.0−43.0
22NORL. Norris1.23 places worse than the model expected ± 1.400.11 places better than the model expected20161 dns · 3 unk213215.5−2.5
23RUSG. Russell2.11 places worse than the model expected ± 1.410.17 places worse than the model expected20182 unk228308.8−80.8
24PIAO. Piastri2.27 places worse than the model expected ± 1.320.26 places worse than the model expected20172 dns · 1 unk131199.6−68.6
25CAMR. Camara† fewer than 5 races4.59 places worse than the model expected0.43 places worse than the model expected11—00.3−0.3

Interval chart: dot = mean race gain, bar = ±2 SE, vertical line = 0 (as the model expected); further right is better. DNFs: inc = incident, mech = mechanical, dsq = disqualified, dns = did not start, unk = unknown (most retirements in the source data are recorded only as “Retired”). Points are compared with the expected points from the forecast made before qualifying.

† Greyed rows have fewer than 5 races: their means are too noisy to read and they are not part of the model check.

Teams

Teams sorted by mean race gain (best first), over every race start by the team's cars. Exp. pts: expected points before qualifying, summed over the races; Pts vs exp.: points minus that.
#TeamRace gain ± SEInterval chart (values in the race gain column)Quali gainRacesClassifiedDNFsPtsExp. ptsPts vs exp.
1RB F1 Team1.42 places better than the model expected ± 0.620.07 places better than the model expected40371 dns · 2 unk10655.3+50.7
2Aston Martin1.16 places better than the model expected ± 0.560.01 places better than the model expected401921 unk70.9+6.1
3Cadillac F1 Team0.55 places better than the model expected ± 0.410.02 places better than the model expected40251 dns · 14 unk00.7−0.7
4Red Bull0.45 places better than the model expected ± 0.960.09 places worse than the model expected40355 unk395330.1+64.9
5Alpine F1 Team0.05 places better than the model expected ± 0.700.19 places better than the model expected40355 unk6745.0+22.0
6Haas F1 Team0.03 places better than the model expected ± 0.610.07 places worse than the model expected40346 unk2715.8+11.2
7Williams0.26 places worse than the model expected ± 0.570.03 places worse than the model expected40321 dns · 7 unk128.7+3.3
8Ferrari0.37 places worse than the model expected ± 0.790.20 places better than the model expected40373 unk449477.1−28.1
9Audi0.64 places worse than the model expected ± 0.650.15 places worse than the model expected40342 dns · 4 unk2649.7−23.7
10Mercedes0.64 places worse than the model expected ± 0.890.06 places worse than the model expected40382 unk587621.7−34.7
11McLaren1.75 places worse than the model expected ± 0.950.07 places worse than the model expected40333 dns · 4 unk344415.1−71.1

Interval chart: dot = mean race gain, bar = ±2 SE, vertical line = 0 (as the model expected); further right is better. DNFs: inc = incident, mech = mechanical, dsq = disqualified, dns = did not start, unk = unknown (most retirements in the source data are recorded only as “Retired”). Points are compared with the expected points from the forecast made before qualifying.