Power ranking and predictions

A rating of every active rider, tested on races it never saw.

Power ranking

Bradley–Terry strength from every heat of the last 4 seasons (older heats count less: a heat's weight halves every 45 days), as of October 16, 2026. Riders with 25+ rides, 5 in the last 45 days.

#	Rider	Strength	Win vs 3 average riders	Elo	30 days
1	Robert Lambert	1.85	62.9 %	2018	+33
2	Bartosz Zmarzlik	1.80	61.8 %	1978	+15
3	Artem Laguta	1.59	56.9 %	1954	+29
4	Brady Kurtz	1.50	54.7 %	1925	+43
5	Emil Sajfutdinow	1.47	53.8 %	1936	-5
6	Anders Thomsen	1.32	50.0 %	1868	-11
7	Ryan Douglas	1.30	49.5 %	1845	-35
8	Mikkel Michelsen	1.24	48.2 %	1909	+13
9	Michael Jepsen Jensen	1.19	46.8 %	1857	-7
10	Jason Doyle	1.19	46.7 %	1859	+26
11	Dan Bewley	1.10	44.6 %	1879	+17
12	Patryk Dudek	1.05	43.3 %	1817	-34
13	Leon Madsen	1.03	42.8 %	1843	+4
14	Andzejs Lebedevs	1.02	42.5 %	1791	-32
15	Max Fricke	1.01	42.3 %	1807	+26
16	Kai Huckenbeck	0.86	38.8 %	1796	+19
17	Kacper Woryna	0.86	38.7 %	1786	+12
18	Przemysław Pawlicki	0.83	38.1 %	1738	0
19	Martin Vaculik	0.81	37.6 %	1782	+36
20	Mateusz Cierniak	0.81	37.6 %	1745	-21
21	Jakub Miśkowiak	0.80	37.4 %	1801	+30
22	Szymon Woźniak	0.78	36.8 %	1751	+11
23	Frederik Jakobsen	0.76	36.3 %	1728	+35
24	Tobiasz Musielak	0.74	35.9 %	1723	-23
25	Maksymilian Pawełczak	0.74	35.9 %	1708	-7
26	Wiktor Przyjemski	0.73	35.7 %	1749	+7
27	Krzysztof Buczkowski	0.70	35.0 %	1709	+4
28	Joshua Pickering	0.68	34.7 %	1688	-67
29	Mathias Pollestad	0.65	33.9 %	1759	+67
30	Matias Nielsen	0.64	33.6 %	1732	+46
31	Benjamin Cook	0.63	33.5 %	1711	-9
32	Maciej Janowski	0.61	33.0 %	1698	-17
33	Jan Kvech	0.58	32.4 %	1665	-15
34	Rohan Tungate	0.58	32.3 %	1700	+43
35	Fredrik Lindgren	0.57	32.2 %	1734	+27
36	Danny King	0.57	32.1 %	1675	+20
37	Dominik Kubera	0.57	32.1 %	1735	-6
38	Piotr Pawlicki	0.55	31.7 %	1700	-23
39	Mads Hansen	0.55	31.7 %	1719	+19
40	Janusz Kołodziej	0.52	31.1 %	1780	-21
41	Norick Blödorn	0.51	30.9 %	1678	+17
42	Marcin Nowak	0.51	30.8 %	1690	0
43	Jacob Thorssell	0.48	30.3 %	1682	-15
44	Luke Becker	0.48	30.2 %	1693	0
45	Adam Bubba Bednar	0.48	30.2 %	1684	+2
46	Nicolai Klindt	0.48	30.1 %	1677	+26
47	Andreas Lyager	0.47	30.0 %	1660	-16
48	Tim Sörensen	0.46	29.8 %	1676	0
49	Chris Harris	0.45	29.5 %	1656	+9
50	Jaimon Lidsey	0.40	28.5 %	1669	-12

Conditions

Track

Weather

Heat predictor

Pick four riders: their chance to win the heat from gates 1–4, from strength, gate and — once a track is picked — their record and experience there (meetings ridden), as far as the backtest kept them.

Gate	Rider	Win vs 3 average riders
1		33 %
2		28 %
3		20 %
4		19 %

Gate effect

Chance to beat an equal rider from gate 4.

Gate 155.4 %
Gate 252.2 %
Gate 349.7 %
Gate 450.0 %

How good are the predictions?

Trained on the oldest 70 % of 18912 heats, extra inputs chosen on the next 15 %, tested on the newest 15 % (from 2026-06-18): 18545 rider pairs, 3338 heats it never saw.

Model	Pairs ordered correctly	Heat winner picked	Log-loss
Coin flip	50.0 %	36.5 %	0.693
Average position points	63.4 %	41.9 %	0.655
Elo	65.5 %	44.2 %	0.618
Bradley–Terry + gates + recency	65.6 %	44.4 %	0.618
Bradley–Terry + extra inputs	67.0 %	46.0 %	0.606

Extra inputs

Each input is kept only if it lowers the log-loss on the held-out heats.

Input	Without	With	Kept
Form (last 20 rides)	0.6062	0.6005	yes
Track record	0.6005	0.5955	yes
Track experience (meetings there)	0.5955	0.5912	yes
Meetings in 7 days	0.5912	0.5911	no
Age curve	0.5912	0.5910	no
Rain × inside gate	0.5912	0.5912	no

Calibration

Predicted chance against how often it happened (test pairs).

2023–2026

Sources, definitions and methods