Football analytics · flagship

Live repoOpen data (StatsBomb)

Opponent-Adjusted Football Metrics

Independent research · 8 months of active development

What this is

Contextual expected goals (CxG) and assists (CxA) over 2,143,146 raw StatsBomb events, benchmarked directly against StatsBomb's own xG on the same shots rather than self-reported in isolation.

Metrics

CxG model

0.80920 AUC

vs StatsBomb xG 0.81957

CxG calibration

Brier 0.07325 · log loss 0.26110

CxA baseline

0.85814 AUC · Brier 0.04046

over 1,091,388 rows

Testing

332 tests across 46 files

Stack

scikit-learnFastAPIPostgreSQLSQLAlchemyAlembicCI: lint+type+test

Provenance

Computed outputs are gitignored; the numbers above live in committed docs, not rendered dashboards.

What I would not claim

LightGBM as headline tech
gradient boosting as headline tech
plotly

Explore the shots

A sample of real shots, my CxG beside StatsBomb's own xG on the same shots. Toggle the metric or the diff, filter by team, match or pressure, and watch the calibration recompute.

Larger, brighter dot = higher CxG White ring = goal

440 shots

59 goals in this selection

Mean CxG0.1420

Mean StatsBomb xG0.1538

Brier score on this selection (lower is better calibrated)

CxG0.0776

StatsBomb xG0.0792

Actual goal rate 0.134. Both are shot-quality estimates; the point is that CxG sits next to StatsBomb's own number on the same shots, not in isolation.

Data provenance. 440-shot sample from the CxG diagnostic model, benchmarked against StatsBomb xG. Open StatsBomb data, served from Supabase; full set in the repo.

Explore contextual expected assists

The CxA diagnostic model against its own baseline over 1,091,388 action rows. Precision among the model's most confident actions is the headline; the leaderboard shows volume against efficiency.

Diagnostic v1 against its own baseline

Over 1,091,388 action rows. There is no off-the-shelf expected-assists model to compare against, so this is an internal baseline, not an industry incumbent.

provisionally promoted

Diagnostic v10.5860

Own baseline0.4627

+0.1233 better than the baseline. Higher is better for this metric.

The headline. Among the actions the model is most confident about, the share that actually created a shot rose from 0.463 to 0.586 — 12.3 percentage points.

scikit-learn GradientBoostingClassifier, sigmoid-calibrated · 37 selected features · promotion gate passed

Volume and efficiency tell different stories: sort by total and by mean to see it. Özil leads on total CxA, but Sánchez creates more per action, and Fàbregas has the most actions of the leaders with the lowest mean.

Mesut Özil

Arsenal

416.09

Total CxA

Total 416.09Mean 0.08914,672 actions417 shot-creating

Alexis Alejandro Sánchez Sánchez

Arsenal

351.93

Total CxA

Total 351.93Mean 0.10703,288 actions324 shot-creating

Willian Borges da Silva

Chelsea

325.34

Total CxA

Total 325.34Mean 0.09763,333 actions277 shot-creating

Francesc Fàbregas i Soler

Chelsea

324.86

Total CxA

Total 324.86Mean 0.06315,150 actions283 shot-creating

Ross Barkley

Everton

316.04

Total CxA

Total 316.04Mean 0.08053,924 actions322 shot-creating

Christian Dannemann Eriksen

Tottenham Hotspur

311.67

Total CxA

Total 311.67Mean 0.08423,703 actions402 shot-creating

Dimitri Payet

West Ham United

311.06

Total CxA

Total 311.06Mean 0.10133,071 actions364 shot-creating

Eden Hazard

Chelsea

296.40

Total CxA

Total 296.40Mean 0.09703,055 actions262 shot-creating

Riyad Mahrez

Leicester City

273.62

Total CxA

Total 273.62Mean 0.10232,674 actions341 shot-creating

Juan Manuel Mata García

Manchester United

#10

268.31

Total CxA

Total 268.31Mean 0.07413,621 actions199 shot-creating

Aaron Ramsey

Arsenal

#11

266.26

Total CxA

Total 266.26Mean 0.06304,223 actions233 shot-creating

Anthony Martial

Manchester United

#12

258.48

Total CxA

Total 258.48Mean 0.11402,267 actions220 shot-creating

Players by diagnostic CxA
Rank	Name	Total CxA	Mean CxA	Actions	Shot-creating
1	Mesut Özil	416.09	0.0891	4,672	417
2	Alexis Alejandro Sánchez Sánchez	351.93	0.1070	3,288	324
3	Willian Borges da Silva	325.34	0.0976	3,333	277
4	Francesc Fàbregas i Soler	324.86	0.0631	5,150	283
5	Ross Barkley	316.04	0.0805	3,924	322
6	Christian Dannemann Eriksen	311.67	0.0842	3,703	402
7	Dimitri Payet	311.06	0.1013	3,071	364
8	Eden Hazard	296.40	0.0970	3,055	262
9	Riyad Mahrez	273.62	0.1023	2,674	341
10	Juan Manuel Mata García	268.31	0.0741	3,621	199
11	Aaron Ramsey	266.26	0.0630	4,223	233
12	Anthony Martial	258.48	0.1140	2,267	220
13	Sergio Leonel Agüero del Castillo	257.14	0.1428	1,801	215
14	Kevin De Bruyne	256.32	0.1028	2,494	283
15	David Josué Jiménez Silva	255.46	0.0900	2,839	231
16	Dušan Tadić	251.42	0.1168	2,153	286
17	Romelu Lukaku Menama	249.28	0.1109	2,247	226
18	Philippe Coutinho Correia	247.98	0.1075	2,306	288
19	Sadio Mané	245.73	0.1025	2,397	251
20	Jesús Navas González	243.40	0.0857	2,840	217
21	Harry Kane	234.90	0.1105	2,125	312
22	Gylfi Þór Sigurðsson	231.15	0.0949	2,436	279
23	Raheem Sterling	226.48	0.1127	2,010	194
24	Erik Lamela	221.96	0.1002	2,215	231
25	Marko Arnautović	218.77	0.0876	2,497	221
26	Matt Ritchie	214.77	0.0699	3,072	226
27	Bamidele Alli	211.71	0.0911	2,325	227
28	Gnégnéri Yaya Touré	209.48	0.0602	3,477	222
29	James Philip Milner	209.12	0.0745	2,808	238
30	Odion Jude Ighalo	208.70	0.1299	1,606	232
31	Moussa Sissoko	208.13	0.0662	3,142	278
32	Jason Puncheon	207.57	0.0764	2,717	202
33	Kylian Mbappé Lottin	206.52	0.1226	1,684	214
34	Diego da Silva Costa	203.75	0.1125	1,811	157
35	Nathan Redmond	203.69	0.0915	2,226	227
36	Robert Brady	199.07	0.0707	2,816	196
37	Troy Deeney	196.49	0.0903	2,176	225
38	Wayne Mark Rooney	194.75	0.0839	2,322	188
39	Pedro Eliezer Rodríguez Ledesma	192.79	0.0855	2,254	209
40	Marc Albrighton	191.34	0.0818	2,338	246
41	Wes Hoolahan	190.04	0.0804	2,364	181
42	Steven Davis	188.19	0.0666	2,827	190
43	Jamie Vardy	184.49	0.1355	1,362	221
44	Wilfried Zaha	182.91	0.0955	1,916	185
45	Oscar dos Santos Emboaba Júnior	179.04	0.0831	2,155	169
46	Roberto Firmino Barbosa de Oliveira	177.24	0.0868	2,043	209
47	Héctor Bellerín Moruno	175.06	0.0467	3,747	142
48	Adam David Lallana	173.32	0.0871	1,990	216
49	Memphis Depay	167.53	0.1048	1,599	119
50	Charlie Daniels	162.56	0.0420	3,871	178
51	Ignacio Monreal Eraso	162.55	0.0435	3,738	136
52	Jordan Ayew	162.55	0.0824	1,972	136
53	Santiago Cazorla González	161.96	0.0665	2,435	167
54	Manuel Lanzini	161.71	0.0695	2,327	202
55	Georginio Wijnaldum	161.44	0.0626	2,578	202
56	Aleksandar Kolarov	159.26	0.0524	3,041	128
57	Nathaniel Edwin Clyne	158.76	0.0479	3,316	177
58	Xherdan Shaqiri	157.44	0.0889	1,771	186
59	Alberto Moreno Pérez	154.56	0.0545	2,835	192
60	Gerard Deulofeu Lázaro	153.70	0.1031	1,491	161
61	Idrissa Gana Gueye	152.02	0.0418	3,635	145
62	Danny Drinkwater	151.71	0.0430	3,527	177
63	Luka Modrić	150.76	0.0513	2,941	149
64	Lionel Andrés Messi Cuccittini	150.59	0.1177	1,279	149
65	Fernando Luiz Roza	149.55	0.0431	3,472	151
66	Kevin De Bruyne	147.73	0.0864	1,710	193
67	Graziano Pellè	145.18	0.0950	1,528	133
68	André Ayew Pelé	144.54	0.0649	2,228	134
69	Yannick Bolasie Yala	143.80	0.0942	1,527	172
70	Jonathan Howson	143.03	0.0590	2,426	161
71	Junior Stanislas	142.71	0.0932	1,532	145
72	Darren Fletcher	142.60	0.0509	2,801	116
73	Antoine Griezmann	141.90	0.0849	1,672	148
74	Daryl Janmaat	140.93	0.0495	2,845	167
75	Ashley Westwood	140.33	0.0487	2,880	121
76	Emre Can	139.97	0.0454	3,084	181
77	Yohan Cabaye	139.58	0.0570	2,448	166
78	Jermain Defoe	139.43	0.1383	1,008	154
79	Yann Gérard M'Vila	138.91	0.0439	3,167	210
80	Olivier Giroud	138.08	0.0964	1,433	129
81	José Manuel Jurado Marín	137.54	0.0705	1,950	146
82	Aaron Cresswell	136.19	0.0426	3,200	184
83	Mark Noble	134.45	0.0389	3,460	189
84	Bojan Krkíc Pérez	133.81	0.0805	1,663	149
85	Ayoze Pérez Gutiérrez	132.48	0.0858	1,544	149
86	Patrick van Aanholt	131.73	0.0536	2,458	150
87	Ryan Bertrand	131.61	0.0506	2,601	112
88	César Azpilicueta Tanco	131.22	0.0372	3,525	99
89	Séamus Coleman	130.35	0.0498	2,619	106
90	Joshua King	129.74	0.1118	1,160	137
91	Andrew Surman	128.96	0.0332	3,879	149
92	Joshua Kimmich	128.46	0.0590	2,177	147
93	Simon Francis	128.33	0.0313	4,098	134
94	Toni Kroos	128.17	0.0559	2,292	129
95	James McClean	128.13	0.0827	1,550	123
96	Dan Gosling	127.83	0.0553	2,312	131
97	Mousa Sidi Yaya Dembélé	127.70	0.0457	2,796	154
98	N'Golo Kanté	127.37	0.0474	2,688	138
99	Daniel Olmo Carvajal	126.87	0.1021	1,243	110
100	Stéphane Sessègnon	125.99	0.0873	1,444	148

What drives the model

end_x0.0896

end_y0.0050

end_zone0.0045

is_through_ball0.0033

start_x0.0031

is_pass0.0031

length0.0017

x_progression0.0014

pass_height0.0010

seconds_since_possession_start0.0010

Progression and location dominate, led by end_x. That is an honest description of what the model leans on, not a claim of subtle contextual insight.

Data provenance. Generated into data/cxa.json from the committed CxA portfolio export (headline metrics, top players and teams, feature drivers). Open StatsBomb data. The comparison is against this project's own baseline, not an industry expected-assists incumbent, and the model is provisionally promoted.