How the match predictions did on past seasons and on this one.
Method
PitStop predicts matches from OPR, the points a team adds to its alliance's score. OPR is solved from the scores of earlier matches, without penalty points. Each team's earlier events and last season give a starting estimate that the matches at the current event soon replace. The win chance comes from the predicted score gap.
Every match below was predicted using only matches played before it. Parameters were set on
2021 to 2023, the model was chosen on
2024 and 2025, and 2026 was not used for either.
Results by season
Model
Winner accuracy
Brier score
Log loss
Score error
2026, BIOBUZZ 6,463 matches. Not used for tuning
OPR, this event only, season mean fallback
68.2%
0.199
0.585
17.6
OPR, this event only, last event fallback
65.4%
0.204
0.595
17.2
Ridge OPR toward the season mean
68.8%
0.189
0.563
15.5
OPR with priors from earlier events (ridge / MAP OPR) Used on this site
73.4%
0.174
0.527
14.3
EPA, sequential updates, no reset between events
72.7%
0.176
0.532
14.3
EPA with a partial reset at each new event
72.8%
0.176
0.532
14.3
2025, DECODE 45,230 matches. Used to choose the model
OPR, this event only, season mean fallback
69.1%
0.198
0.581
27.0
OPR, this event only, last event fallback
70.4%
0.197
0.579
23.8
Ridge OPR toward the season mean
69.3%
0.189
0.559
24.9
OPR with priors from earlier events (ridge / MAP OPR) Used on this site
74.8%
0.168
0.510
21.2
EPA, sequential updates, no reset between events
74.5%
0.170
0.514
21.6
EPA with a partial reset at each new event
74.6%
0.169
0.513
21.4
2024, INTO THE DEEP 39,799 matches. Used to choose the model
OPR, this event only, season mean fallback
70.6%
0.191
0.563
34.6
OPR, this event only, last event fallback
71.9%
0.191
0.561
31.0
Ridge OPR toward the season mean
70.9%
0.181
0.540
32.3
OPR with priors from earlier events (ridge / MAP OPR) Used on this site
76.3%
0.159
0.486
27.8
EPA, sequential updates, no reset between events
76.1%
0.161
0.491
28.0
EPA with a partial reset at each new event
76.1%
0.160
0.489
27.7
2023, CENTERSTAGE 32,809 matches. Used to set parameters
OPR, this event only, season mean fallback
67.8%
0.202
0.592
29.1
OPR, this event only, last event fallback
69.1%
0.201
0.588
25.5
Ridge OPR toward the season mean
68.1%
0.194
0.572
26.1
OPR with priors from earlier events (ridge / MAP OPR) Used on this site
74.0%
0.170
0.516
22.6
EPA, sequential updates, no reset between events
73.8%
0.172
0.520
22.6
EPA with a partial reset at each new event
73.8%
0.172
0.519
22.3
2022, POWERPLAY 28,186 matches. Used to set parameters
OPR, this event only, season mean fallback
68.2%
0.202
0.591
29.8
OPR, this event only, last event fallback
69.4%
0.202
0.589
26.4
Ridge OPR toward the season mean
68.5%
0.193
0.570
26.5
OPR with priors from earlier events (ridge / MAP OPR) Used on this site
74.7%
0.168
0.509
23.2
EPA, sequential updates, no reset between events
74.4%
0.170
0.514
23.2
EPA with a partial reset at each new event
74.3%
0.170
0.513
23.0
2021, Freight Frenzy 19,601 matches. Used to set parameters
OPR, this event only, season mean fallback
66.3%
0.212
0.614
34.9
OPR, this event only, last event fallback
67.3%
0.210
0.611
31.1
Ridge OPR toward the season mean
66.8%
0.203
0.593
30.9
OPR with priors from earlier events (ridge / MAP OPR) Used on this site
71.5%
0.183
0.547
28.1
EPA, sequential updates, no reset between events
71.3%
0.184
0.551
27.7
EPA with a partial reset at each new event
71.2%
0.184
0.549
27.5
Winner accuracy
How often the alliance with the higher win chance won. Ties are left out.
Brier score
The average squared gap between the win chance and the result, where a win is 1, a loss is 0
and a tie is 0.5. Lower is better. Always predicting 50% scores 0.250.
Log loss
Like the Brier score, but it punishes confident misses more. Lower is better. Always
predicting 50% scores 0.693.
Score error
The average gap in points between the predicted and real score of one alliance, without
penalty points. Lower is better.
Win chance calibration
Matches are grouped by the predicted chance that red wins. Each point shows how often red won
in that group, for the model used on this site. Points on the dashed line mean the predicted
chance matched the result. Larger points have more matches.
2024 and 2025
2026
Predicted equals actual
2024 and 2025
2026
Predicted range
Predicted
Won
Matches
Predicted
Won
Matches
0% to 10%
5.5%
5.3%
6,722
5.5%
4.9%
399
10% to 20%
15.0%
14.3%
8,006
15.0%
13.3%
545
20% to 30%
25.0%
24.2%
7,987
25.1%
23.6%
639
30% to 40%
35.0%
34.1%
8,467
35.0%
34.0%
759
40% to 50%
45.1%
45.5%
8,810
44.9%
46.1%
784
50% to 60%
55.0%
56.0%
9,060
55.0%
56.6%
858
60% to 70%
65.0%
66.9%
8,830
64.8%
65.5%
740
70% to 80%
75.1%
76.8%
9,201
75.0%
77.3%
654
80% to 90%
85.0%
87.2%
9,406
84.6%
86.1%
601
90% to 100%
94.6%
95.3%
8,540
94.7%
94.6%
484
Ranking point chances
Chances that an alliance earns each ranking point, checked against 2026 qualification
results.
Ranking point
Earned in
Brier score
Brier score from the base rate alone
Log loss
Swarm, leave and park 16+
62.8%
0.192
0.234
0.564
Pollinator 1, 4+ tips
7.4%
0.036
0.068
0.128
Pollinator 2, 7+ tips
1.0%
0.007
0.010
0.026
Swarm, leave and park 16+
2026
Predicted equals actual
Swarm, leave and park 16+: predicted chance against how often it was earned
Series
Predicted
Actual
Matches
2026
8.2%
4.6%
22
2026
16.5%
9.7%
309
2026
25.0%
23.4%
768
2026
35.0%
38.0%
966
2026
45.3%
50.1%
1,223
2026
54.9%
62.5%
1,284
2026
65.5%
56.4%
2,713
2026
75.1%
77.5%
2,110
2026
85.1%
87.1%
2,080
2026
92.2%
92.4%
811
Pollinator 1, 4+ tips
2026
Predicted equals actual
Pollinator 1, 4+ tips: predicted chance against how often it was earned
Series
Predicted
Actual
Matches
2026
1.3%
1.3%
10,313
2026
14.4%
12.3%
652
2026
24.5%
22.7%
322
2026
34.7%
37.0%
219
2026
45.1%
43.2%
185
2026
55.4%
55.6%
133
2026
65.1%
75.6%
123
2026
75.0%
78.1%
105
2026
84.9%
85.1%
107
2026
95.1%
93.7%
127
Pollinator 2, 7+ tips
2026
Predicted equals actual
Pollinator 2, 7+ tips: predicted chance against how often it was earned
Series
Predicted
Actual
Matches
2026
0.3%
0.3%
11,986
2026
14.2%
11.5%
131
2026
24.0%
28.4%
67
2026
34.5%
30.6%
49
2026
44.1%
57.1%
21
2026
55.0%
54.6%
11
2026
61.3%
33.3%
3
2026
74.9%
80.0%
10
2026
85.4%
80.0%
5
2026
92.2%
100.0%
3
Queue time estimates
Each estimate is made when a match starts and predicts when a later match will start. The
columns show how many matches later that match is.