ICE Hockey League

The model has called 65% of 71 2026/27 games right · always picking the home side: 55%

Sat 24 Oct

Sun 25 Oct

Tue 27 Oct

Wed 28 Oct

Fri 30 Oct

Sat 31 Oct

Sun 1 Nov

Wed 11 Nov

Fri 13 Nov

Sat 14 Nov

Sun 15 Nov

Win chances include overtime and shootouts. Hover the bar for the regulation split; click a game for its full preview.

Predicted vs actual71 finished games

Predicted vs actual 71

46 hit · 25 missed · 3 exact
Winner called right64.8%46 of 71 games
Exact scoreline34.2% of games
Mean goal error±1.53on the total goals in a game
Mean prediction49.5%home actually won 54.9%
Every finished game with the prediction it had beforehand, sortable by any column
23 Oct 2026, 19:45MHCvsKAC37%2.9 – 3.82:32:3Hit★-1.6
23 Oct 2026, 19:30PIVvsHCB11%1.8 – 5.11:53:6Hit+2.1
23 Oct 2026, 19:15OLLvsAVS88%4.8 – 1.84:12:3Miss-1.6
23 Oct 2026, 19:15HCIvsRBS37%2.5 – 3.32:32:4Hit+0.2
23 Oct 2026, 18:30G99vsBWL52%2.9 – 2.82:22:5Miss+1.3
21 Oct 2026, 19:45HCBvsHCI84%4.5 – 2.04:16:0Hit-0.5
21 Oct 2026, 19:45HCPvsPIV85%4.5 – 1.94:13:1Hit-2.3
21 Oct 2026, 19:15BWLvsVIC61%3.3 – 2.63:24:1Hit-0.9
21 Oct 2026, 19:15RBSvsOLL32%3.0 – 4.33:41:0Miss-6.4
21 Oct 2026, 19:15KACvsAVS83%4.1 – 1.84:16:3Hit+3.1
21 Oct 2026, 19:15VSVvsFTC54%3.2 – 2.93:24:1Hit-1.1
21 Oct 2026, 18:30G99vsMHC59%3.0 – 2.52:25:3Hit+2.6
18 Oct 2026, 18:00HCBvsBWL52%3.4 – 3.23:35:0Hit-1.6
18 Oct 2026, 17:30HCIvsOLL21%2.4 – 4.52:42:5Hit+0.0
18 Oct 2026, 17:00MHCvsAVS67%3.2 – 2.23:23:0Hit-2.5
18 Oct 2026, 15:30FTCvsKAC26%2.6 – 4.42:43:2 OTMiss-2.0
18 Oct 2026, 13:30VSVvsHCP39%2.5 – 3.22:32:3 OTHit★-0.7
17 Oct 2026, 19:15VICvsPIV77%3.8 – 2.03:24:2Hit+0.1
17 Oct 2026, 17:00MHCvsAVS69%3.2 – 2.13:23:4 OTMiss+1.7
16 Oct 2026, 19:45HCPvsHCI73%3.6 – 2.13:22:3 OTMiss-0.7
16 Oct 2026, 19:15OLLvsHCB53%3.6 – 3.43:35:4Hit+2.0
16 Oct 2026, 19:15RBSvsBWL47%3.2 – 3.43:30:6Hit-0.5
16 Oct 2026, 19:15KACvsVIC70%3.8 – 2.43:22:3Miss-1.2
16 Oct 2026, 19:15VSVvsG9945%2.5 – 2.82:22:4Hit+0.7
16 Oct 2026, 18:30FTCvsPIV72%3.9 – 2.53:24:3Hit+0.6
11 Oct 2026, 17:30HCIvsVSV39%2.6 – 3.22:33:2 OTMiss-0.8
11 Oct 2026, 17:30KACvsRBS63%3.8 – 2.93:25:3Hit+1.4
11 Oct 2026, 17:00MHCvsHCB34%2.6 – 3.62:33:4 OTHit+0.8
11 Oct 2026, 17:00PIVvsBWL20%2.3 – 4.52:41:4Hit-1.8
11 Oct 2026, 16:00OLLvsFTC77%4.9 – 2.84:25:1Hit-1.7
11 Oct 2026, 16:00AVSvsG9930%1.8 – 2.91:22:3 OTHit+0.3
11 Oct 2026, 16:00VICvsHCP41%2.5 – 3.12:33:2Miss-0.6
09 Oct 2026, 19:45MHCvsHCI75%3.9 – 2.23:21:4Miss-1.1
09 Oct 2026, 19:30PIVvsKAC13%1.9 – 4.81:44:5 SOHit+2.3
09 Oct 2026, 19:15BWLvsHCP42%2.8 – 3.32:35:2Miss+0.9
09 Oct 2026, 19:15AVSvsRBS37%2.2 – 3.02:31:6Hit+1.8
09 Oct 2026, 19:15VSVvsHCB30%2.4 – 3.72:34:3 OTMiss+0.9
09 Oct 2026, 18:30G99vsOLL44%2.9 – 3.32:33:4 OTHit+0.8
09 Oct 2026, 18:30FTCvsVIC46%3.0 – 3.22:33:2Miss-1.2
04 Oct 2026, 17:30BWLvsKAC28%2.6 – 4.12:46:1Miss+0.4
04 Oct 2026, 17:30VSVvsVIC52%2.8 – 2.72:23:4 OTMiss+1.5
04 Oct 2026, 17:00HCPvsAVS80%3.7 – 1.73:13:2 OTHit-0.4
04 Oct 2026, 16:00OLLvsPIV82%4.9 – 2.44:28:0Hit+0.7
04 Oct 2026, 16:00HCBvsFTC79%4.4 – 2.34:24:3 SOHit+0.3
04 Oct 2026, 15:00RBSvsMHC42%2.8 – 3.32:35:2Miss+0.8
03 Oct 2026, 17:30G99vsHCI68%3.4 – 2.23:23:0Hit-2.6
02 Oct 2026, 19:45HCBvsKAC33%2.6 – 3.82:34:0Miss-2.4
02 Oct 2026, 19:45HCPvsOLL48%3.4 – 3.53:34:2Miss-0.9
02 Oct 2026, 19:30PIVvsRBS32%2.5 – 3.62:31:4Hit-1.1
02 Oct 2026, 19:15VICvsHCI52%3.0 – 2.93:24:0Hit-1.9

Showing 50 of 71 rows. Sort or filter to narrow the list.

Each row's prediction was built from the games played before that date only — nothing after it. Sort by Goals Δ or Call to find where the model went wrong. A ★ marks a game whose exact scoreline it also got right.

Biggest upsets

  • OLLvsAVS

    23 Oct 2026, 19:15 · predicted 88% home

    2:3away won
  • VSVvsMHC

    02 Oct 2026, 19:15 · predicted 23% home

    7:2home won
  • MHCvsHCI

    09 Oct 2026, 19:45 · predicted 75% home

    1:4away won
  • FTCvsKAC

    18 Oct 2026, 15:30 · predicted 26% home

    3:2 OThome won
  • HCPvsHCI

    16 Oct 2026, 19:45 · predicted 73% home

    2:3 OTaway won

Games where the model was most confident in the side that lost.

Best calls

  • PIVvsHCB

    23 Oct 2026, 19:30 · predicted 11% home

    3:6as predicted
  • PIVvsKAC

    09 Oct 2026, 19:30 · predicted 13% home

    4:5 SOas predicted
  • HCPvsPIV

    21 Oct 2026, 19:45 · predicted 85% home

    3:1as predicted
  • HCBvsHCI

    21 Oct 2026, 19:45 · predicted 84% home

    6:0as predicted
  • HCIvsHCB

    27 Sept 2026, 17:30 · predicted 17% home

    3:4as predicted

Confident calls that came in. Note these are cheap to collect — a good model should have many of them, which is why the upsets matter more.

Model vs betting market52 games

Model vs the betting market

52 games with closing odds
Model and market scored on the same games
MeasureModelMarketDifference
Winner called right65.4%67.3%−1.9 pp
Brier score0.2160.206−0.010
Log loss0.6210.601−0.020
Market edge0.010Brier points, on the same games
Flat-stake return-3.7%one unit on the model's pick, every game
Bookmaker margin9.8%mean overround in the prices
They disagree21.2%model right 45.5% of those

Both are scored on the same 52 games, using the same overtime coin flip, so the comparison is like-for-like. Odds are de-vigged by normalising the three reciprocals to 1. The market is better calibrated, which is the expected result — closing odds absorb injuries, line-ups and money this model never sees.

The flat-stake figure is a sanity check, not a strategy: the three-way price is on regulation time, so any game decided in overtime or a shootout loses the stake. Treat a positive number on a few hundred games as luck until it survives several seasons.

Accuracy and calibrationBrier 0.227

How well does this model do?

71 games scored

Only 71 games have been scored so far this season — far too few to judge a model on. Switch to a finished season for a stable read.

Winner called right64.8%home-only baseline 54.9%
Brier score0.227baseline 0.253 · lower is better
Log loss0.646baseline 0.700
Beats the baseline by0.026Brier points of skill
Total-goals error±1.53mean absolute error per game
Exact scoreline4.2%hit rate on the likeliest score
Accuracy as the season went onCumulative share of games called right, against picking the home side every time
  • Home-only baseline
  • Model
02550751004122028364452606864.8%
Table view
Games scoredHome-only baselineModel
450.0%75.0%
837.5%75.0%
1250.0%83.3%
1637.5%81.3%
2045.0%70.0%
2450.0%66.7%
2857.1%64.3%
3259.4%62.5%
3658.3%61.1%
4057.5%57.5%
4454.5%61.4%
4856.3%62.5%
5253.8%61.5%
5653.6%60.7%
6055.0%63.3%
6457.8%64.1%
6857.4%64.7%
7154.9%64.8%

Both lines are cumulative, so they settle as the sample grows. The gap between them is the model's edge over the naive pick.

CalibrationWhen the model says 70%, does the home side win 70% of the time?
  • Predicted
  • Actual
025507510020–3030–4040–5050–6060–7070–8080–9080%
Table view
Predicted home win %PredictedActualGames
20–3026%57%7
30–4036%40%15
40–5044%46%13
50–6055%73%11
60–7065%75%8
70–8075%63%8
80–9084%80%5

The two lines tracking each other means the probabilities can be taken at face value. Where "actual" sits above "predicted", the model is under-confident in the home side; below, it is over-confident. Check the game count before reading much into a bucket — the extremes hold far fewer games than the middle. Buckets with fewer than five games are left out.

Every game is predicted using only the games played before it — a walk-forward test, so these are the numbers the model would really have produced. The baseline is "always back the home side", priced at the home-win rate known at the time. Team strengths start from the 2025/26 final table.

How the prediction works
  • Each team gets an attack and a defence multiplier from the goals it has scored and conceded, measured against the league average.
  • A fixture's expected goals are the league's home (or away) scoring rate times the attacking team's attack times the defending team's defence — so home advantage comes out of fitting home and away separately, not from a bonus added on top.
  • Both sides' goals are then treated as independent Poisson draws, which gives a probability for every scoreline rather than one guess, and from that the win, overtime and exact-score numbers.
  • Rates are shrunk toward the prior with six games of weight, so an early 7-1 win does not make a team look unbeatable. This is why fixtures are marked low confidence until both teams have played a few games.
  • Overtime and shootouts are treated as a coin flip. The feeds carry nothing that reliably predicts a shootout winner, and pretending otherwise would just be noise.

Goal models like this one capture scoring rates, not injuries, suspensions, travel, goalie rotation or roster changes. Treat the probabilities as a calibrated baseline, not a forecast.