Predictions

What the rolling-strength model forecast before each kickoff — win/draw/loss probability and expected score — graded against what actually happened. Out-of-sample (the model never saw the result). What do these mean?

This matchday

The model's forecast for the upcoming fixtures — make your own picks before kickoff and see how you stack up.

Upcoming — Matchday 12026/27 · provisional

model forecast · 9 of 342 fixtures

the rolling-strength model's pre-match call for games not yet played — win/draw/loss probabilities. Analysis only. The 2026/27 fixture list is still provisional — dates can move, and forecasts use end-of-2025/26 team strengths.

Predict the matchday

Lock a 1X2 call before kickoff, then see if you beat the model when the results come in. Free — just sign in.

UdinesevComo
locked
IntervMonza
locked
GenoavNapoli
locked
ParmavCagliari
locked
VeneziavLecce
locked
AtalantavSassuolo
TorinovMilan
BolognavLazio
RomavFiorentina

Model probabilities are this site's own forecast. Picks lock at kickoff and are graded from the final result — no stakes, just bragging rights.

Season track record

Every call the model made this season, graded against the result — and, more importantly, whether its probabilities held up.

Outcomes called

197/380

Accuracy

52%

consensus ~53% · always-home ~46%

Avg log-loss

1.026

lower is better

Avg Brier

0.615

lower is better

Calibration

do the model's probabilities hold up? Each dot is a probability band — its height is how often those calls actually happened. On the dashed line = perfectly calibrated.

0.025 calibration error (0 = perfect)

Accuracy is the wrong lens for a three-way outcome: draws are rarely any side's single most-likely result, so ~25% of games are "wrong" by design. What matters is whether a stated 60% really happens ~60% of the time — that's calibration, and it's the number that actually matters.

Model vs Market

how the model's pre-match probabilities compare to the market consensus forecast — the sharpest, most efficient benchmark available. Lower log-loss = more accurate. We don't claim to beat the consensus; we show we're close to it, with the logic fully open. What's this?

Match result (1X2) · log-loss

Model log-loss

1.026

Market log-loss

0.977

Gap to market

+0.049

market is sharper

Over/Under 2.5 goals · log-loss

Model log-loss

0.700

Market log-loss

0.698

Gap to market

+0.002

market is sharper

Cumulative accuracy over the season model market

Over 380 matches, the model lands within 94.9% of the consensus forecast's accuracy. The consensus wins narrowly — as expected, since it aggregates all available information. Source: football-data.co.uk (closing: Pinnacle/Bet365). Analysis only.

Where we disagreed with the market

The games our model and the market consensus read most differently — and who called it. Across these 8 most contested matches the market was closer 5×, our model 3×. Disagreeing with an efficient market usually means the market knew something (news, lineups) our public-data model can't see.

Inter 3–0 Cagliari · Inter win

model 51%(H)market 78%(H)market ✓

Hellas Verona 3–1 Atalanta · Hellas Verona win

The market strongly favored Atalanta (58% away), while the model gave Verona the edge (43% home), likely reflecting information the market priced in beyond a public-data model's reach. Verona won 3-1 despite being outshot on xG (0.99 vs 1.61), vindicating the model's higher home probability even as the scoreline outran the underlying chances.

model 40%(H)market 16%(A)model ✓

Torino 2–2 Inter · Draw

The market strongly favored Inter (68% away) while the model gave Torino a better chance (31% home), likely because the model weighed recent xG more heavily whereas the market may have priced in information a public-data model cannot capture. The result showed a 2-2 draw with Torino outperforming on xG (2.51 vs 1.47) across a game with late goals, validating the model's more balanced view.

model 28%(A)market 19%(A)model ✓

Juventus 2–0 Parma · Juventus win

The model gave Juventus only 41% versus the market's 68%, likely underweighting factors a public-data model cannot capture, and the result favored the market as Juventus controlled play (2.4 xG to 0.6) before scoring at the 59th and 84th minutes for a comfortable 2-0.

model 43%(H)market 68%(H)market ✓

Juventus 1–1 Hellas Verona · Draw

model 29%(H)market 13%(H)model ✓

Juventus 3–1 Udinese · Juventus win

The model rated the match nearly even at 37% for Juventus, while the market strongly favored the hosts at 66%, likely pricing in information a public-data model cannot capture. The result validated the market, as Juventus dominated by xG (3.21 vs 0.54) and won 3-1.

model 42%(H)market 66%(H)market ✓

Hellas Verona 0–1 Milan · Milan win

model 39%(A)market 62%(A)market ✓

Pisa 0–2 Inter · Inter win

model 54%(A)market 77%(A)market ✓

"closer" = the side that gave the actual result the higher probability. Pick badges: H/D/A = home / draw / away. One-line reads are AI-written strictly from each match's data (xG, red cards, goals, score) — no outside news.

Biggest upsets

rarest results the model gave little chance

Milan 0–1 Parma

9% chance

Cagliari 1–0 Juventus

11% chance

Bologna 1–3 Cremonese

12% chance

Confident & correct

bold calls that landed

Cremonese 0–2 Inter

called at 80%

Juventus 5–0 Cremonese

called at 80%

Inter 2–0 Genoa

called at 79%

Biggest misses

high-confidence calls that flopped

Juventus 2–2 Lazio

backed at 79%

Juventus 1–1 Sassuolo

backed at 77%

Milan 0–1 Parma

backed at 75%

Matchday

Matchday 38

Predictions

Season track record

Every call the model made this season, graded against the result — and, more importantly, whether its probabilities held up.

Outcomes called

197/380

Accuracy

52%

consensus ~53% · always-home ~46%

Avg log-loss

1.026

lower is better

Avg Brier

0.615

lower is better

Calibration

do the model's probabilities hold up? Each dot is a probability band — its height is how often those calls actually happened. On the dashed line = perfectly calibrated.

0.025 calibration error (0 = perfect)

Model vs Market

Match result (1X2) · log-loss

Model log-loss

1.026

Market log-loss

0.977

Gap to market

+0.049

market is sharper

Over/Under 2.5 goals · log-loss

Model log-loss