The BTP model is a machine. It does not have feelings, a Bet365 account, or opinions about Wigan. It produces probability estimates. This post is data analysis — it is not a tipster service, a recommendations column, or a suggestion that you do anything at all with any of this. The model has never watched a match in its life. It doesn’t even have a life. It has features.
Every gameweek we run our ML models across every Championship and League One fixture. Usually those probabilities sit in a database doing nothing. This week, we went one step further: we compared them to what Bet365 is actually pricing.
The idea is simple. We’re not looking for certainties — football doesn’t have those. We’re looking for fixtures where Bet365’s implied probability is significantly lower than what our model estimates. When Bet365 implies a 25% chance and our model says 50%, that gap is the interesting thing. That gap is the expected value.
We applied three filters: EV ratio ≥ 1.20 (model is at least 20% more confident than the market), model probability ≥ 25% (no speculative outsiders), and draw probability < 40% (avoid fixtures where the result is genuinely a coin-flip three ways). Twelve fixtures across both leagues cleared those filters. Here are the top six by EV.
One thing worth noting before you scroll: almost none of them are home favourites. When the market is most wrong, it tends to be on away teams.
Championship predictions use our Platt-calibrated Logistic Regression (goals_logreg_v1_cal). League One uses the original Random Forest (leagueone_goals_v1). Why different models? Read this.
HOW EV WORKS — PLAIN ENGLISH
Bookmakers express their probability as implied odds. Divide our model’s probability by that implied probability and you get the EV ratio. EV of 1.0 = the model and the market agree exactly. EV of 1.97 = the model thinks this outcome is almost twice as likely as the market does.
EV doesn’t guarantee profit on any single fixture — the bookmaker might be right and we might be wrong. What it measures is where our model’s estimate diverges most from market consensus. Over a large enough sample, positive EV selections should return more than they cost. One weekend’s Heinz proves nothing either way.
The Full EV Landscape — All 22 Fixtures
All Championship GW42 and League One GW43 fixtures with Bet365 odds available, sorted by EV ratio on the best outcome. ★ = Heinz selection. The Heinz crosses both leagues by design — selections are the top 6 by EV regardless of division, which is why this week’s six contains one Championship pick (Hull City) and five League One picks. That’s what the data returned, not an editorial choice.
★ = Heinz selection (top 6 by EV). Faded rows = didn’t qualify (model_prob < 25%, or EV below threshold). B365 implied = normalised Bet365 probability (overround removed).
WHY COVENTRY ISN’T IN THE HEINZ (AND WHY THAT’S THE WHOLE POINT)
Coventry v Sheffield Wednesday has the highest raw EV ratio in the table — 2.95 on Sheffield Wednesday winning away. But the model only gives them 16.4%, which falls below the 25% minimum. Rightfully excluded. More interesting is what happens to Coventry’s home win: the model says 69.2%, while Bet365 implies 84.0% (odds 1.12). EV: 0.82.
Coventry are a home favourite the market has already priced to the hilt. High model confidence doesn’t mean high EV — it means the bookmaker is equally, or more, confident. When both agree, there’s nothing left in the gap. That’s why the Heinz contains zero home favourites from last week’s most-confident list, and five away teams instead.
★ The Heinz — Six Where the Model Diverges From the Market
Top 6 by EV ratio from strict filter (EV ≥ 1.20, model ≥ 25%, draw prob < 40%). Five away teams, one home win.
LEAGUE ONE · SAT 11 APR
Doncaster v Reading
SELECTED: READING AWAY
EV RATIO
1.97
odds: 3.70
49.8%
25.3%
Reading are 8th, Doncaster 17th. The model gives Reading nearly 50% to win away; the market gives them 25%. The model thinks this is genuinely competitive. Bet365 thinks Doncaster are a home side that usually turns people over. One of them is wrong by a lot.
CHAMPIONSHIP · SAT 11 APR
Sheffield Utd v Hull City
SELECTED: HULL CITY AWAY ←
EV RATIO
1.78
odds: 4.20
40.1%
22.5%
Hull away at Bramall Lane. The model gives City a 40% chance; Bet365 implies 22.5% (odds 4.20). That’s a 78% edge by the model’s estimate. The Blades are hosting a Hull side whose form the rolling features rate more highly than their current league position. The market is backing the home shirt. The model is backing the numbers.
LEAGUE ONE · SAT 11 APR
Burton Albion v AFC Wimbledon
SELECTED: AFC WIMBLEDON AWAY
EV RATIO
1.66
odds: 4.20
37.0%
22.3%
AFC Wimbledon are the away side the model rates most highly relative to their price this weekend. Burton’s home record isn’t convincing the algorithm of much. Wimbledon at 4.20 — the model thinks they should be around 2.70. That gap is substantial.
LEAGUE ONE · SAT 11 APR
Blackpool v Peterborough
SELECTED: PETERBOROUGH AWAY
EV RATIO
1.65
odds: 3.75
41.0%
24.9%
Draw probability here sits at 35.6% — the highest of the six selections, and the one most likely to bite us. The model still favours Peterborough away and the EV is clear. If this is the one that draws, we’ll mention it in the review and explain what we had for breakfast.
LEAGUE ONE · SAT 11 APR
Huddersfield v Wycombe
SELECTED: WYCOMBE AWAY
EV RATIO
1.42
odds: 3.20
41.4%
29.2%
Wycombe at 3.20 while the model gives them over 41% — another case of the market pricing in home advantage more heavily than the rolling form warrants. Huddersfield’s home record hasn’t been convincing; Wycombe’s away form has been serviceable. The model has noticed.
LEAGUE ONE · SAT 11 APR
Rotherham v Barnsley
SELECTED: ROTHERHAM HOME
EV RATIO
1.33
odds: 2.45
50.4%
38.0%
The only home win in the six. Rotherham at home with a 13-point gap in the model’s estimated probability vs Bet365’s implied. A South Yorkshire derby where the model thinks Rotherham are more likely to win than the 2.45 suggests. The only selection where the model likes the home team and the market disagrees.
★ The Heinz — All Six Selections
A Heinz is named after the Heinz 57 varieties slogan — 57 combinations from 6 selections (15 doubles, 20 trebles, 15 four-folds, 6 five-folds, 1 six-fold). It’s a framework for exploring how model disagreement compounds across multiple fixtures. In this context, the EV on the six-fold (all six correct) is the product of individual EVs: 1.97 × 1.78 × 1.66 × 1.65 × 1.42 × 1.33 = 18.1x what the market implies.
The combined model probability of all six being correct is 0.63%. That sounds terrible. The market prices it as 0.035% (implied by the combined odds of 1,919×). The gap between those two numbers is what the EV analysis is actually saying.
For all six to land, the model puts the probability at 0.63%. For context, five are away wins at odds between 3.20 and 4.20. The 55 partial combinations give the Heinz format its structure — at least two correct gets you something from the doubles. Though if five of these are home wins, the model will be in a corner rethinking its priors.
EV Ratio — The Six Selections Visualised
Bar length = EV ratio. The dotted line marks EV = 1.0 (break-even — model and market agree). Anything to the right of the line represents model disagreement with the market price.
1.00
1.97
1.78
1.66
1.65
1.42
1.33
Bar scale: EV 1.0 = 51% width. Green = EV ≥ 1.60. Teal = 1.30–1.59. Amber = 1.20–1.29. Red line = break-even.
Value on the Draw — Three Fixtures Worth Noting
Three fixtures where the draw specifically offers positive EV or high model probability — ranked by draw-specific EV. Worth noting: the draw calibration work we did on the Championship model means these aren’t just raw probability guesses. Full context here.
LEAGUE ONE · SAT 11 APR
Plymouth v Exeter City
30.8%
model draw probability
vs Bet365 implied 23.4% (4.00)
The best-EV outcome for this fixture is actually the draw. The model gives Plymouth 30.8% for a stalemate; Bet365 has it at 23.4%. Neither side convinces the algorithm of a decisive result.
LEAGUE ONE · SAT 11 APR
Cardiff v Bolton
29.8%
model draw probability
vs Bet365 implied 23.4% (4.00)
Draw EV 1.27 — the model sees Cardiff v Bolton as evenly matched enough that the stalemate is underpriced. Bet365 agrees that Bolton may win but hasn’t fully priced the draw probability.
LEAGUE ONE · TUE 15 APR
Luton v Northampton
42.4%
model draw probability
no Bet365 odds yet — EV unavailable
The model’s only outright draw prediction across the weekend. Bet365 odds not yet available, so EV can’t be calculated. At 42.4%, it’s the highest draw probability of any fixture — Luton and Northampton cancel each other out in the numbers.
The BTP model was built to produce honest probability estimates and write interesting data posts — not to tell anyone what to do on a Saturday afternoon. It has identified five away teams and one home win as having positive EV relative to Bet365’s market, which it finds analytically interesting and makes no promises about. The model has no financial interest in any outcome. It simply disagrees with the market on six occasions this weekend, and it stands by every decimal point right up until kick-off, at which point it becomes everyone else’s problem.
Data notes: Championship predictions use the Platt-calibrated LR model (goals_logreg_v1_cal). League One uses the original Random Forest (leagueone_goals_v1). Bet365 decimal odds fetched 9 April 2026 via API-Football. Implied probabilities normalised to remove overround. EV = model probability ÷ normalised implied probability. Luton v Northampton excluded (no Bet365 odds available). Coventry v Sheff Wed away excluded (model probability 16.4% below 25% minimum). 12 fixtures passed strict filters; top 6 by EV shown.
Full model methodology here.
