Club football stops for a fortnight and the Nations League fills it: six fixtures on Thursday night, seven more on Friday, all but one of them at 19:45 UK. Below is where our model disagrees with the market across the two nights, and one category of fixture we are leaving alone entirely.
Start with that, because it is the more useful half. There are also nine friendly internationals across these two days, and we will not be pricing any of them. The two biggest apparent edges on Thursday's entire board - a shade over ten points and nine points - both sit in the same friendly. That is not an opportunity, it is a warning that the model is being asked a question it cannot answer.
Six fixtures, all at 19:45 UK, and two worth writing about.
The model makes both teams to score 52 per cent at 6/5, against an implied 45. Portugal scoring at home is not the question; Wales are. They have taken two wins and two draws from their last five and have not been a side that gets shut out. Full match stats here.
Over 2.5 goals is rated at 48.5 per cent against 13/10, and the Republic of Ireland at 2/1 to win are 35 per cent against an implied 33 - a thin margin, and honest to call it that. Full match stats here.
Elsewhere on Thursday the model has nothing to say. Netherlands against Germany is the tie of the night and prices about right; Austria against Israel is the one fixture where the model sits marginally below the market on every selection.
A better card, and three calls.
The biggest disagreement across either night. Hungary at 13/10 are 49 per cent against an implied 43, having lost one of their last five. Full match stats here.
Both are 58 per cent for goals at both ends at 5/6. Neither has a favourite heavy enough to be expected to keep a clean sheet, which is the whole case. Montenegro v Cyprus and Poland v Bosnia and Herzegovina.
Turkey at 11/2 at home to France is rated 19.7 per cent against an implied 15.4, which is a real disagreement and the second-biggest on Friday's card. It is also a 19 per cent shot, and this column has a long record of those being right about the price and wrong about the result. Worth knowing, not worth backing at a stake that matters. Full match stats here.
A projection model works by learning what a team does and assuming it keeps doing it. A friendly breaks every part of that. Managers rotate freely, often changing six or seven at half-time. Players who have just flown in do not start. Nobody is chasing a result, so a side a goal down at 70 minutes behaves nothing like a side a goal down in a qualifier.
The model does not know any of this. It sees two national teams, applies the form it has, and produces a number that looks confident and is not. When our biggest edge of the day appears in a friendly, the correct read is that the model has mispriced the fixture, not the bookmaker.
Competitive internationals are not perfect either - a Nations League side is still not a club side that trains together every week - but there are stakes, the elevens are close to first-choice, and the behaviour at 0-0 in the 80th minute resembles a real football match. That is the line we are drawing, and it is why this page is shorter than it could be.
What are the best Nations League bets this week? Both teams to score in Portugal v Wales at 6/5 on Thursday, and Hungary to beat Ukraine at 13/10 on Friday, which is the biggest disagreement between our model and the market across the two nights.
Why does Statz not price international friendlies? A projection model assumes teams keep doing what they have been doing, and a friendly breaks that: managers rotate freely, often changing six or seven at half-time, and nobody chases a result. When the biggest edge on a board appears in a friendly, the likeliest explanation is that the model has mispriced it rather than the bookmaker.
When is the Nations League on this week? Six fixtures on Thursday 24 September and seven on Friday 25 September, almost all at 19:45 UK. Georgia v Northern Ireland is the exception at 17:00 on Friday.
Is Turkey v France worth a bet? The model rates Turkey at 19.7 per cent against an implied 15.4, which is a genuine disagreement. It is still a 19 per cent shot, and worth knowing rather than backing at a stake that matters.
18+. Please gamble responsibly - begambleaware.org. Odds correct at time of writing and subject to change. Statz publishes model projections, not advice.