How Delta decides a match has value

Delta doesn't guess and doesn't have opinions. It's a deterministic model: the same inputs always produce the same output, and every prediction keeps a record of how it was calculated. Nobody nudges a number by hand.

Here's how it works, in order.

1. How many goals we expect from each side

It all starts with one question: how many goals should each team score in this match? Not how many they scored last time, but how many the chances they create — and the ones they give away — are worth.

To answer it, the model looks for the best information available and walks down six levels, from sharpest to bluntest: an estimate built for that specific match, our own reading of recent form, the season average split by home and away, the overall average, a strength-based calculation for national teams, and a global baseline as a last resort.

Two details matter more than they look:

• When the best estimate exists, it isn't used alone. It's averaged with our reading of recent form. One source sees the match, the other sees the moment; the average sees both.
• If the model falls all the way to the baseline, it recommends nothing. With no real data for that match there's no honest recommendation, and we'd rather say so than invent one.

2. From expected goals to the odds of every scoreline

With those two numbers the model builds the full table of possible scorelines, from 0-0 to 5-5, and gives each one a probability.

It sounds complicated and it's the opposite: once you have that table, everything else is adding up cells. Chance the home side wins? The sum of every cell where they score more. Chance the match goes over 2.5 goals? The sum of every cell adding up to three or more. Same for goals in each half. There isn't a separate opinion per market: there's one reading of the match, seen from several angles.

3. The context the table can't see

Maths is cold and football has circumstances. Before the verdict, seven adjustments are applied, each with a fixed reason and a fixed size:

• Recent form — how each side arrives across their last five matches.
• Urgency — chasing a title or running from relegation changes how a team plays.
• Money movement — when a price moves hard and in one direction, somebody knows something. It gets logged.
• Head to head — meetings between these two sides, counting only the last ten years.
• Neutral ground — there's no home advantage in a final, so it's switched off.
• Rivalry — derbies produce fewer goals than the stats suggest: they're cut by 5 %, and the biggest ones by 7 %.
• Rotation — a congested calendar or a cup tie within 72 hours means that side isn't naming its strongest eleven.

4. Comparing against what the market pays

This is where a probability turns into a decision.

Every price hides a probability. If you're paid 2.00, the bookmaker is saying this happens once in every two. If our model thinks it happens six times in ten, there's a gap in your favour. That gap is the only thing we chase.

And here's the part almost nobody says out loud: it promises nothing about that match. It promises that, repeated many times, choosing that way wins. A single match can go wrong with the best gap in the world.

5. Nothing gets edited afterwards

Every analysis is stored exactly as it ran: the expected goals that went in, which adjustment fired, which price it was measured against. If the model got it wrong, the wrong analysis stays published.

We're not asking you to believe us. We're showing you the arithmetic.