Elo Grid Mixt

This round's Mixt pick

Each round we take the condition + box pair that came up most often in the season's earlier rounds — the current round is never used to choose. A box is “complete” only if every match in it meets the condition: if a single one does not, the box fails (2 out of 3 does not count).

Other strong boxes

The conditions that did well in the check, on the round's most lopsided matches, when the Mixt pick is on another box. Benchmark = the same number of matches taken at random from the same round.

Combined boxes: 3, 6, 9, 12, 15 matches

Here every match gets ITS OWN condition: the most likely of its conditions inside the band above. We take the matches that have such a condition, in Elo grid order, and build longer and longer boxes. A box is complete only if every match comes up on its own condition; the longer it is, the more rarely it comes up.

On every match: its most likely condition between ≈ 1 in 1.60 and ≈ 1 in 1.30 (62.5–76.92%) · round 23 Oct–26 Oct

The check on past seasons

The check on past seasons, with the same rule: how many complete boxes the model expected (the sum of the probabilities) and how many came up. When the two are close, the share shown on a box is fair — neither inflated nor too cautious.

Condition simulators

Combinations of 3, 6, 9, 12 and 15 matches from the frozen round, on four conditions: 3+ goals, both teams score, 1st half: 2+ goals, both halves: 1+ goal. The matches come from the Elo grid positions where one of the conditions came up most often in earlier rounds; variant 1 puts on every match the condition with the best record, variant 2 the second, variant 3 the third. “By the model” instead takes the most likely matches and conditions by the model's calibrated estimate, frozen before the round (without “both halves”, which has no estimate). “By the model, 67–76.92%” takes on every match its most likely condition among all conditions, only if the estimate is within that band — below the threshold for conditions that are too easy. A combination is green only if every match in it comes up. It shows what happened, not what will happen.

Round 23 Oct–26 Oct · order frozen ❄

How the simulators did

The same rule applied to every round of past seasons, picking only with the rounds before it: how often it would have come out all green and how often it would have had at least one red. The “promised” column = what the model expected (the “model” row) or how often it would have come out if the position rates repeated exactly (variants 1–3); the model is close to what happened, the positions promise far more; in reality far fewer did, and a long combination almost never comes out whole.

The model against the result, round by round

For every round with estimates frozen before kick-off: what the model gave for the complete box (the product of the per-match estimates) and whether it came up. The sum of the estimates = how many complete boxes the model expected; if many more or many fewer come up, it shows here.

What it does not show

What to know before reading a green box: the history describes past rounds, not what comes next.

• A box does not make matches more likely: over 2023-24 – 2025-26, Mixt boxes were complete in 62% of rounds, and the per-match rates multiplied give 60% — the gain comes from the matches chosen, not from any link between them.
• Next to the pick we show only the conditions that beat the benchmark on the most lopsided matches in BOTH checked periods (2023-24 + 2024-25 and 2025-26). 2+ goals, 3+ goals do not: 2+ goals — 53% vs 46% at random, then 78% vs 45%; 3+ goals — 19% vs 15% at random, then 36% vs 14%.
• The history describes past rounds; the model estimate is separate and has its own public calibration.