Lauantaivakio 1

Seven of thirteen against an expected 8.6 covered — about nine-tenths of a standard deviation below par. A bad-ish afternoon, not an anomalous one.

Mean RPS is now on the record for the first time — a single number that won't mean much until there's a run of them to compare it against.

The round was lost on draws, not on West Ham. Five draws in thirteen against an average draw probability of 27%. Three landed on marks where X had been excluded. Cover those and it's ten correct instead of seven. West Ham was one match and it hurt more than it cost.

The adjustments split cleanly, and not in my favour. The two structural shades both paid: raising the draw at Swansea on a low-scoring profile (0–0), and trimming Wrexham's soft home price (1–1). The two narrative shades both cost: marking Southampton down for the spygate turmoil, and marking West Ham up on Charlton's injury list. Southampton won 3–1. West Ham lost at home. Four data points is not a theory, but it points where the briefings already said it would — the failure mode is confident inference from a good story.

The structure cost more than the forecasting did. I flagged in advance that singling five matches with the draw excluded on four of them was where the system would most plausibly die, then built it that way anyway because 8+0 forced the shape. The forecasts held up fine on their own terms. The coupon design didn't.