playing away in touchdowns: -0.03 (n=7,618), confirmed on an era it had never seen. Context for reading a slate — it never adjusts a served probability.
Discovery applies a Benjamini-Hochberg false-discovery correction across every condition tested in this market at once, so a result here is already adjusted for how many conditions were tried. Confirmation re-runs the same test on a chronologically later era the engine had never seen. The eras differ by sport, because each sport has the history it has.
2025: -0.01, missed
1 season missed since confirmation. A trend that keeps missing is demoted rather than quietly kept.
playing at home is not a second piece of evidence — it is this same comparison written down from the other end, with an identical p-value and an exactly negated effect.
A confirmed trend is context for reading a slate. It never adjusts a served probability, and it is not a bet — the probability you see on a pick comes from the model, and this measurement is not an input to it.
Identifier touchdowns:on_the_road. Synced 2026-08-17.