Build a system in plain terms — four wins in the last six at home, priced above 1.40 — and HavilahWin scans the week and shows only what fits. Then it keeps score of itself, in public, against real results.
Every sport gets its own site, its own app and its own maths — because the model that prices a 1–0 is the wrong model for a tiebreak. What they share is the engine underneath, and one ledger that grades all of it.
156 competitions. Poisson with Dixon-Coles, Shin devigging, 26 signals and a calibration loop that learns from settled results.
football.havilahwin.com →Next. Shares the football model — low-scoring, Poisson-shaped — so it is the shortest step from what already works.
icehockey.havilahwin.comHigh-scoring, so the goal model gives way to a Normal distribution around the spread and total.
basketball.havilahwin.comNo draw, and a match is a hierarchy of points, games and sets — the biggest departure from the football engine.
tennis.havilahwin.comIce hockey is next. Basketball and tennis follow it. When there are three, havilahwin.com becomes the all-sports product: ask it for a 5× accumulator and it searches every sport at once instead of one.
Each published prediction is recorded at the price and the probability it had at the time, then graded against the real result. Nothing is edited afterwards, and the losers stay in the table next to the winners.
If it says 70% and 70% land, it is calibrated. If it says 70% and 55% land, the page says so in red and the model is pulled down. A market that keeps under-delivering is demoted automatically.
Under thirty settled results a market is greyed out and left unjudged, because at that size one match moves the number by tens of points. Calling that a finding would be the same mistake as quoting an unmeasured hit rate.
One email when it launches. Nothing else, and no list-selling — tell us which sports and that is all you will hear about.