Accuracy

    Confirm the weather before you bet

    By OverUnderWeather Research Desk

    For every completed game we compare the forecast we showed you against what the venue actually got, and publish the error by lead time. Unflattering numbers stay up.

    Loading accuracy data…

    Forecast versus what the venue actually got

    Every read we published, scored against an independent Open-Meteo archive observation for the same stadium hour. Each sport states how many fixtures sit behind its numbers — a small count means early, not wrong.

    How this is measured

    Method, tolerances and refresh cadence

    For every fixture we store a trail of forecast reads, from three days out down to the hour before kickoff. "Within tolerance" means the read was close enough to confirm conditions: 3 mph of wind and 4°F of temperature for US sports, the metric equivalent for AFL.

    The observed tables score every read against an independent Open-Meteo archive pull for the same stadium hour. That guards against provider drift and is the number we stand behind. The headline numbers at the top of this page come from the day-of reads (inside 24 hours) of that same observed view.

    Rolling 90-day window, refreshed automatically.

    How we handle rain when our two weather providers disagree

    Every forecast is built from a primary provider plus a second, independent rain cross-check. When the two disagree on rain, we currently publish the wetter of the two readings — a deliberately cautious default, so we do not print "Clear" over a drizzly ground.

    That default was a judgement call, not a measured one, and we are not going to pretend otherwise. We have started storing each provider's own rain reading separately alongside the combined figure, so we can grade them individually against the rain that actually fell: how often each provider was right when it called rain, and which one was right on the games where they disagreed.

    Until that sample clears our minimum size, the accuracy of the wetter-reading rule is not yet established and no per-provider hit rate is published here. If one provider turns out to be measurably more reliable when it calls rain, we will weight it accordingly and publish the numbers behind that decision on this page. If the two are level, the wetter reading stays and we will say so.

    How game-time leans are derived

    Each game-time lean is produced by the same published-study coefficients shown on the research page. The model checks the forecast against the bands we back-tested — wind speed and direction, rain chance and intensity, temperature and, where available, roof state, surface condition, and visitor climate-shock inputs.

    A lean only becomes Over or Under when the relevant band clears both our minimum sample threshold (40 games) and a |t| near 2 significance check. Anything short of that is published as Neutral, and thin or pre-set bands are flagged as directional evidence only. This is why a strong-looking forecast can still return Neutral — the numbers behind it are not yet strong enough to call.

    The studies these coefficients come from:

    AFL closing totals and study grading

    The AFL Temperature & Scoring study was built without historical closing totals, so it reports scoring effects only. We are now capturing AFL totals as bookmakers post them, storing each line alongside the kickoff weather read and the final score.

    The model predicted total for each graded game is the season-average combined score plus the measured rain, wind and temperature adjustments from our published studies. We then compare that prediction to the captured closing total and to the actual result. If the market already prices cold Melbourne nights and warm Darwin afternoons, the model will not beat the closing line and the study will stay scoring-only. If a persistent edge appears, we will publish the band-by-band numbers and update the lean model accordingly.

    No AFL hit rate will be shown until each temperature band reaches 40 graded games — the same minimum sample we require before any study claim is treated as reliable.

    Self-scored consistency (not an accuracy claim)

    These tables score every earlier forecast read against our own read taken closest to kickoff. They show whether our numbers stayed stable as the game approached — a consistency check on ourselves, not a measure of how right we were.

    Where to go next

    Check live conditions on the NFL, MLB, AFL or college football desks.

    OverUnderWeather Research Desk

    Independent data analysis on public sports and weather records. Studies match historical game outcomes to archived game-window weather, apply pre-set sample and significance rules, and publish null results when that is what the data show. Editorial policy and corrections log.