Prooty

What is this?

Prooty is a statistical model that predicts AFL game outcomes. It learns from years of historical match data and updates automatically after every game — so predictions always reflect current form.

What goes into each prediction?

Every prediction is based on five things the model tracks for each team:

• Overall strength (Elo) — a running power rating that goes up when you beat strong teams and down when you lose to weaker ones. Bigger margins shift it more.
• Attack rating — how many points a team typically scores, considering how strong the opponent's defence was in the respective games, weighted toward recent games. Higher is better.
• Defence rating — how many points a team typically concedes, considering how strong the opponent's attack was in the respective games, weighted toward recent games. Lower is better.
• Home ground advantage — each team's historical performance boost when playing at their home ground, calculated from past results.
• Head-to-head history — whether a team has consistently over- or under-performed against a specific opponent relative to what the model expected.

How is the predicted margin calculated?

All five factors above are fed into a statistical model trained on thousands of historical AFL games. It learns how much each factor typically contributes to the final scoreline and uses that to estimate the expected margin for any upcoming matchup.

Recent games carry more weight than older ones — last month matters more than last year.

What does the win % mean?

The win percentage reflects how confident the model is, not just who it thinks will win. A 90% means the model is very confident — a 55% means it's basically a coin flip. It's based on the predicted margin and how much AFL games typically vary from what models expect.

What is the projected ladder?

The ladder shows expected wins (xW) — a projection, not the current standings. For every remaining game, the model calculates each team's chance of winning. Those probabilities are added up across the season to estimate how many wins each team is likely to finish with. Games already played count as full wins or losses.

Reading the cards

• The bar shows the home/away win split
• Blue name — model's tip for upcoming games
• Green name — actual winner of a past game
• Red name — who the model tipped but lost
• Past cards show what the model predicted vs what actually happened