⚠️ Our AI model is still learning from match data. All predictions are experimental statistical estimates for information purposes only — not financial advice and not an invitation to bet. Outcomes are never guaranteed. 18+ · Gamble responsibly.
Colour scheme
Language
English ✓ Español ✓ Português ✓ Deutsch ✓ Français ✓
Home/Research/CLV experiment

The CLV experiment

Closing-line value is the professional's early truth test: if you consistently beat the last price, you have an edge long before ROI can prove it. We measure ourselves in public. Updated 2026-08-05.

Where the experiment stands

106Bets measured
-0.24%Avg CLV (all)
-0.24%v1 era (105)
0%v2 era (1)
22/106Beat the close

CLV = (price we took ÷ last tracked price − 1). Our snapshots are taken at each refresh, so "close" means the last price we saw before kick-off, not the true market close — labelled honestly.

Why CLV, not ROI

Win-rate and ROI are what everyone quotes and what almost nobody should trust on small samples: a few hundred bets can sit 15% away from their true expectation by pure chance. Closing-line value converges far faster. The logic is simple — the closing price is the market's most informed estimate, sharpened by all the money that arrived before kick-off. If the price you took is consistently better than where the line closed, you were systematically ahead of the market's final opinion; profit follows over time. If you can't beat the close, no run of wins will save you.

The experiment's rules

Every official pick is archived at publication with its price and timestamp. Nothing is deleted or edited — losing runs stay on record next to winning ones. At settlement, each bet is annotated with the last price we tracked before kick-off and its CLV. The first era of our model (v1, a pure market-consensus approach) averaged negative CLV — exactly what theory predicts for a model with no information of its own, and we say so openly. The current model (v2) blends our own Poisson goal model with the market and bets selectively; its CLV track is accumulating now, in public, on this page and on the performance page.

What would falsify us

If v2's average CLV settles below zero over a few hundred measured bets, the model does not have an edge — and this page will show it. That's the point: a prediction site that can't lose an argument with its own data isn't making predictions, it's making content. Every number here is reproducible from the public settled log.