Every wallet tracker on the market ranks by trailing profit, which guarantees the top of the list is whoever got luckiest recently. The only honest test is out-of-sample: score wallets on one window, then check whether that score predicts a completely separate later window the model never saw.

That is what the backtest below does. It reports Spearman's ρ between train-window and test-window performance, plus the lift of the top-decile wallets over the median. If ρ is not meaningfully positive, the model is decoration and the correct response is to stop using it — not to reshuffle the weights until the number looks better.