Wispr Flow: Dictation That Works EverywhereStop typing. Start speaking. 4x faster.
Promoted
Convexly
Skill or luck? Check any Polymarket wallet in 30 seconds
7 followers
Skill or luck? Check any Polymarket wallet in 30 seconds
7 followers
Every prediction market ranks traders by profit, but over a few dozen bets, profit is mostly luck. Convexly is an independent skill-vs-luck audit for Polymarket wallets. Paste any wallet address and in about 30 seconds get its realized edge with a 95% confidence interval, calibration by market category (Brier score), and an Edge Score against a frozen 8,656-wallet reference cohort. Free, no signup. Frozen methods, public data, and negative results that stay published.
How does the Edge Score actually get calculated against the 8,656-wallet cohort, and are you weighting newer wallets the same as ones with years of historical data?
@cemilalpermgp8 Fair question. Short version on the second half: not by age, by sample size.
The score is a percentile against a frozen 8,656-wallet cohort. Under the hood it's three z-scored pieces: how calibrated the wallet is (baseline Brier minus its own), how concentrated its PnL is (that one carries the most weight), and how many resolved positions it has. Rank that against the cohort and you get a 0 to 100. In-sample Spearman is +0.42, and I keep saying in-sample because that's what it is. I haven't shown it holds out of sample yet and I'm not going to pretend otherwise. Full coefficients are on the methodology page.
On age: no, we don't fade old bets or favor recent ones. It reads the whole resolved record. What actually protects against a thin or brand-new wallet is the interval, not a date filter. The 95% band is a clustered bootstrap, so it blows out when there aren't many resolved positions, and under 30 the tool just won't call anyone skilled. Ten years and 12 resolved bets gets the same shrug as a week-old wallet with 12.
If you think recency should count for more, make the case and I'll listen. Or paste a wallet you think it's getting wrong and I'll look.
@cemilalpermgp8 Correction to my own comment above: the composite's in-sample Spearman is +0.514, not +0.42. I conflated it with the calibration-only correlation from our whale audit. The methodology page has the exact number. That's on me.