Every tracker shows a different PnL for the same wallet — the venue's own page, the hype terminals, us. That isn't a bug in one of them; it's a difference in what gets counted. Most numbers are built to look big. Ours is built to survive checking. Here is exactly what we count, what we refuse to count, and how the score is made — so you can audit the method instead of trusting the brand.
| Excluded | Why |
|---|---|
| Unrealized / open positions | "Up 90%" on an open bag is a screenshot, not a result. It counts when it settles. |
| Funding & fees (perps) | We measure trading skill from trade fills; funding flows are reported separately by the venue. This is the main reason our perps number sits below the venue's account PnL — the relationship is measured and published in our receipts. |
| Tokens with no traceable purchase | If a wallet sells tokens it never verifiably bought on-chain (airdrops, transfers in, unindexed sources), we drop the position rather than invent a cost basis. Windfalls aren't trading skill. This can make our number lower than inclusive trackers for wallets that dump received tokens. |
| Partial histories | If we couldn't pull a wallet's complete record, its dollar figures are withheld and the row is marked partial. We don't publish numbers we haven't finished computing. |
Two questions, two axes. Is the edge real? — profit factor (gross wins ÷ gross losses), breadth (what share of gains come from the top 3 trades; concentrated gains are luck-prone), sample size, consistency. Can you actually follow it? — a wallet firing 90 trades a day can be genuinely skilled and still uncopyable: by the time you mirror it, you're buying the price it already got. Real edge and followable edge are different things, and we label them differently.
The Tells Score (0–100) is a weighted blend of edge 40%, breadth 25%, sample robustness 20%, and consistency 15% — then capped by the verdict, so the number can never contradict the label: a wallet with a losing record can't score above the low 20s no matter how pretty its components, and a verified real edge has a floor. Grades: A ≥85 · B ≥70 · C ≥55 · D ≥35 · F below. Wallets with too little history get no score at all rather than a misleading one.
We reconcile the engine against external ground truth across many wallets per venue — the venue's own published numbers, independent trackers — and keep a running public log of the results, including the misses. Where our number differs systematically (funding/fees, unrealized, untraceable tokens), the difference is quantified, not hidden. A number you can't check is just a claim.
If a verdict concerns a wallet you control and you believe it's wrong, tell us what's off — we'll re-run it publicly and correct or annotate it within 48 hours. For a record-checker, the correction is the brand. We keep the raw data behind every published verdict so any number can be reproduced as of its post date.
Scores and labels are our opinion of what the public on-chain record shows, stated with the math open. They are not financial advice, not a recommendation to copy anyone, and not an accusation that any person has committed fraud. We have no signal to sell, no copy group, and no referral code — the only honest read in this market is the one from someone who doesn't want you in a trade.