What this tool measures
And, more importantly, what it refuses to measure
The question every wallet tracker answers is “who made the most money?”. That question is useless. In a market where thousands of wallets take lottery-ticket bets, the top of any 7-day PnL ranking is populated by whoever got lucky, and the ranking resets every week.
The question here is “whose edge would still exist if I copied it?” — which subtracts three things a naive tracker keeps: the profit that only existed because they were faster than you, the profit that was never repeatable, and the profit extracted from coordinated wallets selling into their own followers.
The composite score, component by component
Copyable ROI
40 pointsEvery closed position is re-simulated as if you had entered it after a delay of 5, 15 and 30 seconds, filling at the price the market actually printed at that moment. Entry pays +2% slippage and 1% fees; the exit pays the same in reverse. If no price printed within 120 seconds of your simulated fill, the position is discarded rather than guessed at.
This is the single most important number on the site, and it is routinely negative for wallets with spectacular raw PnL. That is not a bug — it is the whole point. A wallet that wins by being 400ms early is not a strategy you can rent.
Out-of-sample persistence
30 pointsThe window is split into sub-windows. A wallet's percentile in each is compared against its percentile in the others using Spearman's rank correlation with tie-averaged ranks. Consistency scores; a single outlier trade does not.
When the dataset is too short to split, this component reports n/a and contributes nothing, which deliberately caps the maximum achievable score. A short history cannot produce a high score here, no matter how good the numbers look. See the evidence page for whether persistence is currently predictive at all.
Win rate
12 pointsShare of closed positions with positive realised PnL. Saturating, not linear — the difference between 40% and 55% matters far more than the difference between 80% and 95%, because the latter usually means a tiny sample or a wallet that never cuts losers.
Median ROI
8 pointsMedian, never mean. One 100x on a $12 position should not outweigh forty losses, and with a mean it would.
Sample size
10 pointsCredit for having enough closed positions across enough distinct tokens to be worth reading. Below 8 trades or 4 tokens the entire score is additionally multiplied by 0.4: an unproven wallet cannot rank highly by accident.
Penalties applied after scoring
Ring membership
Wallets clustered by union-find over co-firing events — buying the same token within 30 seconds of each other, repeatedly, across at least four distinct tokens. Confidence combines how often they repeat, how dense the cluster is, and how tight the timing is.
Penalty: up to −75% of the composite score, scaled by confidence. Rings are hidden from the leaderboard by default. Detail on the rings page.
Bot likelihood
Derived from the median delay between a token's launch and the wallet's entry. Consistent sub-2-second entries imply infrastructure — private RPC nodes, mempool access, co-located execution — that you cannot reproduce by clicking a button.
Penalty: up to −60%. Bots are informative to watch and impossible to copy, so they are hidden by default but kept in the dataset.
Where the data comes from
Public on-chain Solana data, indexed through GeckoTerminal, DexScreener, Jupiter, Pump.fun and RPC. Nothing here scrapes Axiom or any other terminal: Axiom has no public API, and the wrappers that claim otherwise drive a headless Chrome through Cloudflare Turnstile — a terms-of-service violation that breaks whenever the challenge changes. The underlying trades are public anyway.
Positions are reconstructed with FIFO lot matching. Each open→flat cycle is a separate position; a sell with no tracked entry is skipped rather than booked as pure profit, which is how naive indexers manufacture phantom winners.
Known limitations
Indexing latency. The tape is minutes behind the chain. This tool is for deciding who to follow, not for reacting to individual fills.
Price reconstruction. Simulated fills use printed trade prices, not order-book depth. On thin tokens, real slippage will exceed the modelled 2%, so copyable ROI is an optimistic upper bound.
Unrealised PnL is ignored. Open positions contribute nothing. A wallet currently holding a winner will look worse here than on trackers that mark to market — deliberately, since unrealised gains are not evidence of an exit strategy.
Coverage is partial. Only tracked pools are indexed. A wallet's score reflects its behaviour in the tokens this system watches, not its entire portfolio.
Scores are a filter, not a recommendation. The realistic best case for this tool is that it removes most of the wallets you would otherwise have wasted money copying. Most memecoin traders lose money, and that includes the ones at the top of every leaderboard, including this one.