Every memecoin influencer's track record, every signal group's "90% win rate", every viral backtest thread shares one property: the dataset only contains tokens that lived long enough to be counted. We run a scanner that logs every alert it fires — 475,000+ and counting — at the moment of firing, with live market data, before knowing anything about the future. That design choice is the whole difference between real numbers and marketing. Here's what the unfiltered data says.
Survivorship in memecoin data isn't one bug — it's a pipeline of disappearances, each removing losers before anyone counts:
- Death before first snapshot. A meaningful share of new tokens rug within minutes. Any tracker that polls on an interval — including ours, before we fixed it — simply never records them. They existed, they took money, they're not in your denominator.
- Delisting from the data source. Price APIs stop returning dead pairs. Backtests that query "all tokens" a month later are querying all survivors.
- The influencer's memory. Calls that rugged get deleted or just never mentioned again; the 20x stays pinned. This is survivorship as a content strategy.
- Lifetime-peak accounting. Our own early stats used each token's lifetime peak — which credits an alert with a pump that often happened before the alert. Fixing this plus the dead-token gap moved measured 2x rates by about 30 percentage points.
Survivorship exists everywhere, but memecoins maximize it: token lifetimes are measured in hours, the death rate is extreme, and the winners are so loud that a 5% hit rate can fund an entire ecosystem of screenshots. When we studied exits, we found the median rug goes to zero faster than a human can react to an alert — which is why our research concluded that on-chain memecoin edges live in entry selection, not exit tactics. No stop-loss saves you from a contract that blocks selling.
- Ask when records are created. Honest: at signal time, before the outcome. Poisoned: backfilled later from whatever still has a price.
- Ask for the denominator. "We called 12 tokens that did 10x" means nothing without the total number of calls. Ours is public: the honest 10x rate on our flagship detector is 6.8%, and we published the downgrade from the fake 45%.
- Check for impossible peaks. Data glitches print billion-dollar market caps on manipulated pools; without sanity caps they become fake wins.
- Prefer forward-validated claims. A rate measured on alerts fired after the methodology was frozen can't have been curve-fit to the past.
This is the standard we hold our own public track record to, and the same logged-at-fire-time data now powers the memecoin mode of the backtest Studio — so when you test an entry filter, the rugs are in your denominator too.