Movement Radar

How it is measured

A launch counts as having moved if it at least doubled off its opening mark or lost at least half of it. The 5× column is the same test at five times and one fifth. Everything is measured from the token's own trade history, with no position and no exit rule, so these are properties of the token rather than of anybody's trading.

• The model was fitted on the earlier half of the launches we have recorded and scored on the later half, which it never saw. Every number on this page comes from that held-out half.
• Bands are three, not five, because the data only supports three. We can tell the top ~27% of launches from the rest; we cannot reliably rank inside it, and we do not pretend to. That is also why rows are ordered newest first within a band rather than by score: ordering inside a band would be ordering noise, and freshness is the thing you can actually use.
• The window is deliberately short. Measured across thousands of flagged launches, the first doubling lands at a median of 15 seconds and 89% of them inside the first minute, so a launch more than a few minutes old has usually already done whatever it was going to do. We would rather show you a short honest list than a long stale one.
• The accuracy table above is the check on all of this. Every launch created since the model went live is graded automatically once its trade history is complete, and the result is compared against what we published. We did not want a page whose only evidence was a backtest.
• A name that launches repeatedly inside the window is shown once, with a count. That repetition is almost always a single operator recycling one name, and it is worth seeing as a count rather than as six identical rows.
• Coverage is honest and partial: we hold full trade history for roughly a fifth of launches, and those skew toward the better-observed ones. That history also runs out a few minutes in, so the timing figures above describe how fast the early move arrives, not a claim that nothing can happen later.