What the engine actually did.
Nothing here is a projection. These are decisions the engine made on live Solana data in paper mode, and what the coins it refused went on to do. Snapshot taken Wed, 19 Aug 2026 12:40:06 UTC.
What the filters cost us
Every refusal is graded again later. If a coin we skipped went on to do 2x or better within 6 hours, that counts against the gate that refused it. A gate with no cost is usually a gate nobody is testing.
| Gate | Refused | Later doubled | Regret | Median skipped | Best skipped |
|---|---|---|---|---|---|
Outside your universe outside_universe | 333 | 10 | 3% | x0.55 | x17.9 |
Holders too concentrated holder_concentration | 16 | 0 | 0% | - | - |
Volume growth too low vol_growth_below_min | 9 | 0 | under 12 samples | x0.52 | x0.5 |
No baseline yet warming_up | 3 | 0 | under 12 samples | x0.48 | x0.5 |
"Best skipped" is the highest multiple any refused coin reached. Above x1 it is money the filter cost us. Below x1 it means the best of everything that gate turned down still went down, which is the filter doing its job. A rate is only shown once a gate has at least 12 graded refusals. Below that the count is published and the percentage is left blank on purpose: three refusals and one winner is not a 33% failure rate, it is noise, and putting a decimal point on noise is how people talk themselves into turning a filter off.
Why it said no
Every decision carries a category. These are the counts across the whole store.
The wallet lists
Screened against the holders of every token the engine looks at. A wallet on more than one list keeps its most dangerous label.
Paper mode means the engine made every decision exactly as it would live, and simulated only the fills. No live trade has been placed on the platform. These numbers are a snapshot of one machine over a short window, not a track record, and they are not a prediction of anything.