Why most of what we detect never becomes an alert
The engine has evaluated roughly three times more candidates than it has published. The gap is the product.
About one in three candidates clears the funnel. The rest are logged, scored and suppressed and that silence is what you are paying for.
Every coordinated pattern the engine notices becomes a candidate. A candidate is not an alert. It is a row in a table with a score, the evidence behind it, and the result of every gate it was put through. Since the v2.0 engine went live, only about one candidate in three has been published. The other two thirds were either rejected outright or scored and held back.
Nothing is deleted. A suppressed candidate keeps its score, its evidence and the reason it was stopped, which is what lets us audit our own filter rather than argue about it. If we are wrong about something, the record of being wrong still exists.
The gates fall into a handful of families. Evidence: is there actually enough independent activity here, or is this one participant making several transactions? Independence: do the wallets involved look like separate actors, or one actor split across addresses? Tradability: is there enough depth that a normal position could be entered and exited without the exit being the whole move? Contract risk: does the token itself carry the kind of structural problem that makes the flow irrelevant? Freshness: has the window in which this was actionable already closed?
We publish the families, not the numbers. The exact floors, windows and weights are the part of the system that took the longest to get right, and they are tuned continuously. What matters to you is the shape of the decision, not the constants inside it and the shape is deliberately conservative.
There is a cost to this, and we would rather state it than hide it. A conservative filter misses things. There are days when the feed is quiet not because nothing happened but because nothing happened cleanly enough. We accept that trade because the alternative is worse: a feed you learn to ignore is worth nothing, no matter how much of the market it covers.
The way to judge us, then, is not by volume. It is by whether the things we do publish were worth reading. That is a harder standard and it takes longer to prove, which is exactly why we record every measured outcome wins and losses, and publish them rather than quoting a headline number.