Five gates: how an ABEBIT signal is actually made
Most signal services will not tell you where their calls come from, which is usually because the answer is a person with an opinion. Here is ours, stage by stage. It is the actual pipeline, not a simplified diagram.
One. Scan. Every enabled asset is scanned continuously across three timeframes — 4H, 1D and 1W — in parallel. This is the widest point of the funnel and almost nothing survives it. A scan is not a signal; it is a question being asked.
Two. Rules and model. A rule ensemble votes on the setup, and that vote is blended with two machine-learning models trained from scratch on real market data — gradient-boosted trees and a linear model. Not a general-purpose API with our logo on it; models fit to this problem, on this data.
The blend is deliberately asymmetric. The model's influence is capped, so it can shift a decision but never override the rules on its own. A model that is confidently wrong should not be able to publish by itself.
Three. Regime and confluence. This is where most surviving candidates die. A setup has to agree with the higher timeframes around it, and with the market regime it is sitting in — trending, ranging, volatile. A long that looks perfect on its own chart and contradicts everything above it is not a good trade; it is a chart pattern. The macro calendar is checked here too, because a technically clean setup an hour before a major economic print is a coin flip wearing a suit.
Four. Geometry. A surviving candidate gets levels: an entry zone, a stop-loss, and three take-profits, all derived from the asset's own recent volatility rather than a fixed percentage. A 2% stop means something completely different on Bitcoin than on a thin altcoin, and using the same number for both is how people get stopped out by noise on one and wiped out by a real move on the other.
The full reasoning is published with the signal — which rules fired, what the confluence looked like, what the regime was. You are meant to be able to disagree with it. That is only possible if you can see it.
Five. Outcome. Once published, the signal is tracked candle by candle until it resolves as a win, a loss, or a timeout, and written to the record either way. Resolution is first-touch, and when one candle covers both the stop and the target, it is recorded as the stop — the conservative reading, because intrabar order is unknowable and the alternative flatters every result.
What this pipeline does not do is guarantee anything. Five gates make the surviving candidates more considered than an opinion typed into a group chat. They do not make markets predictable, and any service telling you otherwise is selling something other than analysis.
What we can say is that every stage above is inspectable, every outcome lands on the same record, and nothing is quietly removed for looking bad afterwards.