How we will grade ourselves
The measurement rule is already running and the results are public. What we will not do is quote a number before it means anything.
Entry pinned at publication, prices recorded at 1h, 4h and 24h, wins and losses both published. No hit rate until the sample is real.
Every signal-selling service eventually shows you a track record, and almost all of them are constructed after the fact. The pattern is familiar: the losses were 'not really signals', the entry was taken at a convenient price, and the sample starts wherever the run of good luck starts. We designed our measurement to make that kind of editing impossible for us to do quietly.
The rule is simple and it is already running. At the moment a signal publishes, we freeze its reference price and the score it published at. Those values never change afterwards. We then record the token's price at fixed offsets, one hour, four hours and twenty-four hours against that frozen entry, along with the best and worst points reached in between. The primary grade is the twenty-four hour move, because that is the horizon most of our patterns actually play out over.
Losses are published on exactly the same footing as wins. A signal that went nowhere, or went against the direction we published, appears in the log with its number attached. So does a signal whose measurement failed for a data reason, those are marked as such rather than silently dropped, because a missing row is the easiest place to hide a bad one.
The honest state of things today: the v2.0 engine has only recently started publishing, and the measured sample is still too small to say anything meaningful about. So we are not saying anything. There is no hit rate on this site, no precision figure, and no average return, because any number computed from a handful of signals is noise with a decimal point on it. The track record page shows what has been measured so far and tells you plainly that it is still collecting.
When the sample is large enough, the number goes up whatever it is. If it is unimpressive, that will be visible, and it should be, a track record you can only trust when it is flattering is not a track record. Until then, the thing worth judging us on is the verifiability of individual alerts: every one carries the transaction hash, and every one can be checked against the block it came from.