Accuracy

How the follow-up reader is tested

Follow-up posts are the messages a channel sends after a signal: "close half", "move the stop to entry", "close now". We collected real posts from the channels we monitor, with the instruction each one gave and the trades that were open at that moment, and replay them in order through the copier. The number that matters most is the share of posts that would close a trade the channel did not ask to close.

The rules were written while looking at those posts, so figures on them are best-case. To test honestly we then took 789 management-looking posts the rules had never seen and had them labelled blind. There the reader went wrong far more often: 3.17% of posts would have closed a trade the channel did not ask to close, against 0.42% on the posts it was built on. Most mistakes were announcements ("ready to close some soon"), the trader describing his own trade ("I'm taking partials"), permissions ("you may take some profits") and closes meant for a setup the copier never took. We fixed those; the fixed reader's rate on new posts is what the live logs will show, and we will publish it.

What the copier never does on a follow-up

• Open a trade, add to one, widen a stop loss or remove it.
• Act when the post does not say which trade it means: it shows a notice instead.
• Act on instructions only the AI found: in our tests the AI alone was wrong in 18 of 33 such posts.
• Touch trades from other channels.

The signal score

For every signal the copier shows how random trades with the same stop and target distances fared on the same symbol in the previous 45 days, at similar volatility and time of day. A tight target with a wide stop wins often and can still lose money; the score makes that visible. It is a description of the levels, not a prediction of profit. Channel skill and market context did not improve it in our out-of-sample tests, so they are not part of it.