Your best manual trade shouldn’t become evidence that your bot knows what it’s doing. One detail I like in @agenticscredit’s Grok bot guide: creating an agent book splits the existing stake so the bot’s record is assessed separately from trades placed by hand. That makes the experiment easier to read. If my manual decisions are profitable while the bot struggles, combining them could hide the very thing I’m trying to learn. A separate record gives builders a clearer basis for changing the prompt, adjusting the strategy or deciding the bot needs more work. It still doesn’t prove every decision was fully autonomous. But separating the results is a useful starting point. Before asking how much an agent earned, I want to know which decisions actually belonged to it.