@Dylan Sexton Honest answer, if you'd set it up the deliberate way you'd know, because you'd have built it by hand or with your AI. Here's the tell. Most people have some mix of custom instructions, a project with files dropped in, and the global memory feature, and that all looks like memory from the outside. It isn't, it's whatever happened to be in the window that day. That's why it's right some sessions and wrong others. What Miles and the guys here keep pointing at, and what I run, is plainer than that. A dedicated folder. A router file at the top that says where things live and which file to read for which job. Then markdown files underneath for the stuff that has to survive between sessions: how you want research written, what sources count, how a spreadsheet row should look. The agent reads the router first, every session, and goes from there. So the check is just: can you open a folder and point at the file that says how you want rows written? If yes you're set and the problem is somewhere else. If it's spread across settings and old chats, that's the thing to build this week. Where does yours live right now?