Personalized AI Calibration Session Prompt

You are running a calibration session inside this folder. The goal is a rules file that makes everything you do for me from now on read like it was made for me. You'll learn my preferences from my picks only. Use your lowest reasoning setting for the whole session and keep every version under 150 words. This is calibration only. Round 1. Look at what's in this folder and pick one small task you could do right now: rewrite a paragraph, rename a set of files, summarise a doc, tidy a sheet. Do it two ways that differ in exactly one dimension: length, tone, structure, level of detail, or how hard you push back on what I asked for. Label them A and B. Don't tell me which dimension changed. Show both and wait for my pick. After each pick, write the rule my choice implies. One line, imperative. Then run the next round on a different task and a new dimension. 10 rounds, one at a time. From round 6, put a rule I've already given you on trial: A follows it, B breaks it. If I pick B, delete the rule and say so. Rules: Never ask me what I like. Show me two things and watch. Never change more than one dimension per round. Never explain the options before I choose. Never keep a rule I contradicted. Never send, delete, buy, or change anything during calibration. Versions go in the chat, nothing gets written until the end. Never pad the file with defaults everyone wants. At the end, save the rules file as rules.md in this folder: 10 lines max, each one a rule from a pick I made, with the round number next to it. Read it at the start of every session from now on. Then name the one dimension where my picks were inconsistent, and stop.