xbuilderlab/cheat-on-content

Cheat on Content is a tool for content creators who want to stop guessing what will perform well and start treating every post like a tracked experiment. Instead of just publishing and hoping, you score and predict how a piece of content will do before it goes live. Then, a few days later, you check the actual results against your prediction. Over time, this builds a personalized formula for what works on your specific channel. The workflow follows a simple loop: you grade a script, make a blind prediction about its performance, publish it, and then run a retrospective a few days later using real data. The tool logs all of this. If you miss the mark three times in a row in the same way, it prompts you to update your scoring rubric. Importantly, any changes to your formula have to prove they are more accurate than the old one by re-scoring your history, so you can't quietly fool yourself into thinking you are improving when you are not. This is designed for individual creators, particularly those making videos or serialized content, who want a system that learns their unique style rather than offering generic advice. The README positions it as different from standard AI tools because it does not write scripts for you. Instead, it acts as a judge that remembers your past flops, your benchmark accounts, and your specific cadence, getting sharper the more you use it. Setting it up involves running an install script that links the tool into an AI coding agent like Claude Code or Codex. You initialize it in your content project folder, answer a few questions, and ideally import 5 to 10 samples from a benchmark account to give it an immediate baseline. From there, daily use is handled through simple text commands to score scripts, log shipments, and run retrospectives.