How a ChatGPT Prompt Log Became a 43-Module AI Reasoning Framework
In 2023, I had a ChatGPT Project with a growing stack of files in the knowledgebase.
Not a framework. Just files.
Instructions I'd written. Patterns I'd noticed. Things I wanted the model to carry between sessions. I was running a one-person shop, no team to sanity-check the thinking. The only second opinion I had was the model itself.
The problem: the files weren't linked. ChatGPT pulled from them independently, so the output was inconsistent. Same question, different session, different answer.
So I started asking for improvements.
At some point I noticed a pattern. The way the files were being organized when things worked — there was a structure to it. Not one I'd designed. One that had emerged from what actually produced consistent output.
I had it generalize that pattern. Turn it into something reusable.
That became KnowledgeForge. The name made sense: it was forging the knowledgebase into something that held its shape.
Patterns became principles. Principles became modules. Modules accumulated.
By the time I migrated from ChatGPT to Claude in 2024, it had a name: KnowledgeForge. And it had stopped being a knowledge base. It had become a reasoning framework, a structured way of thinking through problems before asking the model to solve them.
The core idea: LLMs are good at execution. They're less reliable at knowing when to push back, which failure modes to watch for, or whether a question even needs a direct answer. KnowledgeForge fills that gap. Not by constraining the model, but by constraining the inputs the model gets. Garbage-in is a human problem, not a model problem.
By late 2025, it had 43 modules.
Forty-three.
Each one addressed a specific failure mode: decision classification, confidence calibration, adversarial verification, knowledge accretion. Not invented. Extracted from things that actually went wrong.
The 43-module version wasn't a framework. It was anxiety in Markdown.
I'll get to what happened to those 43 modules in a later post. The short version: I did something I should have done much earlier.
If you're building solo with AI, you're accumulating reasoning patterns whether you write them down or not. The question is whether they stay in your head, where you can't audit them, or somewhere you can actually iterate on them.
The knowledgebase was the right instinct. Forty-three modules was too far.
What's the version you've built? And where did you hit the ceiling on it?