ChatGPT disclosed things in this session that I did not expect it to disclose, and there is no guarantee it stays available. I would rather people had time to use it than had a tidier version of it later.
So: every word of every ChatGPT response here is verbatim, and that part is checked automatically on every build. What is not finished is the presentation. The color coding on the code blocks is incomplete and some of it is imprecise, and there are notes to myself still sitting in the page.
Parts 1 and 2 are close to empty and are getting a lot of detail over the next week or so, along with a cleanup pass on everything else. Worth checking back.
How to use this guide
How to Read This Guide
The layout, the reading order, what is verbatim from the session and what we wrote.
This resource is built around a real chat session, kept whole. Reading it takes a little orientation, because the interesting part is not only what ChatGPT said but which parts of it can be trusted.
Every page is one turn
Part 4 and Part 5 follow the same layout. A context box we wrote, then a short summary of the response, then the prompt in a darker box, then the response in a lighter box.
The context box and the summary are always labeled. They are the only text on those pages that we wrote. Everything below them is unedited.
Everything is verbatim
The prompts were dictated, so a tightened rewrite is what you see, labeled as one. The original of every prompt is preserved in the page source.
The responses are untouched, including the places where ChatGPT contradicts something it said earlier, and the places where the source text is garbled. That is checked automatically on every build: the response text on every page is compared word by word against the original transcript.
The code blocks are colored
ChatGPT shows a lot of JSON in this conversation. Some of it is what it actually sent and received. Some of it it made up on the spot to illustrate a shape. The two look identical on the page.
So every code block carries a color and a label. Green means the values are real. Blue means they are invented. The color key has the full breakdown, and it is worth reading before Part 4.
The order
Parts 1 and 2 are about the prompt itself, and use a test where one irrelevant word changes the whole answer. Start here if you want the short version of why this matters.
Part 3 is the practical response to Parts 1 and 2. What a prompt has to contain before the result means anything.
Part 4 is the transcript, eleven turns. It builds, so it reads best in order. If you only have time for one page, read the live tool tests.
Part 5 runs the whole pipeline on five commercial questions. Skip to whichever is closest to your business.
Part 6 is the prompt that reproduces all of it on your own category.
What this is not
This is not a leak. Nothing here came from inside OpenAI. It is one model reporting on its own tool interface, pushed hard on the difference between what it can observe and what it is guessing, and then made to demonstrate rather than describe.
That distinction is why the colors exist. Read them as the difference between evidence and a story.