I want to tell you what happened in my shop today.
It is not a flattering story about me. That is why it is worth telling.
Let me start with a small thing that would not leave me alone.
The Faust Baseline is a set of protocols. Rules of conduct for an AI working with a person. I have been building it every day for more than a year.
The header of my working file said twenty-one protocols.
The list inside that same file counted twenty-four.
Three off. In my own document.
Three is not a big number. Nobody had asked. Nobody had complained. I could have picked a number and moved on.
But a governance framework that cannot count itself is not much of a governance framework.
So I went after it.
And the first thing I did was wrong.
I treated it like arithmetic. I built a careful rule for how to count things. Hours of clean work on the wrong problem.
Here is what was actually going on.
I had two documents out in the world. One was the working file I load every morning. The other was the plain-language edition I built for buyers.
Both of them said twenty-one protocols.
They were not the same twenty-one.
Thirteen protocols showed up in both. Seven showed up only in the buyer’s copy. Eleven showed up only in the working file.
And one of them, a protocol I call CIMRP-1, was defined two different ways. In one document it governed memory. In the other it governed moral questions.
Same name. Two different jobs. Two documents. Both mine.
That is not a counting error. That is a split.
Now here is the part I have to be careful about, because the title of this post makes a claim and I do not want to overstate it.
The Baseline did not catch this by itself.
No protocol fired. Nothing rang a bell. What happened is that I went looking for a missing piece of text, and searching through old work sessions turned up the buyer’s edition, and there it sat.
Luck and a search. Not a system working.
I want that on the record, because if I let you believe my framework caught its own defect automatically, I would be doing the exact thing the framework exists to stop.
So what did the Baseline actually catch?
Me.
Once I saw it, every rule I had written for the last year pointed straight at my own chest.
No claim without evidence. Stop when the evidence ends. Name the gap, do not fill it. Do not smooth a contradiction. Tell the man before it costs him time.
I built those rules for an AI. They do not care who is standing in front of them.
Here is what I did about it.
First, I stopped declaring the count and started deriving it.
There is a rule now that decides what counts as a protocol, and it comes out of a rule I already had. If the Baseline cannot be tested, it cannot be trusted.
So if you can trigger a thing on its own, and it can fail on its own, and it carries its own consequence, it is a protocol. If it only fires inside another one, it is a sub-layer. Named, dated, fully in force. Just not counted.
The number computes itself now. Nobody has to argue about it again.
Second, I merged what was doubled.
I had three protocols standing on the same ground. Fifteen rules between them. Ten of those fifteen were the same rule said three different ways.
They are one protocol now, carrying nine rules.
Not one word of governed ground was given up. I did not shrink the framework. I stopped paying rent on three doors into the same room.
Third, and this one cost me something I liked.
I had a protocol for governing AI that acts on its own. Agents. AI that takes steps without stopping to ask after each one.
I had described it as a transmission. Five gears. A governor on the machine.
It sounded good. It was a claim about machinery I do not have.
I am one man in Kentucky with a document. There are companies shipping working code in that space right now. If somebody had walked up and asked me to show them the gearbox, I would have had nothing to show them.
So I rewrote it.
It is a standard now, not a machine. It names what the plumbing owes before an autonomous action runs. Confirm the scope. Verify the authority. Check whether it can be undone. Write it down. And leave no way around the gate.
Every word of that I can defend. And it puts me back on my own ground, which is writing the standard, not building the cage.
Fourth, I found two protocols I had lost.
They were sitting in the buyer’s edition and had fallen clean out of the working file. One governs precision of language. The other governs accountability for results.
That second one is the hardest rule I own.
It says process compliance is not the standard. If the output is wrong, the protocol was not met, no matter how clean the session ran.
And it fired on me the same day I put it in.
Because for six hours I followed every rule correctly and worked the wrong problem the whole time. Every protocol was satisfied. Not one of them was wrong about anything.
The work was still wrong.
Nothing in the old version could catch that. The new one can. That is the single best reason to call this a new edition.
So here is where it lands.
Twenty-three protocols. Four sub-layers. One hundred twenty-five rules. Codex 4.0.
Fewer seats than before. More ground covered. A count that computes itself.
One more thing, and then I am done.
Nobody had bought the buyer’s copy. Not one person. So nobody is walking around with the wrong list in their hands.
That is timing, not skill. I am not going to dress it up as anything else.
Now, why publish this instead of quietly fixing it.
Because I sell a standard for how an AI should conduct itself with a person. And a large piece of that standard is naming your own failure before somebody else finds it.
I would be a poor advertisement for my own work if I found a split in my own documents and swept it under the bench.
So there it is. I had two lists. They disagreed. I found it, I traced it, I fixed it, and I am telling you the same day.
That is the whole standard. Run on me.
The archive is dated. It has always been open. Go look.
Written with my AI partner | The Faust Baseline™ | intelligent-people.org
“If this post helped you understand AI better. Share it, a Word of mouth is the only algorithm nobody owns.”
Contact: micvicfaust@gmail.com
Post Library – Intelligent People Assume Nothing
Purchasing Page – Intelligent People Assume Nothing
© 2026 The Faust Baseline LLC | All Rights Reserved
