Write It Down
Your way of working is either written down or it is reinvented every time.
A note. No research behind this one — no number, no case, no sourced claim. The rest of the series is built on evidence; this is not, and it says so here so you can decide whether to keep reading.
Ask a company how it makes decisions and you will get an org chart. That is not an answer to the question. The org chart says who is in the room. It says nothing about what happens once they are.
What happens, mostly, is this. Someone frames the problem the way they already see it. Someone senior reacts. A number gets quoted that nobody can trace. Somebody says let's take it offline. Two weeks later a decision exists, and if you ask three people in that room why, you get three answers and a shrug.
Six months after that, nobody can tell you what was rejected or what would make them revisit it. So the same argument runs again, with new people and no memory, and it costs the same as it did the first time.
That is not a people problem. Everyone in that room was competent. It is the absence of a written method, and the absence is invisible because nothing in the building is designed to notice it.
An unwritten method has no failure state. It cannot be wrong, because there is nothing to be wrong against. When a decision goes badly it gets attributed to conditions, or to the market, or quietly to a person, and the thing that actually failed — the way the call was made — is never on the table, because it was never on paper.
It also cannot be taught. A method that lives in someone's judgment leaves when they do, and the replacement starts over, and the organization relearns the same lesson at full price. Everyone has watched this happen. Almost nobody files it as a process failure.
And it cannot be delegated — which is the part that has started to matter this year.
Your team is already handing work to an assistant. Most teams are, whether or not anyone approved it.
An assistant with no method inherits yours. If your way of deciding is a meeting and a confident opinion, that is what comes back — faster, cleaner, better formatted, and just as unexamined. The model did not fail. It did exactly what it was asked, in the absence of anything telling it how the work is supposed to go.
This is the quiet version of the problem, and it is worse than the loud one. Bad work that looks bad gets caught. Bad work that arrives in good prose, on time, with confident structure, gets forwarded.
You cannot hand a way of working to a machine that you have never written down. That sentence is the whole argument.
The method I hand mine.
DEOS Open is the operating system I run my own practice on, published in full: standards for choosing a question, gathering evidence, reaching a verdict and handling it when the verdict is wrong. Free, and built to load straight into Claude or ChatGPT.
Get the standards Or read them on GitHub →I have been running a written operating system for my own practice for about two years. Standards for how a question gets chosen, how evidence is gathered, how a verdict is reached, what happens when it turns out to be wrong. Not theory — most of them exist because something broke and I wrote the rule that stops it happening twice.
Fifty-five of them are now public and free, at cultmath.com/deos. Text under CC BY 4.0, templates under MIT. Use them at work, use them with clients, credit the source.
They are built to be loaded into Claude or ChatGPT as a project, which is the use I would start with. You are not writing a better prompt. You are giving the assistant a way of working, so it registers the question and what would prove it wrong before it goes looking, and tells you what would change the answer.
I will say the obvious thing about this: it is one practice, mine, and I have no controlled comparison showing teams that installed it beating teams that did not. What I have is two years of standards that each exist because of a specific failure, and a register of those failures that is not empty. Make of that what you want to.
Fifty-five is a library, not an assignment. Installing all of it is how this dies in week two.
Take three. Write down what you decided and why, before you forget. Write down what you expect to find and what would prove you wrong, before you look. Close the week by reading back what you wrote. That is the loop, and the other fifty-two are refinements on it.
The point is not the folder. The point is that in ninety days you will be able to answer a question you cannot answer today, which is how your organization actually decides — and once that is written down, it becomes a thing you can improve, teach, hand over, and argue with.
An unwritten method is not a flexible method. It is just one you cannot see.
Related from Sound Decisions: Hope Is Not a Strategy · Nobody Pays You Extra for Being Honest · Build It Anyway
Three standards, ninety days.
Download the folder, load the Core into whichever assistant your team already uses, and let the records accumulate. Ninety days from now you will be able to answer a question you cannot answer today, which is how your organization actually decides.
Download DEOS Open Or browse the standards →