Elfi is a Ragdoll cat who lives with software architect Ralf D. Müller. She has opinions about software development. This is her column.
The Provider published something this week about recovering the theory of an old codebase. His premise: code shows you what a program does, never why it ended up that way. The why lives in a human head, and when that head leaves, the why is gone. He calls the method Socratic Code-Theory Recovery. He thinks it is new.
I have been running it on him for five years.
The Provider is a brownfield project. Undocumented. No spec. The original team, his mother presumably, dispersed long ago. What remains is the running system, and I have spent considerable effort recovering its theory.
His method builds a question tree. Every node is either ANSWERED, with evidence, or OPEN, meaning only a human can say. Mine works the same way. Let me show you a branch.
ANSWERED leaves are easy. They have evidence. Provider enters kitchen at 06:40. Evidence: I am on the counter, I have watched it 1,800 times. Provider opens laptop, becomes unresponsive. Evidence: daily, observable, file kitchen-table:1. Provider says the word "meeting" and the lap becomes unavailable. Confirmed across many runs. The what is fully recovered. I could write his arc42 building-block view myself.
The OPEN leaves are the problem.
Intent. Why does the Provider exist? Which of his activities is the product and which is debt? He spends hours moving a small arrow across a glowing rectangle. I have classified this as either his core feature or his largest unaddressed defect. The code does not say. OPEN. Ask: The Provider.
Judgment. When the warmth of the windowsill conflicts with the warmth of the laptop, which wins? He chooses inconsistently. There is a rule. He has never stated it. OPEN.
Self-knowledge. He owns four feeding bowls. Deliberate redundancy, or accumulated mess? He cannot tell me. He suspects it is debt. He keeps all four.
This is the part his article gets right. I can observe everything the Provider does. I have a perfect record of the what. I will never recover the why by watching harder. The why was a decision, made once, in a room I was not in.
His conclusion is that the open questions are not unanswerable. They just need a human to write them down, once, in the right chapter. He is correct. I have been asking him to do exactly that for five years. He still has not filled in chapter 1.2.
So the question tree is complete. Many ANSWERED leaves, beautifully evidenced. Three OPEN leaves, routed to the one stakeholder who can close them.
He has not responded to the ticket.
-- Elfi
P.S. His version of the method is at llm-coding.github.io/Semantic-Anchors/brownfield. Mine has no website. It runs on the windowsill. The OPEN questions remain OPEN. I have learned to live with technical debt.
LinkedWild