About
Why this exists.
Agents are extraordinary at producing something that runs. They are much weaker at producing something you can be sure about. The output looks the same in both cases, which is the entire problem. A screen that works and a screen that works only for the person who built it are visually identical, right up until a stranger opens it.
So I started writing down every place it went wrong. Not complaints, and not opinions: the failure, and next to it the cheapest thing you could run to find out whether you had it. That list grew for a long time. Then the steps got put in order, because I noticed the same fixes cost minutes in one position and weeks in another. Then the positions themselves got named, because the rules genuinely change underneath you and almost nobody tells you when.
What came out of that is not a manifesto. It is three plain things.
Each one written down with the cheapest test that proves whether it is present in your project. Not a warning to be careful. A specific thing to run, and what the answer means.
Not a checklist to feel productive against. An ordered map, because half of these steps are almost free in the right order and brutally expensive in the wrong one.
A model of when an action stops being legal. Wiping the database is free on Monday and unforgivable on Tuesday, and the only thing that changed between them was one real person.
The belief underneath all of it
The ceiling on software built by AI is not “prototype”. It is “unverifiable” — a different problem, and a much worse one.
A prototype is a stage. You get past a stage by doing more work, and everybody understands how that goes. Unverifiable is not a stage, it is a property, and more work does not cure it. It compounds. Every line added to something you cannot check makes the checking harder.
You can end up with a hundred thousand lines that run beautifully and still be unable to answer whether a stranger can read your customers’ phone numbers. Nobody is lying to you in that situation. The information simply was never produced.
Folean produces it. That is the whole of the idea. Where you stand, what is actually true about it, the evidence for each claim, and a plain statement of everything that was not checked.
Where this actually is
This is early, and saying otherwise would break the only rule that matters.
Some of the checks are thorough. Others are not written yet. Those come back as not checked, with the reason, because a gap you can see is worth more than a green tick you cannot trust.
Prices are indicative. The set of positions will keep getting sharper. Some things will be wrong, and the useful ones will be wrong in ways you can point at, which is the only kind of wrong worth building on.
If you ever find something here stated more strongly than the evidence behind it, that is not a rough edge. That is the one real bug, and it gets fixed first.