When you build fast with an LLM, the first thing you have to learn not to trust is the press release it writes about your own work. The model that wrote my code also wrote the prose describing it, and that prose came back impressive by default — every module a contribution, every script a system.
So before I published anything, I did the opposite of a launch. I took the forty-nine numbered novelty claims my portfolio had accumulated, ran each one through a textbook prior-art check and a literature search, and reclassified honestly. Forty-nine went in. Zero came out as confident novel contributions.
I published that number. It's a paper now — Zero of Forty-Nine.
I know how that reads. It reads like I just told the world my work isn't original. But that's not what the audit found, and it's not why I published it. The deflation is the credibility. When you've shown you'll knock down forty-nine of your own claims, the one that's left standing — a system-level discipline for keeping LLM-collaborative work honest — is worth something, precisely because you can see I didn't grade it on a curve.
This is the same habit the trade taught me. Twenty years as a Master Electrician on a federal medical campus: what's behind the wall has to do what the paperwork says, and the test that proves it isn't optional or flattering. You wire it, then you test it, then you energize it — in that order, every time. I just pointed the same order at code.
There's a print of a phononic test part sitting on my desk. I haven't taken the acoustic measurement yet. If it matches the model, the discipline produced an honest claim. If it doesn't, that's the strongest evidence I have that the discipline is at least falsifiable. Either outcome is information — which is the whole idea.
The three papers are on Zenodo. The code is open. If you want to argue with the audit, the audit is public.
→ Zero of Forty-Nine · the methodology · the experience report