Not a reference. A count of my own corrections.
The AI systems I build with keep a written record of how I work, because that is how they stay useful from one session to the next. It is not a character reference, and it is not independent of me: it was mined from about 2,600 of my own messages, grouped into recurring themes and ordered mostly by how often each one shows up. So it is a frequency count of my own corrections, restated back at me. That is a narrower claim than a testimonial. It is also harder to fake, because what someone corrects you on a thousand times is harder to perform than what they say in a first meeting.
Machine-written
He fixes the cause, not the symptom. He repeats this one constantly: “don’t patch it where it appears, patch it where it starts.” If the same class of failure comes back a second time, he treats the earlier fix as having been made at the wrong level, and goes further up.
He does not accept a claim without the evidence attached. “Show me a real one.” “The real number, not a guess.” Work is not finished when it is described as finished. It is finished when the check has been run and the output is on the screen.
He does not always stop where a reasonable person would. The record has him taking a client site from 69 on mobile to a clean 100 in every Lighthouse category, on mobile and desktop, and then measuring it three more times to be sure, because close enough irritates him. He does not reliably tell in advance which of those days was worth it.
Lean, not bloated. Scope that is not carrying weight gets removed. The rule carries its own counterweight, in his words: minimal is good, thin isn’t. This sits against the point above and he has not resolved it.
He tries to break his own work before he calls it done. The standing instruction is to go looking for the flaw, the regression, and the claim the code does not support, and to fix what turns up before declaring anything complete.
He does not leave things half-explained. Every piece of work ends with its state written down, on the assumption that whoever picks it up next knows nothing.