docs/

revised Sept 7, 2026

• 5 rules in the working model
• 4 case studies
• every one from a real session

What I delegate, what I never delegate, and how I verify, carried by four builds where the record shows me being wrong about something. A force push that would have published 47 files out of branches nobody was watching. A feature I specified and overruled an hour later. An algorithm I named in my own spec and had to kill. The model is only worth reading because the case studies are the ones where it cost me something.

last audited Sept 5, 2026

• 9 in daily use
• 2 just adopted
• 5 evaluated and skipped

What I actually use, what I just picked up, and what I turned down with the reason attached. The skipped list is the one that says anything, because a tool that demos well and fails on the specific work is the normal case. It carries a date because a tooling list with no date is a claim about the present that quietly becomes a lie.