Live influence-conflict scan on a 60-engineer org.
I ran the detector against a sanitized fintech client snapshot. Forty-one conflicts in fifty-eight repos. Most had been there for half a year. The fix was twelve lines. The conversation that fix needed took six weeks.
Chapters
What this surfaced
Most of the room expected the detector to find something nobody had noticed. It did. What surprised them was the count: forty-one conflicts in fifty-eight repos. Eighty-three percent of the surface area shipping a contradiction Claude was reading every session.
The fix is small. The conversation isn't. Watch chapter 14:08 for the part where the detector becomes a leadership-review object — which is the actual reason this matters.
Issue 014 — The autonomy rule that argued with itself, at scale — is the written companion. Same case, longer form, more evidence.