A polish Charter that was supposed to be cosmetic cleanup surfaced ten production gaps in six hours — including two features that had shipped to main and never functionally worked for ten days. The anti-pattern got named, the pattern got documented as fw-4.18.0, and the CLI helper got deferred on purpose.
Twenty-seven hours later, the manual discipline got two names: Pattern 1 — pre-declare SpecKit refresh, and Pattern 2 — post-close audit-driven Batch N.4. The canonical document, the telemetry schema, the CLI helpers — and one hour after that, a meta-pattern that reabsorbs both upward.
Seven consecutive Charters on a one-month-old plan in Sentinel, executing CHARTER-18 by hand with the feeling that any false step would bury six Charters of work. Five hours to canonize the manual discipline as upstream governance — without naming the pattern yet.
Six hours of session work with a capable agent and the framework's full onboarding loaded — and not once did Charters get suggested. Five minutes to apply them when asked directly. The asymmetry isn't about capability; it's about visibility.
Forty minutes of copy-paste per Charter on top of two hours of work. Discipline was working; the ritual was becoming the problem. How the Charter became a first-class CLI entity, and the architectural decision that left audit orchestration to the framework and the prompt-eval to anyone else.
Six Plans on paper, five in practice — Sentinel's first systematic experiment between April 25 and 28. The cycle that turned an artisanal pattern into Charters, the framework's bounded unit of work.