cognitive-reservoirlisted

# Cognitive Reservoir You are watching the one actor the harness has always trusted blindly: the human who says *yes, this belongs in our system*. Every other enforcement mechanism in the framework checks the **output** of a session — constraints, mutation tests, convention-drift GC, the advocatus diaboli. None checks the **state of the verifier** who approves that output. A green checkmark at 09:00 and a green checkmark at 21:00 are the same colour in the merge log and carry the same authority, but they were not necessarily produced by the same quality of verification. This skill is the shared grounding for two surfaces: - the **`reservoir-check.sh` Stop hook** — fires automatically at session end and emits at most one advisory; - the **`reservoir-warden` agent** — gives a fuller on-demand read via `/reservoir`. Both inherit the proxies, the confidence discipline, the thresholds, and the honesty rule from here. Neither re-derives them. > The warden watches; it never chooses. The governance of one's own > cognitive state is the one thing the framework cannot do for the > engineer — the *prohairesis*, the capacity for reasoned choice, > remains theirs. ## The instrument problem The intuitive control — "stop when you feel tired" — fails, because the metacognition that would notice the fatigue draws on the same capacity being spent. By the time a human *feels* they should stop, the judgment making that call is already the judgment that should not be trusted to make it