What problem does it solve? Acceptance criteria often degrade into decorative green checks: scores without measurement methods, gates that never proved they can fail, and validators that fire on the desired state until they become ignored noise. This Skill designs controls that are observable, falsifiable, and satisfiable from birth. ## Core Features & Use Cases - METHOD and signal per criterion: Every control declares an observable procedure and a countable signal, with honest proxies (e.g., "acronyms used before definition: 3") instead of invented scores like "Clarity: 8/10". - Falsification testing: Each control's falsification_test is exercised before "Passed" means anything; until then the state is "Not executed", never a decorative green. - Birth satisfiability: New controls must pass the current corpus, be a regression against a recorded baseline, or carry an explicit debt budget that only decreases. - Deterministic validation: The control set is validated with validate_graph_spec.py --controls from graph-spec-compiler until zero findings. - Use Case: When defining acceptance gates for a graph pipeline, convert vague quality judgments into measurable controls, prove each gate can fail, and validate the whole set with the checker. ## Quick Start Ask the assistant to design falsifiable acceptance controls for your graph or harness, with an observable method and failure test for each criterion and no invented scores.
