094 design review, before human trial
The implementation preserves the candidate A/B/C starts and six-node geometry.
The hypothesis is a visible cause and effect: closing a stall opens a public shortcut, and the guests keep following their own wishes. It is not a demand for high score, a compulsory moral dilemma, or a claim about what people found fun.

All legal six-slot sequences were enumerated after exited guests were removed from legal guide targets and already closed stalls were removed from legal close targets. Counts: A 3453, B 4594, C 6609. Basic successful sequences: 1161, 1430, 1609.
Revenue, completed wishes, and sum of guests' exit times were reviewed separately. The maximum-revenue plan does not dominate all outcome triples: faster exits with other wish outcomes exist in every scenario. That is a mechanical alternative, not a recommendation to disregard wishes or a measure of human satisfaction. The specific last closure, use of public paths, who was guided, and farewell text are additional expressive outcomes; no total score assigns them a moral rank.

Guide-then-close and later-light policies can share some numeric outcomes; they still produce different paths and remarks. The search does not claim every named policy is Pareto-optimal. Representative plans are illustrative reproducible inputs, while exhaustive_results.json contains separate undominated numeric examples.

The six-slot target is intentionally brief. In every enumerated A/B/C sequence all guests are home by slot6. The generic post-six-slot walkout phase is therefore not reached from these starts; its no-extra-sales, pause, and auto-play contracts are covered by explicit boundary fixtures. This safety implementation is not advertised as extra content.

Human questions remain open: Was the shortcut readable as a consequence of closing? Did the last open stall matter to the player? Was guiding understandable without taking away the guest's wishes? What kind of ending did the empty square leave? No test or agent operation answers those questions for a human.
