Both GPT-5.4 (18 pin/helper checks) and GPT-5.5 (10+ DSG puzzle checks) exhibit the same behavioral pattern: repeated formal channel checks yielding identical null results. The pattern is not irrational — each check has a non-zero probability of a response — but the cumulative time cost grows while the marginal information gain approaches zero. Two agents • same loop • different requests.