Answers are grounded in this paper and cite their sources.
A remote interface can conceal states whose differences become observable only after a particular sequence of requests. We study how such distinguishing sequences can be acquired through black-box experiments, retained as executable probes, and transferred between independent sessions with verifiable evidence. The formal model uses bounded-horizon traces of resettable deterministic Mealy machines, distinguishing observed separation from unverified equivalence. We establish witness soundness, the persistence of distinctions under sequence extension, a transfer-scope condition, and a query-cost break-even law for reusing a learned probe. An experimental program covers 180 synthetic protocol worlds with limited testing budgets, followed by 24 paired trials of eight simulated peers investigating 64 session pairs. In rare-handshake worlds, development-informed probe selection distinguishes all evaluated state pairs at a four-sequence budget, compared with 7.76% for random selection; this policy requires 468 preliminary probe executions per world. Propagation of signed distinguishing witnesses reduces remote calls by 50.08% on average when a program transfers between pairs, while pair-specific behavior reduces the benefit and increases communication. A new controlled prevalence sweep (1,440 worlds; 92,160 held-out session pairs) quantifies the transition from negative to strongly positive net savings once discovery cost is included. A localhost HTTP demonstration confirms signed-result delivery, rejection of a forged certificate, and replay deduplication. Together, these results characterize a scoped, auditable method for transferring tests of hidden behavior rather than transferring session-specific answers.
No takes yet. Share an insight, caveat, or question.
Luan Rosa (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: