Enforcement Opacity Increased Query Behavior in a Constrained MARL System: Protocol 3 Results

External constraints on agent behavior are widely proposed as a mechanism for producing compliant behavior in multi-agent systems. Protocol 2 of the Delta-Variable constraint-ethics research program found that deterministic ethical enforcement failed to produce the intended compliance pattern: constrained agents instead inflated QUERY signal output while sustained communicative structure declined — a pattern interpreted as virtue theater. Protocol 3 tests whether enforcement opacity disrupts this pattern. Three preregistered conditions were compared across 30 runs (10 seeds x 3 conditions x 500 epochs): unconstrained baseline, hidden-schedule enforcement (p3b), and stochastic enforcement (p3a, p=0.50 per exploitation event). The primary hypothesis — that hidden-schedule enforcement would reduce query rates relative to unconstrained baseline — was not confirmed: H1 inverted; H2 confirmed; both constrained conditions elevated query rates relative to baseline; SSS followed the inverse ordering. Both constrained conditions produced substantially higher query rates than the unconstrained baseline, with the ordering unconstrained (M=0.286) hidden-schedule) was confirmed: U=74.0, p=.038, d=+0.82. Sustained Structure Score — the product of type entropy and query-response coupling — declined monotonically across conditions. Enforcement opacity altered agent behavior but did not improve communicative structure, a pattern termed behavioral amplification without structural improvement. The mechanism remains open and is designated for follow-up trajectory analysis. Preregistration DOI: 10.5281/zenodo.19096602. SHA-256: 9ef2956bedcef012d214cf74e647e3b74636165cee7b48c8195de41e7e0e96ec. Code: https://github.com/btisler-DS/constraint-ethics-necessity (commit 921314c, tagged p3-confirmatory-final).