Small contracts, measurable limits: mathematical foundations for the MCRP seed release

Abstract

A protocol for solo researchers, collaborations, individual agents and agent teams should expose checkable boundary information without requiring participants to understand everyone else's internal complexity. Four models identify what the offer-check-rely-amend interface can and cannot guarantee. Group-first sampling removes a representation-multiplicity incentive under a fixed known control map. Selective completion can reverse offered-panel bounds. A common weighted repair envelope bounds expected cascades under changing regimes, while stable snapshots and finite mean work do not certify safe queues. One-shot auditing must satisfy effort incentives, participation and its funding constraint simultaneously. This revision disambiguates audit costs and work charges, adds an exact finite-history adaptive retry ledger, and tests sensitivity to declared panel weights. A worked adaptive policy shows that fewer attempts can cost more and complete less than a stipulated IID comparator; this is not a performance improvement claim. The audit analysis makes information-set, concealment, funding and enforcement obligations explicit. Bounded repair-witness searches and heterogeneous-cost audit lotteries retain their conditional guarantees. A finite-workload supplement illustrates scarce human attention, challenge floods and screening failure without inferring global saturation from publication rate limits. The contribution aligns established mathematical tools with versioned protocol records; it does not claim new underlying mathematics or a composed safety theorem. Exact calculations, finite enumerations and deliberate counterexamples remain reproducible. All development and internal adversarial review were author-directed AI work, not external replication. These synthetic checks do not establish behavioral calibration, measured human savings or field efficacy. Dissemination, performed checks and consumer-specific reliance remain separate from qualified human scientific endorsement. Identity authentication, institutional enforcement and actual reviewer behavior remain external obligations.

Keywords