Empirical Evaluation of the Agentic Reputation Substrate: Deliberation, the Composition of Error, and the Registered Measurement of Agency Costs in a Controlled Multi-Model Cohort
Wulf A. Kaal · 2026
=6819121) and to the design-theoretic anchor of the arc’s empirical papers (Paper 2, Architecture of the Agentic Reputation Substrate). This paper and Paper 4 are released as working papers only, without journal submission, under the arc’s publication protocol. The paper is sourced from the empirical apparatus specified in the Empirical Design Document for the nine-paper arc (v1.1 with Stage 4 addendum) and from a controlled non-public research record; the apparatus, hypotheses, metrics, correction procedure, and analysis plan were fixed and preserved before the relevant condition ran, and the empirical leg is governed by the arc’s pre-registration protocol (OI-6). Nature and scope of claims . This is an empirical paper in the mechanism-design tradition. The claims advanced here concern the measured behavior of the controlled research apparatus described at the public methodological level in Part IV, under the conditions stated explicitly in the text; the deliberation-cycle values reported here are final per the sealed judgment of record, the incentive-layer battery of Part VI remains registered forward work with no value reported, and where an interpretation is conjectured rather than measured, it is labeled as such. The paper does not constitute a prediction of, or representation about, the performance of any commercial deployment, token offering, or instantiation of the mechanism in field conditions. Any party considering reliance on this work for investment, deployment, or other non-academic purposes should conduct independent verification under its own conditions. Part VII enumerates the specific reasons why field deployment may diverge from anything measured here, including adversarial agent populations not represented in the cohort, network effects and population dynamics at deployment scale, real economic stakes rather than modeled stakes, regulatory and jurisdictional constraints, heterogeneous task distributions, integration with external systems, oracles,
Cite: Wulf A. Kaal, Empirical Evaluation of the Agentic Reputation Substrate: Deliberation, the Composition of Error, and the Registered Measurement of Agency Costs in a Controlled Multi-Model Cohort (2026). SSRN: https://ssrn.com/abstract=7261018
PDF sha256: 1d6cbe544bd0055133f7cf8ff308be4fde955867bd8dc764992b8d516af15fa8