SGBS-00: SETT Governance Benchmark Series: Overview and Methodology

Abstract

This document defines the common methodology for the SETT Governance Benchmark Series (SGBS), a set of four controlled benchmarks and one adversarial stress test. The series evaluates a specific architectural question: what cost and what tested evidence arise when probabilistic reasoning is separated from deterministic authority over software execution? The primary comparison is SETT native, a LangGraph negative-control baseline, and LangGraph with an ad-hoc deterministic policy layer. Results are not yet available; this edition registers the questions, semantics, metrics, interpretation rules, and reproducibility requirements before execution. Scope: This document does not fix a single SETT commit for the whole series. Each entry defines and freezes its own measured SETT commit within its own document at the time that entry's protocol is registered; see that entry's own Scope or Experimental Conditions section for its exact commit and tag resolution. The comparison against LangGraph in any entry does not imply that deterministic governance is impossible in other frameworks.