From Retrieval to Defensible Decisions: A Protocol for Time-Bounded Evidence States

Retrieval systems can find information, but retrieval alone does not establish whether a claim is supported, current, independently validated, applicable to a specific use, or sufficient for a decision. This paper presents StructureEvidence, a temporal evidence protocol for converting real-world questions into versioned, bounded evidence states that can be inspected and reused by agents, consultants, and humans. The protocol separates event time from knowledge time; distinguishes claims, evidence, decisions, and unresolved unknowns; and represents each public claim with an explicit state, `as_of` boundary, supporting evidence, `does_not_support` constraints, provenance, and next observable. Incoming questions pass an intake gate and are matched as EXACT, ISOMORPHIC, PARTIAL, or NONE. Where public evidence is insufficient, the protocol identifies minimum missing evidence, proposes a candidate research depth, and can generate a bounded verification quote while requiring explicit human authorization. Paid research purchases a controlled process rather than a predetermined outcome, and new evidence is classified for publication eligibility before any append-only public state transition. We report a production implementation spanning three heterogeneous pilot domains, 26 machine-resolvable claims, a deterministic public resolver, a method contract, and a public change feed. The present work is a methods and systems contribution, not a comparative performance claim. An open replication and challenge kit is released to test discovery, claim matching, boundary preservation, freshness, partial-match behavior, and public/private evidence controls.