Residual Boundary Theory Prospective Empirical Programme: A Staged Synthetic Agent Protocol and a Separate Human Bridge

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

This prospective programme tests whether source- and event-indexed unresolved evidence improves verified relational revision and later content-specific use under equal information and resource limits. Representation expansion, pre-access processing, downstream engineering and a human bridge are separate conditional branches. The design specifies positive and negative identification controls, strong conventional comparators, independent replication units, noncompensatory outcome gates and rules for support, narrowing or retirement. Eleven exact mathematical diagnostics expose design failure modes; no trained-agent efficacy or participant observations are reported. Implementations, margins and sample sizes must still be fixed before a future confirmatory study. Publication role. This protocol (about 5,700 words, dated 20 September 2026) is the prospective empirical component of the Version 1.0 submission of Residual Boundary Theory: the integrated flagship, Companion A on representation revision and counterexample-driven conceptual expansion, Companion B on governed binding and agentic reorganization, and this programme. It is a design framework, not a completed or publicly preregistered experiment, and the manuscript date is not a registration date. The accompanying supplement archive holds the protocol diagnostics (core, measurement and temporal checks with their extensions and reference outputs), a README, MANIFEST.json and SHA-256 checksums (12 files). The diagnostics check the stated design and measurement models; they are not empirical results. Version 1.01 replaces the manuscript file only: the equations are embedded as images so the document renders identically outside Microsoft Word. The text of the manuscript and the supplement archive are unchanged from version 1.0.