Latent Article Discovery in Transcript-Sufficient Corpora: A preprint on governed next-paper generation in structured research systems

This preprint introduces latent article discovery as a formal research operation in transcript-sufficient corpora. In many research environments, archives contain not only completed papers but also partially articulated ideas, recurring conceptual patterns, and unrealized article-level distinctions. These latent objects are typically identified informally through editorial judgment rather than systematic analysis.This paper argues that, under specific structural conditions, next-paper discovery can be made methodologically tractable and governance-compatible.The paper studies the Reflexive Laboratory, a transcript-visible research system in which:the corpus is explicitly bounded,artifacts (papers, packages, registry entries) have stable identifiers,relationships are represented in an artifact graph,and candidate outputs are returned to structured review rather than silently accepted.Building on the bounded AutoResearch framework defined in P12, this paper isolates the discovery stage as a primary methodological object and formalizes it as a five-step operation: 1, corpus ingestion2, concept extraction3, relation mapping4, candidate generation5, governance return The paper contributes:a coverage-classification scheme (already covered / partially covered / not yet developed),a candidate ranking framework (novelty, independence, transcript support, drafting readiness),and a governance-return model that preserves the distinction between proposal, review, and canonical publication.The central claim is that the first meaningful threshold for automated or semi-automated research systems may not be autonomous publication, but the ability to generate credible, inspectable, and reviewable next-paper candidates from a structured corpus. This release is a preprint. It includes:the full manuscript,a claim-to-source trace,figure specifications,structured metadata,and supporting transcripts (.mht) that ground the discovery process. The package is designed to be transcript-sufficient, enabling readers to reconstruct how candidate article objects were derived from the underlying corpus.