Thread Constellations as Internal Peer Review: Criticism, Repair, and Enactment in Transcript-Mediated AI-Assisted Research

Writing-intensive research using large language models (LLMs) can generate more plausible ideas, drafts, frames, objections, commands, examples, and possible article paths than any one thread or paper can responsibly absorb. This paper argues that transcript-mediated research using LLMs needs thread constellations that separate proposal, criticism, repair, quarantine, enactment, and publication before artifacts are stabilized. It introduces a compact formal kernel, C = (T, R, ρ, A, H, G, J), where threads, roles, role assignments, artifacts, handoffs, governance constraints, and human admissibility judgment jointly define a review-capable architecture. The paper distinguishes multi-thread drafting from internal peer-review-like architecture and bounds the claim: internal review does not replace external peer review, field recognition, or formal verification. Rather, it internalizes functional precursors of peer review, including adversarial reading, scope discipline, source checking, artifact repair, proof-status classification, and governance-visible acceptance. The motivating case is the Whittle Enterprise Theory relaunch, treated as process evidence for proposal, critique, repair, classification, and enactment. A bounded comparison with Terence Tao’s Lean-centered mathematical workflows clarifies the stronger architecture of decomposition, local checking, repair, coordination, and explicit acceptance rules.