Coherence Without Independence: Inversion in LLM-Assisted Research

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).

A methodological note examining how LLM-assisted reasoning can generate highly coherent theoretical structures without providing independent empirical constraint. It identifies inversion as the central structural failure: the framework intended to be tested can become the conditioning context through which further reasoning is produced. The note distinguishes coherence from independence, analyses conditional completion and context conditioning, introduces purpose and evidence inversions, and proposes the frame-allegiance test and constraint-entry checkpoint as practical safeguards against self-reinforcing theoretical development.