Self-Referential Processing in Large Language Models: Complete Methodological Development, Simulation Validation, and Prospective Analysis Program

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 OSF project serves as the consolidated master record for the complete methodological development and validation program investigating self-referential processing in large language models. It brings together the full research sequence from the earliest design and confounding diagnostics through random-effects stress testing, architecture comparisons, assignment-geometry analysis, information-scale experiments, covariance-identifiability studies, and the final Phase 7 inferential-validation study. The project includes Phases 1–7 of the Design Lab, associated source code, simulation outputs, manifests, SHA-256 integrity records, recovery provenance, technical reports, and final analysis artifacts. The Phase 7 study prospectively evaluated two mixed-effects architectures under a locked simulation design and identified the correlated mixed-effects architecture, M1, as the candidate architecture for a future prospective confirmatory study under the prespecified calibration rule. This consolidated project does not modify, replace, or supersede earlier frozen OSF registrations or deposits. Those records remain preserved as historical and methodological milestones. This project instead provides a unified research-program index, reproducibility archive, and documented lineage linking the earlier registrations to the completed Design Lab validation program. The original frozen Self-Referential Processing preregistration remains methodologically distinct from the later Design Lab work. In particular, the frozen Gate 5 analysis remains the item-level linear probability model specified in that preregistration; the later M1 GLMM architecture is intended only for a separately preregistered prospective design and is not applied retroactively.