Dual Temporal Reasoning for Human-Centred Early Warning: An Extended BSPG Architecture with Historical-Analogue Confidence Modulation and FEELLLM Felt-State Briefs

Abstract

This paper introduces BSPG-I / FEELLLM, an extended Black Swan PreCog Generator architecture in which early recognition of risk and meaningful felt-state patterns is formalised as dual temporal reasoning: an endogenous longitudinal-memory axis (participant baseline, past self-reports, prior actions and outcomes) coupled to an exogenous historical-analogue retrieval axis (reduced from the Radinsky lineage into a governance-constrained confidence modifier), with a present-state felt interface provided by the FEELLLM instrument family. The system projects a bounded possibility space rather than a predicted future: outputs illuminate what is already in the evidence, not beyond it. Companion methods and systems-design paper to BSPG V8 (DOI 10.5281/zenodo.20225247). This version constitutes the pre-data methodological freeze preceding the preregistered BSPG-I / FEELLLM pretotyping study.