Preprint
Article

This version is not peer-reviewed.

Learning from Prediction Errors in an HLbC and IGTC Architecture

Submitted:

20 November 2026

Posted:

20 November 2026

You are already at the latest version

Abstract
We study how a system can use recorded forecasts and observed outcomes to introduce testable predictive distinctions in a framework based on Human Language-based Consciousness (HLbC) and the Information-Geometric Theory of Creativity (IGTC). In a six-input Boolean benchmark with fixed search and observation budgets, ranking rules by signed prediction errors modestly reduced forecast loss relative to counting previously selected rules when the historical rule frequencies persisted, while retaining the original forecast gave the lowest loss under uniform or reversed target frequencies. For covariance transfer, we derive separate identification criteria for the calibrated mean and the conditional distribution under a Gaussian observation model and show in a controlled example that choosing informative historical measurements can lower prediction error when the covariance contrast is large enough relative to estimation error. In 14 held-out sessions from one permanent-magnet synchronous motor, projection preserved the estimated historical variances under artificial additive shifts, and the strongest simple correction of the neural forecast achieved lower temperature error with a single reading.
Keywords: 
;  ;  ;  ;  ;  
Copyright: This open access article is published under a Creative Commons CC BY 4.0 license, which permit the free download, distribution, and reuse, provided that the author and preprint are cited in any reuse.