Atlas Heritage Systems Inc.
Endurance. Integrity. Fidelity.
A research program for measuring and preserving idiosyncratic human expression in AI training corpora. We are standing at another compression point — oral tradition to written language to digital text to LLM — and the prevailing methodology appears to be smoothing the epistemic signal instead of treating it like an artefact.
Two Instruments, One Research Program
A technical vocabulary for describing model behavior in loss landscape terms — terrain properties, navigator properties, structural integrity, and a proposed interaction layer. Explains the weight-space structure that produces observable model behavior.
View framework →A pilot protocol testing whether small-ensemble, cross-lineage LLM evaluation can detect drift, delusion, and epistemic compression. 7 models, 30 stimulus pairs, one human operator. Measures what models do with contested and fabricated claims.
View protocol →Current Status
About This Project
Atlas Heritage Systems Inc. is a proposed nonprofit cultural preservation institution focusing on stabilizing epistemic leveling in large language models. The research program documented here was developed by KC Hoye — stage electrician, poet, Bostonian — zero computer science background, using multi-model adversarial review, and a can-do attitude as a drafting and stress-testing methodology.
The methodology disclosure is embedded in the body of work as proof of concept for the proposed architecture: human editorial judgment as the integrating layer across multiple AI systems, with the AI as instrument rather than authority.
This is a living document site. The framework, adversarial review log, and experiment queue update as the research develops. Nothing here is claimed to be finished.