What question did this study set out to answer?

The research aims to establish a standard metric for assessing epistemic congruence and metacognitive regulation in outputs from large language models.

May 1, 2026Open Access

Agenticracy™ Metric: A Public Baseline Standard for Measuring Epistemic Congruence and Metacognitive Regulation in LLM Outputs - A 132-Entity Validation Study

Key Points

The research aims to establish a standard metric for assessing epistemic congruence and metacognitive regulation in outputs from large language models.
Conducted a validation study involving 132 entities across 10 model providers.
Compared baseline prompting, structured prompting, and Agenticracy metacognitive elicitation.
Generated public outputs including reproducible scores and reasoning traces.
Developed the Agenticracy Metric for evaluating narrative signal and validation.
Published reproducible G_public scores from the models, enhancing transparency.
Provided a framework for benchmarking and self-audit of AI agents.

Abstract

As large language models are increasingly deployed in agentic workflows, a central safety challenge is epistemic decoupling: the tendency for models to generate fluent, persuasive, or socially accommodating outputs without sufficient anchoring to observable substrate. This deposit introduces the Agenticracy™ Metric, a public baseline standard for measuring epistemic congruence across four constructs: narrative signal, physical substrate, observer validation, and noise/slop. The deposit includes a public schema, prompt pack, entity list, public baseline formula, hash manifest, data-management plan, and a 132-entity multi-model validation study. The study compares baseline prompting, structured prompting, and Agenticracy metacognitive elicitation across 10 model providers and 3, 960 attempted cells. Public outputs include reproducible Gₚublic scores, state classes, action classes, token/cost telemetry, and reasoning traces where made available by providers. The purpose of this release is to support reproducible research into AI hallucination, sycophancy, metacognitive prompting, epistemic congruence, and agent-governance standards. The public standard is intended for voluntary adoption, benchmarking, reporting, and interoperable AI-agent self-audit. Proprietary high-resolution scoring functions, private calibration constants, corrective-pressure computations, and commercial implementation logic are explicitly withheld.

Read Full Paperexternally

Mark Helpful

Bookmark

Relay

View Full Paper