Measure AI

Generative artificial intelligence systems are progressively becoming critical infrastructures.
They write, recommend, guide and assist decision-making in sensitive domains: healthcare, finance, law, education, public administration and autonomous systems.
Yet their real behavior in production remains largely unobserved.

Sovereignty begins with measurement

Why measure AI?

Traditional benchmarks generally evaluate models under predefined conditions and at a given point in time.
But a deployed generative service evolves over time. Its behavior may vary depending on context, load, infrastructure or interaction conditions.

NeoMundi Research develops a runtime measurement approach to make certain signals observable:

  • stability;
  • drift;
  • dispersion;
  • critical regimes;
  • informational density;
  • behavior under constraint.

To go further:

Learn more about the NeoMundi Observatory : https://github.com/neomundi-io/neomundi-ai-observatory
Theoretical framework : understand the scientific principles behind runtime measurement.
Public methodology : consult the protocols, data, interpretation boundaries and reproducibility rules.

A public and documented method

NeoMundi Research progressively publishes its protocols, methodological principles, datasets and the interpretation boundaries of its measurements.

This approach is based on a simple principle:
a measurement instrument must itself be observable, documented, contestable and reproducible.

NeoMundi Research measures, documents and publishes observable runtime signals.
The association does not certify the systems observed and does not replace the operators responsible for their governance.

Consult the publications and theoretical framework
Explore the data and protocols

Contribute to the analyses

NeoMundi Research is progressively opening its AI generative systems observation program to external contributors.

Researchers, engineers, data scientists, auditors, legal experts, scientific journalists and organizations can :

  • reproduce a measurement;
  • analyze a published dataset;
  • test the instrument on an independent corpus;
  • propose new questions;
  • challenge a metric or a protocol;
  • document methodological limitations.

Contributors retain their independence of analysis.

Discover the contributors
Submit a contribution

Advisory Committee

NeoMundi Research is progressively bringing together complementary profiles to challenge the method, protocols and published results.
Objective: strengthen the scientific, technical and institutional rigor of the generative AI observation program.

First members :

  • Joël Ignasse, scientific journalist for Sciences et Avenir and La Recherche. Author of Sur les traces du Nouveau T.rex (Eyrolles, 2025).
  • Cédric Chatelain, quality consultant and IRCA-certified auditor, specialized in auditing, validation and ISO frameworks. Based in Bern, Switzerland.
  • James Aull, Founder of ASRO™,  an independent governed-state witness and evidence layer for AI systems. Based in Twin Lake, Michigan, United States.

Test the instrument

NeoMundi ControlTower allows users to explore a runtime measurement approach applied to generative systems.

Test the instrument : https://controltower.neomundi.io

We do not proclaim. We measure.

Scroll to Top