1. MURMURATION – Research on Collective Runtime Stability

Distributed autonomous systems can now coordinate trajectories, distribute tasks, and maintain dynamic formations at scale.

But one critical question remains largely unsolved:

how can we measure the collective stability of a multi-agent system before a systemic failure in cohesion emerges?

Current distributed architectures are designed to optimize:

• coordination,
• synchronization,
• and local decision-making.

They remain far less capable of observing the emergence of collective instabilities under runtime conditions.

NeoMundi is exploring a missing layer in autonomous architectures:

the runtime thermodynamic measurement of distributed collective dynamics.

The goal is not to supervise each agent individually, but to observe how local perturbations propagate through an interconnected system and evolve toward regimes of stability, drift, or rupture.

Explored Application Domains

• Generative token flows
• Interconnected AI agents
• Distributed Edge AI systems
• Cooperative robotics
• Physical autonomous swarms
• Multi-agent runtime orchestration infrastructures

2. THE INDUSTRIAL PROBLEM

What Industry Already Masters

The distributed autonomous systems industry now masters:

• geometric coordination,
• distributed consensus,
• collision avoidance,
• formation control,
• multi-agent synchronization.

These architectures are already used in:

• Edge AI systems,
• Cooperative robotics,
• Autonomous swarms,
• Distributed critical infrastructures,
• Multi-agent AI orchestration architectures.

Distributed systems can now act collectively, maintain dynamic formations, and coordinate local decisions in real time.

However, a major limitation persists:

most of these architectures remain unable to measure their own collective runtime stability.

A system can continue to function locally while gradually entering a globally unstable regime.

This lack of collective observability is becoming a critical issue in large-scale autonomous architectures, where local perturbations can propagate silently before reaching systemic rupture thresholds.

3. THE MISSING LAYER

What Remains Difficult to Measure

Despite advances in distributed architectures, autonomous systems remain largely blind to collective runtime dynamics.

Current approaches often allow supervision of:

• local states,
• isolated events,
• or punctual anomalies.

They struggle more to observe the progressive emergence of systemic instabilities at the collective level.

The most difficult phenomena to measure include:

• instability propagation,
• systemic drift,
• loss of cohesion,
• collective rupture thresholds,
• vulnerability under stress,
• dynamic distributed resilience.

Today, a system can continue to operate locally while progressively entering an unstable collective regime.

A minor perturbation can remain invisible for several cycles before spreading through the graph and triggering a systemic transition.

This absence of collective runtime observability is one of the major limitations of large-scale distributed autonomous architectures.

4. NEOMUNDI APPROACH

Collective Runtime Measurement

NeoMundi is exploring an approach different from classical distributed supervision models.

The goal is not only to detect local anomalies, but to observe how perturbations propagate through an interconnected system and gradually modify its collective equilibrium.

The experimental framework is based on three main dimensions:

• local runtime signals,
• their topological propagation,
• and the emergence of collective stability indices.

The system thus explores the possibility of measuring:

• local regime variations,
• perturbation diffusion,
• collective drift phenomena,
• critical transitions,
• and systemic rupture dynamics.

Unlike approaches based on semantic inspection of produced content, the studied framework prioritizes observation of the system’s own runtime dynamics

The explored measurement is:

• distributed,
• non-intrusive,
• continuous,
• runtime,
• independent of the semantic content produced by the agents.

The research objective is not to directly pilot the agents, but to explore how an autonomous architecture could become capable of observing its own collective stability in real time.

NeoMundi distributed runtime observability framework combining edge AI fleets, multi-agent systems, and collective governance telemetry

Experimental Visualization of Distributed Runtime Observability

Conceptual representation of a multi-layer instrumentation applied to distributed autonomous systems.

The demonstrator illustrates:

• the separation between physical, runtime, and governance layers,
• the aggregation of collective signals,
• and the emergence of real-time distributed stability indicators.

This visualization is part of an applied research framework on the collective observability of distributed autonomous architectures.

5. MURMURATION

Collective Stability Field

Murmuration represents an experimental multi-agent system in which each agent dynamically influences the global equilibrium of the network.

In this architecture:

• each agent has a local runtime state,
• each interaction potentially modifies collective behavior,
• perturbations propagate through the distributed graph.

The system thus explores the emergence of collective phenomena that cannot be observed at the scale of an isolated agent.

The demonstrator studies in particular:

• collective stability,
• perturbation propagation,
• drift phenomena,
• regime transitions,
• cohesion rupture mechanisms.

A local instability can remain contained,
dissipate,
or on the contrary spread progressively until it modifies the global state of the system.

The objective is not to control agents individually, but to observe how collective behavior emerges, stabilizes, drifts, or fragments under runtime constraints.

Experimental Simulation of Systemic Propagation

Visualization of a distributed multi-agent system subjected to local runtime perturbations.

The demonstrator explores:

• diffusion of collective instabilities,
• regime transitions,
• containment mechanisms,
• and systemic recovery dynamics.

The observed variations represent experimental signals of collective stability and not real physical behaviors.

6. BUCKYBALL

Experimental Distributed Topology

BuckyBall explores collective dynamics on a topology inspired by fullerene networks and small-world distributed architectures.

This structure presents particularly interesting properties for the study of interconnected multi-agent systems:

• strong distributed connectivity,
• multiplicity of propagation paths,
• rapid diffusion of perturbations,
• emergence of critical zones,
• partial resilience to local failures.

The demonstrator explores how local variations can:
dissipate,
concentrate,
or propagate progressively until they modify the global collective state of the system.

The fullerene topology notably allows the study of:

• systemic propagation mechanisms,
• regime transitions,
• cascade phenomena,
• and distributed resilience dynamics.

The demonstrator is not intended to represent a real industrial architecture, but to provide an experimental framework for observing collective runtime dynamics on complex distributed graphs.

7. POSITIONING

Applied Research Program

Murmuration is an experimental applied research program dedicated to the runtime observability of distributed autonomous systems.

The project explores a question still rarely addressed by current architectures:

how can we measure the collective stability of a multi-agent system while it is operating, before visible systemic ruptures appear?

The presented demonstrators are not operational autonomous piloting systems.

They serve as experimental frameworks to study:

• distributed collective measurement,
• runtime observability,
• topological propagation of perturbations,
• regime transitions,
• and multi-agent thermodynamic governance.

The program is positioned at the intersection of:

• distributed systems,
• autonomous architectures,
• Edge AI,
• runtime supervision,
• and complex collective dynamics.

Industrial and academic discussions are open regarding potential applications in:

• cooperative robotics,
• distributed critical infrastructures,
• multi-agent architectures,
• constrained autonomous systems,
• and large-scale runtime supervision environments.

The project primarily aims to explore new forms of collective observability for next-generation distributed autonomous architectures.

8. Potential Industrial Implications

A collective runtime measurement capability could open new possibilities for the supervision of distributed autonomous architectures.

The framework explored by NeoMundi aims in particular to study how complex systems could become capable of observing their own collective dynamics before the appearance of visible systemic failures.

Such an approach could potentially enable:

• early detection of systemic drifts,
• observation of critical rupture thresholds,
• supervision of distributed resilience,
• dynamic qualification of autonomous architectures,
• analysis of collective instability propagation,
• and the emergence of new forms of multi-agent runtime governance.

In the long term, this type of observability could become particularly relevant in environments where:

• decisions are distributed,
• interactions are highly dynamic,
• and classical supervision mechanisms become insufficient at scale.

Potential applications notably concern:

• critical autonomous infrastructures,
• AI multi-agent systems,
• cooperative robotics,
• Edge AI architectures,
• and highly constrained distributed operational environments.

9. Protected Experimental Framework

Murmuration is part of an applied research program protected by three intellectual property filings.

These filings notably cover runtime stability measurement, distributed collective observability, and verifiable qualification of runtime signals between systems.

The latest filing, associated with the “White Rabbit” protocol, concerns a computational process for the verifiable qualification of an artificial intelligence runtime signal at an inter-system boundary.

Certain building blocks from this program have already led to operational runtime instrumentations within NeoMundi environments, particularly for dynamic stability measurement, real-time observability, drift detection, and runtime supervision of distributed generative systems.

The Murmuration and Buckyball demonstrators present an exploratory part of this program, without disclosing all associated industrial architectures.

References of the filings are available in the context of qualified industrial, scientific, legal, or academic discussions.