Can an algorithm observe and interact with another algorithm as an operational object? This paper traces the genealogy from Observability of the Observer to the first experimental investigation of inter-algorithmic observability — including the methodological failure that transformed an initial experiment into a comparative research programme.
Frontier AI progress is no longer paced only by capability. OpenAI’s pause in frontier RL training signals a deeper transition: model development is now conditioned by the capacity to observe, monitor, align, and contain emergent cyber-critical capabilities. In this shift, the observer enters the causal chain of AI progress.
This provisional edition establishes a dated public record of the framework while the full manuscript remains under development. Its purpose is to preserve conceptual priority, make the emerging ontology publicly observable, and provide a stable reference for subsequent audit, refinement, and formal publication.
When AI surprises its observers, the unexpected behaviour reveals more than the system itself. It exposes assumptions, blind spots and limits within the frameworks used to interpret it. The Observability Factor begins where astonishment stops: by treating perplexity as data about the observer.