CALLIOPE · BCOM

This short note supplies the formal mathematical vocabulary used in the Perspective papers WP0161 (Pattern, Persist!) and WP0162 (the Nature condensation thereof). A pattern is not a material object inside the world: it is a useful emergent compressive regularity revealed by an observer-relative projection of an evolving algorithmic soup. Persistence is the temporal conservation of projected algorithmic structure, measured by normalized mutual algorithmic information between projected descriptions at different times, conditional on the background dynamics of the soup. This observer-relative formulation absorbs into a single operator the distinctions between single instances, lineages, collectives, ecosystems and planetary systems. Active persistence and the telehomeostatic objective follow as specializations. The note is intended as a citation target for KT papers that require the formal substrate without rehearsing it.

: pattern, persistence, projection, coarse-graining, Kolmogorov complexity, mutual algorithmic information, telehomeostasis, algorithmic agent.

A compact mathematical toolkit for saying precisely what a "pattern" is and when it "persists" — without smuggling in hidden metaphysics.

The core move is simple but powerful. Instead of treating patterns as things that exist out in the world independently of any observer, the paper defines a pattern as what you get when you apply a projection — a computable, lossy map — to an "algorithmic soup" (any computable dynamical system: a cell, an ecosystem, a Turing machine, a civilization). The projection collapses the full microscopic state into a compressed description. That compressed description is a pattern candidate if it satisfies five simultaneous constraints: it must be genuinely compressive (simpler than the raw state), non-vacuous (not just a constant), intelligible (the projection rule itself can't be arbitrarily complex), useful (it helps an observer predict or act), and persistent (it stays self-similar over time).

Persistence is the technical heart. The paper measures it using normalized mutual algorithmic information — essentially, how much algorithmic structure the projected description at time t shares with the projected description at time t+τ, after subtracting what's already explained by the background dynamics of the soup. A value near 1 means the pattern is faithfully conserved; near 0 means it has dissolved. Crucially, the conditioning on background dynamics B₀ prevents you from counting trivial persistence — the fact that everything in the universe obeys the same physics doesn't make everything a pattern.

One elegant payoff: the single projection operator ρ absorbs distinctions that usually require separate ontological categories. Whether a pattern is one organism, a species, a lineage, a culture, or a biosphere is not fixed in advance — it's a choice of projection. The framework doesn't need separate foundational slots for "type" versus "token" or "individual" versus "collective." This is genuinely unifying rather than just terminologically convenient.

Agency then falls out as a special case. A persistent pattern becomes agent-like when its projected structure is not only self-similar over time but actionable for itself — when it supports model-based regulation of its own continuation. The paper formalizes this as the "telehomeostatic objective": a discounted sum of expected future persistence scores, which an evolved or engineered objective function can approximate without explicitly representing the formula. The paper is honest that this is a formal companion note, not a standalone argument — it exists so that WP0161 and WP0162 can cite precise definitions rather than re-derive them each time.