ATSA — Testable Architecture for Adaptive Systems / Arquitectura Testeable para Sistemas Adaptativos

This document presents ATSA (Testable Architecture for Adaptive Systems), an engineering framework based on the [Q↔F]→A architecture. The model describes a cycle in which an adaptive system generates variability (Q), filters the available possibilities according to its accumulated history (F), and selectively amplifies certain outcomes (A), whose feedback subsequently modifies the state of the filter. ATSA formalizes these components and distinguishes between the classical branching ratio (σ_B), used in discrete-event systems, and an operational gain metric (G) proposed for continuous-activation systems. The document presents a minimal computational model that allows both magnitudes to be calculated and examines their relationship to the system's amplification regime. Building on this formalization, quantitative and falsifiable predictions are proposed for continual learning systems, including the possible relationship between the amplification regime and catastrophic forgetting, as well as between the separation of historical adaptation timescales and performance during sequential learning. The goal is to establish an explicit architectural framework that can be implemented, compared against existing architectures, and subjected to experimental validation.