Systems operating under finite resources must selectively allocate computation while preserving organized behavior over time. Most contemporary computational architectures treat resource limitation primarily as an external scheduling or optimization constraint rather than as an intrinsic determinant of system dynamics. Here, we propose a viability-conditioned constraint framework in which system behavior is governed jointly by resource support and perturbation-responsive integrative capacity. The framework is motivated by a two-dimensional biological state space defined by metabolic support (M) and integrative complexity (C), translated into a substrate-independent artificial analog using variables representing finite resource support (Mₐ) and perturbation-responsive integrative complexity (Cₐ). Within this formulation, we introduce the Residue Agent, an architecture defined by five operational commitments: (i) a finite resource variable with an absorbing terminal boundary, (ii) a recurrent integrating core capable of sustaining perturbation-responsive dynamics, (iii) viability-weighted structural plasticity that persistently modifies recurrent transition structure without episodic reset, (iv) stochastic trajectory exploration, and (v) action selection conditioned on internal viability state. To evaluate these commitments, we implemented comparative-control simulations consisting of the Residue Agent, a reward-plasticity control, and a viability-reset control under matched recurrent dimensionality and perturbation procedures. Under matched environmental decay, both the Residue Agent and reward-plasticity control exhibited comparable regime-sensitive end-state structural divergence. However, the reward-plasticity control accumulated substantially greater update-driven structural modification across regimes, particularly under metabolic stress and integration disruption conditions, reflecting the absence of viability-dependent suppression of plasticity under reduced resource support. These comparative signatures remained preserved across moderate perturbations of principal architectural parameters. The framework is intended as a constructive model of viability-conditioned integration under finite resource constraints rather than as a theory of consciousness attribution.
Shivashanmugam et al. (Tue,) studied this question.