Conventional predictive monitoring of space systems assumes stable causal mappings between early signals and later outcomes. This assumption holds in low-variety, well-bounded systems but breaks in the high-variety adaptive architectures where diagnostic attention is most needed. We present KA-Space, a framework grounded in information thermodynamics — distinct from physical heat-flow thermodynamics — that addresses a different problem: not outcome prediction, but detection of the degradation of the regime in which prediction itself remains reliable. The central claim is that structural coupling collapse between subsystems — measured by mutual information decay — precedes observable metric degradation and constitutes a thermodynamically interpretable early warning signal. We formalise this via Proposition 1 (a variational interpretation, not a fully proved theorem under measure-theoretic conditions), which maps the fused fCER signal to a variational upper bound on the local entropy production rate of the coupled system under weak-coupling and approximate separability assumptions. The C2 channel is grounded in the Parrondo–Horowitz–Sagawa information–thermodynamics inequality. An ablation study confirms that C2 provides the primary lead-time contribution (lead of 8.7 steps with full model vs. 1.3 steps without C2), while C1 and C3 contribute non-redundantly across failure scenarios. Synthetic proof-of-concept simulations across six failure scenarios and N=1 000 stochastic realisations provide distributional lead-time estimates. All results are preliminary; real-telemetry validation is required
Karimov et al. (Thu,) studied this question.