Multi-Horizon Cost Prediction: Comparing Near-Term and Long-Term Execution Costs for the Same Computational Decision

Computational systems frequently make decisions whose costs extend beyond the immediate execution step. A scheduling action, resource allocation, caching decision, model configuration, or computational optimization may reduce near-term cost while increasing subsequent execution time, energy consumption, memory pressure, or resource contention. Conventional cost estimation often focuses on a single prediction horizon or aggregates multiple cost components into a single scalar estimate. Such estimates may fail to distinguish decisions that are locally efficient from those that remain efficient over longer execution periods. This paper proposes Multi-Horizon Cost Prediction (MHCP), a conceptual framework for estimating and comparing the near-term and long-term execution costs associated with the same computational decision. Given a current system state and a set of feasible actions, MHCP predicts action-conditioned cost trajectories over multiple horizons, quantifies prediction uncertainty, and selects a decision according to the relevant temporal objectives and resource constraints. The framework distinguishes immediate execution cost, accumulated intermediate cost, terminal-state cost, and prediction overhead. It supports direct multi-horizon prediction, uncertainty-aware cost estimation, horizon-dependent decision weighting, and optional feedback-based policy learning. A mathematical formulation, training objectives, inference algorithm, and experimental protocol are presented. The proposed framework is intended to be compatible with diverse computational environments, including cloud scheduling, machine-learning inference, workflow execution, and resource management. No experimental results are claimed in this paper. Its effectiveness and methodological distinctiveness remain hypotheses to be established through controlled comparisons with single-horizon prediction, conventional cost models, multi-step forecasting, and predictive scheduling methods.