Title: Adaptive Inference for Sequential Pricing Under Resource Constraints

Abstract:

In scenarios where pricing controllers face strict resource limitations, executing inference for a fixed price may become unfeasible. Even if every realized action possesses a known positive density, the controller’s resource state can eliminate the neighborhood surrounding the target price from the set of feasible options. This phenomenon, termed support-exclusion failure, is formalized through a local non-identification result and the introduction of a realized information clock.

To address this, we propose a target-aware pricing controller that certifies feasible target bands and records continuous local densities. By applying localized debiasing, we derive studentized intervals, the width of which is determined by the aforementioned information clock. The resulting regret–information accounting, accurate up to pilot re-solving error, demonstrates that inexpensive exploration may not suffice for effective inference. Specifically, while a polynomial target mass yields polynomial rates, a pure $1/t$ target branch fails to produce shrinking fixed-target intervals without supplementary local movement. Experimental results confirm calibration within the certified bands and highlight the system’s ability to abstain from inference when the resource state causes a collapse in target support.

Source: arXiv Generated at: 2026-06-03 00:00:00 UTC