The Finite-Calibration Limit: What a Device Characterisation Can Claim Between Calibrations

Quantum control practice runs on characterisations: measured coherence times, gate errors, frequencies, and crosstalk figures, taken at calibration and consumed by every layer of the control stack until the next calibration replaces them. A premise rides silently in that consumption — that the characterisation remains current between calibrations — and this paper examines and refuses it. We argue three points, each held at its own strength. First, a characterisation is a finite observer's estimate, not a standing fact: at calibration time it already carries a strictly positive uncertainty floor from finite probes, and the act of probing is itself a physical intervention on the device. Second, between calibrations the device's parameters move — relaxation times fluctuate on hour scales through two-level-system dynamics, and environmental drift reshapes the error landscape — so the uncertainty a consumer should attach to a characterisation widens with time since calibration under any honest model. Third, these two facts compose into a limit with an operational consequence: under the declared drift model a characterisation carries a finite validity horizon, and in every case it carries expiry semantics — a control stack that consumes a characterisation without an expiry is acting on an unexamined exactness claim. The finite horizon is conditional and the finite-calibration limit section states its converse: a mean-reverting drift confined within tolerance admits an infinite horizon, and nothing here forbids that case. We then instantiate the external contract of the companion limits paper for this setting: a characterisation ships with a declared validity budget and a declared expiry model; consumers check validity before acting on it; and on expiry the fail-closed forms are refusal of the stale figure and escalation to recalibration with a declared deadline. We state a minimal stochastic validity model — latent device state, calibration posterior, hazard-widening between calibrations, derived expiry rule — as a declared model whose parameters an installation must estimate, not a law derived here. The authority boundary of the resulting layer is advisory by design choice, and we give the rationale rather than presenting the choice as a consequence. Related work has built the detection and tracking instruments this contract would consume; what is added here is the register those instruments lack: validity as a first-class, expiring, refusable claim.