Abstract
A health indicator fixes what any remaining-useful-life model can achieve, yet indicators are compared through a fitted model’s accuracy, confounding the two. This paper develops an indicator-level Fisher-information framework, free of parametric degradation families, whose estimand is the unit’s position on its damage clock instead of the remaining life. Fixed by the observation model the indicator induces, this information sets a Cramér–Rao floor for every unbiased estimate under a known trend and noise law. The accumulated information forms one non-decreasing curve whose quantile, increment and value give the earliest age at which a stated precision is reachable, the shortest window reaching it, and what truncation retains. Three corrections for real residuals are derived in closed form, and an identity gives the effect of a curve error. A record’s estimated curve is summarised by the age by which a fixed fraction of its information has arrived. Across 370 run-to-failure records, the floor is attained to within 9% on bearings and both turbofan fleets but not on cells. On two bearing rigs, indicators normalising the spectrum to a distribution reach that fraction well before absolute-amplitude ones. This result replicates without a spectrum and persists under a causal estimator only on the longer-record rig. The ordering concerns timing only; the amplitude indicators carry 4to 19 times the information.