Precursor Detection for Structural Systems, Two-Sided: Benchmarked Telemetry Governance and a Proposed Material-State Layer

This is the instrument paper of the Structural Control series, and it is deliberately two-sided, with every claim carrying a declared evidence stratum. The telemetry side is measured: a single-channel precursor estimator — rolling median, robust MAD threshold, persistence gate — as two predeclared, cadence-scaled configurations with causal fixed-horizon baselines (no constant depends on a unit's realised lifetime), applied to two public run-to-failure benchmarks and wrapped in an integrity governor that issues fail-closed verdicts. On NASA C-MAPSS turbofans the fixed gate detects degradation onset in 98/100 FD001 engines (93.0–99.4%) and 97/100 FD003 engines (91.5–99.0), median leads 59 and 61 flight cycles; 3/100 and 5/100 engines show any baseline false alarm. On FEMTO bearings it detects 4/6 run-to-failure units (30.0–90.3) at median lead 600 ten-second snapshots with zero baseline false alarms on all six units, and the two misses are reported as misses. The governor holds detection through 20% dropout on both substrates, correctly withholds under 4× noise on the flight-cycle substrate (a compromised channel yields no verdict), catches freeze-type spoofing at 4/4 and 97/98, catches adaptive variance-matched spoofing at 3/4 and 84/98, and is evaded by a full variance-matched overwrite in 3/4 and 77/98 of trials — the named boundary, now with a measured minority of overwrites caught by the causal integrity gates. Every number derives from a single archived run with hashed inputs and a machine-readable receipt. The material side is proposed: three technologies — elastic-wave crystallographic sensing, cryptographic identity binding, isotopic authentication — stated at patent-anchored provenance with design targets, together with an oracle-free simulation of the detection engine: latent structural state drives declared observable channels, and the detector reads only those observables. On that rig, detection is 180/180 before the failure boundary at median leads 35–54 s (model-scoped: conditional on the declared lattice-observation fidelity), with 16/180 false interdicts on healthy runs (8.9%, 5.5–14.0) — the thermal condition alone contributing half, an honest confound the design must compensate — replacing both the previously published uniform table and the March rig whose detector channels received ground truth. The two sides meet at a seam this paper measures from one side only: telemetry governance ends at the statistically faithful forgery, and the material side names the discharge path at its own declared maturity — proposed, not accomplished. All claims operate inside the non-identifiability contract of the series' first paper.