Ordinal Memory for Synchrophasor Reconstruction and Generator Coherency Monitoring Under Explicit Measurement Constraints

Phasor measurement unit (PMU) streams in smart grids lose packets in transit, yet compact learning mechanisms for filling these gaps are often compared without separating reconstruction accuracy from the measurements each method may acquire. This study evaluates a finite-memory ordinal mechanism (FMOM) under three explicit measurement contracts: passive packet arrival, paid queries to fixed snapshots and a common downstream generator-coherency tracker. Nine signal families combine public field recordings, generated events and nonlinear generator simulations; four loss regimes and three masks per episode yield 684 matched passive cases across 14 implementations. Under 50% independent loss, the passive ordinal adapter improved on its predecessor, route–offset recalibration (ROR), in seven of nine families but attained no family-level minimum, ranking third of ten primary implementations behind rolling singular-value decomposition and median shift. In the active pilot, five families required at least as many paid measurements as direct reading, and anchored sorting produced comparable errors. In the coherency pipeline, the IEEE 39-bus mean adjusted Rand index rose from 0.8284 (ROR) to 0.8417, while median shift reached 0.8505. Ordinal memory therefore offers useful but restricted gains, and measurement access, fallback state and reference definitions should be reported when smart grid monitoring algorithms are compared.