When Predictive Signals Fail: Why Some Time-Series Data Cannot Support Prediction

In this work, we show that predictive feasibility depends on three observable properties of the signal: the presence of structural behavior the consistency of that structure across observations and the predictability of the signal itself Across widely used benchmark datasets (IMS, C-MAPSS) and independent vibration data, we show that signals can be: predictable but not informative structurally active but inconsistent or consistent and suitable for prediction These cases reveal two fundamental failure modes: inconsistent structure, where patterns do not generalize non-informative signals, where prediction is possible but meaningless This leads to a key conclusion: Predictability is not equivalent to predictive usefulness. The results suggest that the primary limitation in predictive systems is not the model, but the signal itself.