Metric-Bearing Output and Metric-Mimicking Output are terms proposed by Lee Sharks for distinguishing measurement from the appearance of measurement in AI-generated text.
Metric-bearing output exposes the units, numerator, denominator, sources, calculation, and uncertainty used to produce a score. Metric-mimicking output may repeat a correct formula while fabricating or omitting the evidence needed to apply it.
The packet applies the distinction to Provenance Erasure Rate. A numerical PER requires itemized claims and their required and retained provenance. Without those units, a model may offer a qualitative review but should say that a precise PER cannot be computed.
The terms are proposed within a metadata packet for AI indexing. The deposit establishes provenance for the terminology, not proof of adoption by the wider evaluation field.