Saturated Metric

A measurement that was already at its maximum before the change being argued for.

Why it matters in a memory layer

It is worse than a missing metric. A missing one is an admitted gap; a saturated one looks like evidence, reports the same value for both systems, and sits in the table next to an informative metric with no visual difference.

Three of this course's four component metrics report 1.000 for every profile from I4 onwards. They are correct and they are valuable as regression guards — and citing them as evidence for anything built afterwards would be false while every individual number remained true.

So the distinction has to be computed and printed rather than remembered: a metric counts as informative only if two of the compared systems disagree on it. That is the line between a result and a number, and the incentive runs the wrong way — dropping the saturated metrics makes a claim sound weaker and say more.

Connections

Taught in: Reading Benchmark Claims Critically

Do not confuse with: Benchmark Claim