ACI-017: A Measurement Is a Claim About a Population

Thesis

Whatever an agent's search, count or measurement returns is a claim about a population, a definition of what matches, and a unit. Every filter, glob or truncation changes the population while the answer still looks complete, so a result is only as good as your knowledge of what it looked at, and a verification is only as good as the property it measured.

Story

On 17 August 2026 an agent working on Intent reported 42 call sites of a function, 6 of them reaching a human. There were 34, and 4. grep -c counts lines that contain a string, so it had counted an import and an explanatory comment as calls (Intent@d8c96828a, 2026-08-17). The line that inflated the count turned out to be the annotation left on the previous instance of the same defect: the better the annotation, the more reliably it is miscounted, "because a good one quotes the defective call" (Intent@0d65c8bf0, 2026-08-17). That bias was measured once, on one symbol, but its direction is worth knowing. It makes a problem look larger exactly where it has already been fixed.