An indicator is its own control group

The smell test's first sweep returned 41,350 findings — 5.4% of every observation in the graph, which is a broken instrument rather than a result. 23,172 of them were negative percentages, and nearly every one was correct: the Percent unit covers bounded proportions and signed growth rates, balances and changes, with nothing in the unit string to separate them.

What separates them is frequency. A rule broken by most of an indicator's observations is its definition; a rule broken by three country-years out of three thousand is an error. So checks are suppressed per-indicator when they fire often enough to be structural, and the outlier check compares each value against its own indicator's 99th percentile rather than against what a unit is supposed to mean.

Suppressions are published, not dropped. The list of indicators that publish signed values under a Percent unit is a finding about the graph's vocabulary, and a report that hid it would look cleaner than the data is.