A simulation of the model behind the change in qrp-mcp. Every report draws one spelling of one operation. The orange line fixes only the spelling it was shown. The teal line closes the whole operation the first time any of its spellings appears. The shaded band is the gaps the second method never has to meet.
Closing an operation turns each family into at most one surprise, its first appearance, plus the spellings its documentation misses. The chance of meeting a family is at most the sum of the chances of meeting each of its spellings (the union bound). So the teal line never rises above the orange one, for any number of reports.
If spellings are long-tailed, as words in a language are, the number of distinct ones seen in n reports grows like a power of n (Karlin, 1967). It slows down but never levels off. Closing operations replaces that endless tail with a finite set of families, which runs out like a coupon collection.
This is the model, not our measurement. The families, frequencies and documentation share are set by the sliders, and each curve averages 30 seeded runs. On our own fix list (438 items, 1 Oct – 17 Nov 2026), relabelled by operation, about a third of detection gaps were later members of a family, the most this method could have saved there. We are measuring the real effect for two weeks after the change.