The index holds three states for the question "did it come back", and almost every citation collapses them into two. That collapse is the single most common misreading of this dataset, and it runs in the direction that flatters enforcement.
| State | Incidents |
|---|---|
| A successor is documented | 50 |
| Searched, no return established | 22 |
| No return finding either way | 33 |
Only the first row is evidence of anything happening. The second is evidence of work: somebody put the question, opened sources, and recorded what they did not find. The third is evidence of nothing at all, and it is the row people quietly convert into a success.
Why the middle row exists
Return evidence does not live in the document that establishes the takedown. A prosecutor's release describes what was seized on the day it was filed. It cannot describe what happened eight months later, because eight months later had not happened yet. Everything this index knows about returns came from somewhere other than the release that documents the action: vendor telemetry, specialist reporting, a later indictment that mentions the earlier service by name.
So a blank is not a finding. It usually means the follow-up research has not been done for that record, and the index now says so explicitly rather than leaving you to infer it. Where the search has been done, the record carries the evidence of the search itself.
The rate you can quote, and the one you cannot
50 of 105 counted incidents carry a documented successor. That is 48 percent, and it is a floor rather than a rate, because it divides by every incident including the 33 nobody has looked at.
Among the 72 incidents where the question was actually put, 50 have a documented successor. That is 69 percent. It answers a different question, with a denominator you can defend, and it is the figure to use if you are going to use one.
Neither number supports a claim about permanence. A takedown with no documented successor has not been established as permanent. It has been established as unresearched, or as searched without a successor being found, and those two are not the same claim as each other either.
Language that survives review
Usable: no successor recorded, the index documents no successor, searched without a successor being established, not researched for return.
Not usable: stayed down, no return, still offline, permanent, successful, and any percentage of takedowns described as having worked. The index cannot see whether a service is running. It can only see what published sources establish, which is a much narrower thing.
The coverage gaps table records this limitation in the dataset's own words, and do takedowns hold? works through the whole cohort with the same three states in view.