Four months after a July 9 measurement, 270 of the 349 servers that answered then answered again on November 15. Of the rest, 67 were still listed but did not answer our probe, and 12 were no longer in the registry1.
The sample is concentrated. One publisher runs 102 of the 349 servers, built from shared templates. Counted by publisher, 153 of 208 had at least one server answer again.
Most of the tool definitions that changed belong to that publisher. Across all 270 servers, 55.5% of the July tools had a different fingerprint2 in November. Leaving that publisher out, 60.5% of the July tools were unchanged, 29.7% had changed and 9.8% were gone. Counted by publisher, 81 of 153 had at least one July tool changed or removed.
Two changes are reasons to look again at an earlier approval. One tool that returns private session state changed its documented authentication requirements. One paid tool changed the currency and payment rail its definition asks clients to use. Both were observed before and after. The evidence is the definitions themselves3.
How the sample was taken: one in twelve names from the official registry. From August 7 an attempt cap skipped some names late in the registry walk, so figures between August 7 and November 14 cover only part of the sample. The November 15 run revisited a fixed panel of 897 earlier names with none skipped. The figures above use only that run and the July 9 one.