Capture Registry › capture leesharks-datasets-aio-20260922
One record of the canonical Capture Registry (EA-WG-CAPTURES-01), cited at https://www.alexanarch.org/captures/leesharks-datasets-aio-20260922/. the canonical Capture Registry (version 12.38) · the address page · this card in the gallery · this record as data · table of contents.
The rerun prescribed on 2026-09-21. That record, 'tiger-leap leesharks datasets', resolved the datasets correctly but pulled in three unrelated 'tiger leap' referents, and its rerun line asked for this string to separate the homonym load from the author scope. With the token removed, the homonyms are gone: six cards, all archive-controlled, none homonymous, and the datasets named as the archive's own (negative-of-the-negative, provenance-erasure, the capture registries).
Across the pair, the only change in the issued string is the removal of 'tiger-leap', and the three homonym cards leave with it (3 of 12 on 2026-09-21; 0 of 6 today). That is consistent with the hypothesis the earlier record stated: the homonym load belonged to the token, not to the author scope. It is two observations on two different strings, one day apart, and it establishes that the outcome is reachable, not a rate. The composition itself is thin but correct, and it carries an identifier in the body, which is rare for this surface.
The second observation at this address holds the first one's result: every card archive-controlled, no homonyms, author and archive named. Two things moved. The DOI present in the 2026-09-22 body is absent, and the registry is described through its June Zenodo versions ('v7.0', 87 captures) although it stands at v11.7 with 469 addresses. The heteronyms dataset is described as mapping 'writer positions (such as Antioch, Damascus Dancings, and Jack Feist)': Antioch is a volume of poems (#331), not a position. Two of the Hugging Face cards are commit pages from the rebuild pushes, not dataset cards.
Two observations one day apart, both clean on sources. The layer's view of the registry is frozen at the last Zenodo versions, the ones the severance left standing, while the live projection it also cites (leesharks.com captures) is current. The work-for-author slip is small and specific: a title in the heteronym list.
The rerun prescribed on 2026-09-21. That record, 'tiger-leap leesharks datasets', resolved the datasets correctly but pulled in three unrelated 'tiger leap' referents, and its rerun line asked for this string to separate the homonym load from the author scope. With the token removed, the homonyms are gone: six cards, all archive-controlled, none homonymous, and the datasets named as the archive's own (negative-of-the-negative, provenance-erasure, the capture registries).
Across the pair, the only change in the issued string is the removal of 'tiger-leap', and the three homonym cards leave with it (3 of 12 on 2026-09-21; 0 of 6 today). That is consistent with the hypothesis the earlier record stated: the homonym load belonged to the token, not to the author scope. It is two observations on two different strings, one day apart, and it establishes that the outcome is reachable, not a rate. The composition itself is thin but correct, and it carries an identifier in the body, which is rare for this surface.