A fetch-back into an empty directory found `vworld verify` exiting nonzero for anyone who fetched: the raidex axis was in the git repository and not in this dataset, and `corpus fetch` filtered the tree to eden_e* so uploading it alone did not reach a reader either. Both halves are fixed; a third clean fetch now gives 481 cells + raidex_pool.json, manifest MATCHES, verify exit 0.
The card no longer states the figure count (it said 17; verify reports 21) and explains that the axis is fetched beside the cells but is not part of the manifest digest. Cells and digest unchanged: 481, fb2234415b322451.
Provider is boundary zero; the card stops writing counts it cannot keep
[CORRECTION] 11's sweep. Four boundaries now, not three: provider joins generation, because the corpus holds Together, DeepInfra and OpenAI columns and a model's rate may not be compared across them.
Three counts here were stale against the data they describe: 357 vs 481 cells, VERIFIED 13 vs 69, and a manifest digest for 259. The cells and manifest were current the whole time; only this card lagged. It no longer states any count that can grow -- `vworld corpus status` and `vworld corpus identity` print them. Digest fb2234415b322451, 481 cells, unchanged by this update.
Retraction 12: relabel 161 cells to their served model (issue #113)
