Biodiversity inventories in Brazil: 2,850 species occurrences from 28 publications

2,850 species occurrences from 322 sampling events, transcribed in Darwin Core terms from the species tables of 28 open-access Brazilian publications (1982-2020): 23 from Embrapa's open repositories and 5 master's dissertations (UFU, UFV, UFMG). Fish, birds, mammals, bats, frogs and reptiles, dung beetles, butterflies, moths, bees, trees and mangrove plants, from the Amazon, Cerrado, Caatinga, Pantanal, Atlantic Forest, Pampa and the coast, in 13 states. Two files joined on eventID: events (study, site, campaign, point, plot, habitat unit or stratum, with locality, printed and ISO date, printed coordinates, habitat, protocol, effort and fate of the material) and occurrences (scientific name as printed, qualifier, family, order, presence or absence, count or quantity, basis of record, page and table of origin). Names and coordinates are kept exactly as printed; an ISO eventDate is derived from the printed date where it is unambiguous (1,087 occurrences); there are no decimal coordinates yet. Each document was checked against GBIF before transcription and none had its records there. Each document was transcribed twice, independently, by AI agents, and every disagreement (3,199) was settled against the PDF; the distinct names were matched against the GBIF backbone as a check. A human reviewed the result before publication. Version 2.0.0 adds a round of 20 documents to the 8-document pilot (v1.0.0). Part of the Paper Rerun project (Open Tables).