Datasets

The deposit holds the corpus, the pipeline that produces it, and the 36 hand-annotated pages the parser is scored against. Of the 233,408 passages, 151,761 are speech and 31,955 are typed stenographer's notes, and the rest are section headings, page furniture, and stray italics, kept so that any passage can be traced back to the page it came from. Each passage carries the speaker label the page printed, the senator that label resolves to, and the caucus that senator sat with on the day of the sitting, which is not the list they were elected on and differs from it for nearly a fifth of floor speech. The source PDFs are not redistributed. A manifest identifies each one by checksum and URL, and the shipped downloader fetches them again.

Coverage. 559 sittings, February 2000 to December 2024. Complete from 2004 onward and thinner before that, because the chamber's portal lists the earlier sittings but no longer serves the files.

Caveats. Two sittings of November 2001 are character recognition of scans and should be excluded from text analysis, and one sitting does not parse at all. 1,109 passages carry no speaker. The caucus is observed on 336 days across 25 years rather than continuously, and every row says which day it was read from.

Citation. Cite the concept DOI for the corpus, or the version DOI when the exact files matter. This is version 0.4.37, whose own DOI is 10.5281/zenodo.22661020.

Covers the national elections of 2011 through 2025 and the city elections of 2013 through 2025, for every office on those ballots, from president and national deputies to Head of Government, the City Legislature, and the Communal Boards. The results were curated from BA Data/GCBA, the historical references of the Tribunal Superior de Justicia, DINE, and Datos Argentina, and are shipped with a manifest and coverage reports that say which source each row came from.

Coverage. National elections 2011 through 2025 and city elections 2013 through 2025, at polling-table level.

The table ranges line up with the polling tables in the results dataset above, so the two join. The public files keep the original establishment, address, and locality fields as they were published, and add curated canonical establishment names, canonical addresses, opaque entity identifiers, and WGS84 coordinates.

Coverage. Voting-location assignments by year, voting location, and table range, for elections from 2011 through 2025.