South Korea counts its paper ballots with optical sorting machines. Ballots the machines cannot read are set aside and confirmed by hand, and every counting station produces a written record of both streams. Those records are the primary evidence in any serious discussion of how the machines behave — yet for years they circulated, if at all, as scattered scans and spreadsheet files that were hard to find and harder to check.
We have now assembled the counting records of four consecutive presidential elections — the 18th through the 21st, spanning 2012 to 2025 — into public archives that anyone can browse and download, free of charge. Each archive pairs the original records with the National Election Commission's official precinct-level results, so that every figure can be traced back to its source.
Ballot-sorter operation results for all 251 cities, counties, and districts, together with the official precinct-level results: 17,871 rows covering six candidates. The original counting documents are also public: 73,363 scanned pages from all 251 districts, republished with permission from Project Boo’s 2012 archive (projectboo.com) with attribution; files the source board never embedded are marked as partially missing rather than silently omitted.
https://storage.googleapis.com/k18-archive/index.htmlBallot-sorter operation records for 13,949 counting units, organized by province, together with the official precinct-level results: 22,214 rows covering thirteen candidates.
https://storage.googleapis.com/k19-archive/index.htmlScanned originals of the official counting reports (tally sheets): 17,377 pages from 224 districts, together with the official precinct-level results: 22,752 rows covering twelve candidates. Originals for 26 districts — including all of Busan, Jeju, and Sejong — were not obtainable and are absent; the archive states its coverage plainly rather than papering over the gaps.
https://storage.googleapis.com/k20-tally-archive/index.htmlScanned originals of the official counting reports: 18,949 pages covering all 252 districts, each district linked directly to its source documents, together with the official precinct-level results: 22,685 rows covering five candidates. Rows excluded from analysis — hand-counted reports and misfiled pages — are itemized, not silently dropped.
https://storage.googleapis.com/k21-tally-archive/index.htmlClaims about ballot-sorting machines — including the widely debated statistic K, the ratio of winner-to-runner-up votes among machine-rejected ballots relative to machine-accepted ballots — can only be tested against the counting records themselves. When the records are public, a claim becomes a calculation that anyone can repeat: the number of ballots each machine classified, the number it set aside, and how the hand-confirmed ballots split among candidates, precinct by precinct and district by district.
These archives underlie our own research on machine rejections across the four elections, including a transcription of the 2022 and 2025 counting reports validated by arithmetic constraints and blind re-reads. But they are not published for our work alone. Journalists, researchers, election officials, and citizens can use them to check any figure — ours included — against the originals.
All four archives are hosted on Google Cloud Storage and are readable without an account or fee. Source documents are the National Election Commission's published counting reports and result files; where a record could not be obtained, the archive says so. The archives are a project of the Electoral Insights Hub (OhMyCode LLC), a non-partisan, bilingual platform for Korean election data that favors transparent and reproducible analysis.
If you find an error, we would rather hear about it than have it stand: every page in these archives exists so that the numbers can be argued with — on evidence.
These archives are free to use. If they saved you time, you can help cover their hosting costs via PayPal.