How We Keep the Record
The E Don Kast methodology: every sheet fingerprinted at capture, most sheets read by machine with automatic checks, keyed blind by people where readings disagree or a sheet is unclear, and traceable from every published number back to its sheet.
Updated 8 October 2026
Every figure E Don Kast publishes traces back to a polling unit result sheet, or to a recorded cancellation. A result sheet is the handwritten form completed at the polling unit and published by INEC to its IReV portal, or photographed and submitted by an observer on the ground. This page explains what happens between that sheet and a number on our board.
The sheet is fingerprinted the moment we capture it
When a sheet enters our archive, we compute a cryptographic fingerprint of the exact image and keep the image under it. If anyone, including us, altered the image afterward, the fingerprint would no longer match. Nothing in the archive is overwritten or deleted: when IReV replaces sheets on election night, we keep all versions, each under its own fingerprint. The archive holds what was published, permanently.
Machine reading, with automatic checks
Each machine reading extracts every party's figure, the registered and accredited voter counts, and the valid vote total.
Most sheets are read by machine and pass automatic checks before anything appears. A clean reading of INEC's own sheet, as published to IReV, can stand on that single reading once it passes those checks. Where readings disagree, where a sheet is unclear, or where the sheet did not come from INEC's own upload, it goes to people.
We also do the math on every sheet ourselves, checking whether the party figures sum to the stated valid votes, rather than trusting any reader's arithmetic. A sheet that does not reconcile is flagged whether or not the readers agree.
Where machines disagree, humans key blind
Where readings disagree or a sheet is unclear, the sheet goes to people, who key every figure blind: no keyer sees the readings or another keyer's entry. A single blind transcription that corroborates the automated readings settles it. Where it cannot, two independent transcriptions must agree exactly, figure for figure. Anything still unresolved goes to a human adjudicator, who rules with the sheet in front of them. No published figure is final: when a later reading or check disagrees with it, it is reviewed and can be replaced.
We measure our keying accuracy the same way we measure everything else, including with sealed test batches whose answer key is opened only after keying is complete.
What we publish, and what we do not
We publish what the sheets say, with each line traceable to its sheet. We publish coverage honestly: how many polling units our record holds and, where INEC never published a sheet at all, the missing polling units by name, because a gap should be a specific, answerable question. We flag anomalies for scrutiny with the checks they fail stated, and the benign explanations acknowledged. We do not declare winners, adjudicate disputes, or present our aggregation as an official result. Only INEC declares official results.
Why this design
Any single reader, human or machine, makes mistakes. Machine readings are checked automatically, sheets are keyed blind where readings disagree or a sheet is unclear, and a figure is reviewed whenever a later check disagrees. That is the whole method: machine readings checked automatically, escalation to people, review when a later check disagrees, and a fingerprinted trail from every published number back to the sheet it came from.