IReV is the receipt, not the proof: Yiaga's AI auditor targets the validation stages the public never sees in Osun

On Saturday, when Osun State voters go to the polls, Yiaga Africa will run an AI-assisted audit alongside its human observers. The AI is not watching the vote. It is watching what happens to the vote after the polling unit closes.

Nigeria's election results leave the polling unit in plain sight and then pass through a series of validation stages that voters, observers, and reporters cannot see. What eventually appears on the Independent National Electoral Commission's Result Viewing Portal, IReV, is the public receipt, not the proof that the number is right. That gap is what Yiaga's Election Result Analysis Dashboard, ERAD 2.0, is built to measure.

ERAD 2.0 is one half of a two-layer system. The other half is PRVT, or Process and Results Verification for Transparency, Yiaga's statistical observation method in which trained observers at a random sample of polling units record votes cast and compare them later to the official results. The PRVT sample is sized so that if the official count diverges from the polling-unit count in a non-random pattern, the divergence is statistically detectable. ERAD 2.0 then takes the official results posted to IReV and the human-verified PRVT polling-unit records and runs them through an AI comparison layer to flag discrepancies a manual review would miss or arrive at too late.

Saturday's Osun governorship is the first major Nigerian off-cycle vote in which both tools run together at this scale and before any official declaration, according to Yiaga's disclosure in Osogbo on Friday. The group previously deployed ERAD to analyze electronic transmission of results during the 2022 Ekiti and Osun governorship elections, but the combination with PRVT and a Saturday-only vote sharpens what the deployment is meant to prove.

Yiaga's executive director, Samson Itodo, framed the deployment as an independent verification layer on top of INEC's IReV pipeline rather than a replacement for it. The argument he pushed in Osogbo is that the presence of a result on IReV says nothing about whether the validation stages between the polling unit and the portal accepted or altered it. Those stages, in his telling, are where the public loses visibility and where an auditor built to compare two independent records earns its keep.

ERAD 2.0 is not a competing tabulation system. INEC still runs the count, the collation, and the publication. Yiaga is running a parallel measurement whose value depends on the official pipeline staying intact and publicly accessible. If IReV were taken down or throttled, the audit loses its second input and the comparison collapses.

Wire copy on Saturday night will read "AI monitors Osun election." That understates the statistical-sample frame and overstates how much of the results chain the public can already inspect on its own. ERAD 2.0's verdict will land days after the vote, and it will answer the question IReV cannot answer on its own: did the number on the portal match the number at the polling unit?

The deployment and Itodo's reasoning are detailed in Premium Times Nigeria's reporting from Osogbo.