Case study
Citizen verification at scale — Watch the Votes pilot cohort
A 400-citizen pilot cohort photographed EC8A result sheets during a bye-election. The parallel count converged with the official result in six of seven wards; the seventh triggered an investigation.
- Observers enrolled
- 400
- Wards covered
- 7
- Consensus convergence
- 6 of 7
- Average upload latency
- 43 sec
- Discrepancy flagged
- 1 ward
Challenge
Citizen observation of election results in Nigeria has historically been limited to spot-checks and hearsay. Photographing an EC8A sheet at a polling unit is useful. Photographing seven at the same polling unit, extracting the tallies, reconciling against official results, and doing it for every polling unit in a state — that requires infrastructure nobody had built.
Approach
We recruited a pilot cohort of 400 observers across seven wards. Each received the Watch the Votes PWA on their own device, the glossary, and a 30-minute onboarding. No other training. The app was deliberately designed to work for first-time users with no election-observation background.
Deployment
The vote took place on a Saturday. The app accepted uploads from 9am to 6pm. OpenAI Vision extracted the tallies; each citizen confirmed; median consensus across multiple uploads per polling unit produced the parallel count. The INEC comparison was run at T+12 hours.
Outcome
The parallel count converged with the INEC-published result in six of seven wards to within 0.8% of the reported total. The seventh ward showed a 6.4% discrepancy that was referred to the relevant state commission and is the subject of an ongoing review. No observer was compensated for individual uploads; the system was designed to reward trust-tier progression instead.
"For the first time I could hold the result sheet from my ward in one hand and the national tally in the other. They matched. That is the whole point."