What this is / is not. This piece discusses general patterns in election integrity infrastructure. It does not endorse or critique any specific candidate, party, or past electoral outcome.
The paper that holds the result together
Form 34A is the polling station results declaration used in Kenyan presidential elections. One sheet per station, completed on election night, signed by presiding officers and agents present at the count, and posted publicly outside the station before the original is transmitted upward. The form is deliberately simple: the count is performed manually, the figures are written by hand, and copies stay with people who were physically in the room.
That simplicity is the source of its credibility. The whole document chain - station-level form, copies retained by agents, scanned image uploaded for public verification - exists so that a result is reconstructable from artifacts produced under direct observation. Digitizing that chain is therefore not a question of replacing paper. It is a question of preserving the properties of paper while adding the speed and consistency of software.
What digitization actually changes
Three things shift when station-level forms move through a structured digital workflow.
Capture latency drops to minutes. A field agent photographs the posted copy of the form outside the station the moment it is displayed. OCR extracts candidate vote counts, rejected ballots, and station identifiers. A reviewer confirms the figures against the photograph. The result is a verified station total available to a campaign or observer organization before the original reaches a constituency tallying center. In the legacy workflow the same data point would arrive through a chain of phone calls, often hours later, often with translation errors.
Source documents stay attached to numbers. Every figure in an aggregated total can be traced back to the photograph it was extracted from, the agent who uploaded it, the reviewer who confirmed it, and the timestamp at each step. A spreadsheet total that disagrees with the source is no longer a guess - the disagreement is visible at the row level.
OCR speeds up capture but does not become authoritative. The pattern that matters here is a soft one: extracted figures are a draft. Confirmed figures, verified by a human reviewer against the photograph, are the source of truth for any downstream aggregation. This is deliberate. OCR misreads smudged ink, faded carbon copies, and handwriting from agents who have been awake for thirty hours. A model that trusted OCR output would erode the manual verification that gave the paper chain its credibility in the first place.
What stays the same
Digitization does not move the trust boundary. The credibility of a Form 34A still rests on the same things it always has: agents from contesting parties were present at the count, the figures match what was announced in the room, and the document was posted publicly before transmission. A photograph of a forged form does not become a real form by being uploaded to a platform. A chain-of-custody record that begins with a verified agent photographing a publicly posted form on a known station, on the night of the election, is materially different from a spreadsheet number that arrived by SMS through three intermediaries.
The integrity story is not "software replaces paper". It is "software makes the paper trail legible, fast, and reviewable at a scale where humans alone cannot keep up". That distinction is what separates a workflow that survives a courtroom challenge from one that adds a layer of process without adding a layer of evidence.
What this means for campaigns and observers
For a campaign chief technologist or an observer organization planning for the 2027 cycle, the question is no longer whether to digitize. Phones are already at every station. The question is whether the digital record matches the integrity properties of the paper record it is meant to support. Three concrete tests are worth running against any platform under consideration:
- Can every aggregated total be traced back to a specific photograph, agent, reviewer, and timestamp?
- Is OCR output marked as a draft until a human reviewer confirms it against the source image?
- When the platform exports a result for a campaign, journalist, or court, does the export include the source documents and the chain-of-custody log, not just the aggregated numbers?
Any platform that cannot answer yes to all three is offering speed without offering integrity. The two are not the same.