Laboratory Open-source the methods
Eighteen thousand walls.
By Violeta Ayala — Tecnologa Creativa
The wall before the server
How to replicate Taiwan's public count, and where Bolivia, Colombia and Peru fail.
Start with what is not the problem. Bolivia, Colombia and Peru already count votes in public, by hand, at the table, with the parties watching. They do that well. The count is not the flaw. The flaw sits in what gets inserted between that public count and the official number: a layer of transmission and digitization that becomes, in fact or in perception, the source of truth.
Taiwan closes that layer with a single move. Each table's result is called out loud, posted on a board on the wall of the polling place, and only then travels to a central system that merely adds it up. The number that circulates on election night is born on almost 18,000 walls, not on a server. That makes it one of the fastest counts in the world, and the reason is not technology. It is institutions.
What follows breaks that model into six replicable guarantees. None requires machines. None requires Taiwan's context. They require moving the authority over the result to the place where there are witnesses, and publishing it right there.
AThe protocol
Manual, public count at the table
Paper ballots, counted by hand, in the same place where people voted, the moment it closes, with the door open to anyone.
The count happens where there are witnesses, not in a distant centerReading aloud with redundant verification
Each ballot is pulled out, held up for the room to see, read aloud, and two independent people repeat it and mark it. Anyone present can call out the error on the spot.
No vote is altered in silenceThe tally sheet is fixed and published at the polling place, instantly
When the count ends, the table's result is posted on the wall of the venue and the photo of the original tally sheet is published openly, before it enters any central system. This is the piece that changes everything.
A public, fixed record exists that does not depend on the centerThe public number does not depend on the software
The central system only adds up results that are already public. The first number, the loudest, the one that circulates that night, is born on thousands of walls.
The software reports and never decidesPer-table data, open and tied to the original
Each tally sheet is published in open format, with an integrity seal (hash and timestamp) that binds it to its original and makes it impossible to change without a trace.
Anyone adds up the whole country and reconciles the official total against the tally sheets, table by tableCustody of the physical record and verifiable recount
Ballots and original tally sheets are kept under documented chain of custody until proclamation, and the physical recount of disputed tables is done in public.
Any challenged table can be recounted against the paperBThe matrix
The same six guarantees, scored across the four systems. Green meets it, amber is partial, red fails. Taiwan's column is full. Latin America's failures cluster in rows 3, 4 and 5: the publication and authority layer. They count and store well. They publish and rule badly.
CWhere each one fails
Bolivia
Counts well and in the open. Fails at publication. The tally sheet is not posted for everyone at the venue, it only goes to the delegates, so the country's only fast number is the transmission system's. In 2019 that layer, not the count, decided the crisis: the TREP was suspended, resumed a day later with the trend flipped, and that was enough to set off everything.
Fix: post the tally sheet on the wall of the venue, publish the photo of every original tally sheet openly and instantly, and make every step of the Sirepre auditable in real time. Let the fast number be born at the table, not at the center.Colombia
Counts in public and even posts the E-14 at the polling place, more than Bolivia does. Fails at the binding. The digital image it publishes is not verifiably tied to the original, traceability degraded (corporation as "XXX", table as "000", metadata as NULL), and the preliminary count is the night's number. Verifying the published against the original gets hard, and the parallel count becomes a fight.
Fix: an integrity seal (hash and timestamp) on every published tally sheet, restore full traceability, and an open per-table dataset. Let anyone reconcile published against original in a minute.Peru
Counts in public and has the best piece of the four. The physical recount of the observed tally sheets is done by the Special Electoral Juries live on YouTube, and since 2026 it keeps the ballots until proclamation. Fails at uneven custody and slowness. The overseas vote was not transmitted digitally, it traveled physically, arrived late and, per the complaint, without proper custody. And the slow count opens the vacuum that rumor fills.
Fix: uniform, documented chain of custody for every vote, including the overseas vote; open publication of the per-table tally sheet instantly; and keep and expand the live recount, which the rest should copy.DWhat not to copy
Taiwan has features that should not be replicated. There is no early or overseas voting. Everyone returns to their registered polling place, which pushes hundreds of thousands to travel on election day. That prioritizes authenticity over access, and for large, dispersed countries it is a cost there is no need to pay. What is replicable is not restricting access. It is the public count and the immediate publication at the table, which works with or without an overseas vote. In fact, Peru's weak point was precisely the overseas vote, so the lesson is the opposite: give the vote from abroad the same custody and the same publication as the vote at home.
EThe limit
A protocol closes the evidence gap, not the trust gap. If an actor decides that losing is theft, no wall convinces them. We already saw Keiko hold the fraud claim for years, after every verdict. But the wall does something decisive: it makes the accusation cheap to disprove. When each table's result is public and photographed before the server, whoever cries fraud has to show which table and which number, against an original anyone can see. That forces no one to accept. It isolates bad faith and puts the burden back on the accuser.
The rule fits in one line. Moving the authority over the result from the server to the table, and publishing it right there, is the only thing that makes fraud have to be proven instead of just shouted. Everything else asks people to trust a box they cannot see.
