Eleksyon 2025: Tech, Misinformation, and Fact-Checking

Archive note: this is a backdated post, written years later while rebuilding this site. It’s dated to the moment it covers, but the hindsight is real.

The May 12 midterms are behind us — twelve Senate seats, the whole House, every local post — and as usual I have nothing to say here about candidates and everything to say about the machinery. Philippine elections are, among other things, one of the world’s largest deployed information systems, and this cycle ran on more new technology than any since automation began in 2010.

The counting layer: new vendor, same questions

This was the first national run for the new automated counting machines after COMELEC switched vendors — ending the long Smartmatic era. The engineering observations transfer from every cycle: transparency of the source review, the random manual audit as the actual trust anchor, and transmission integrity mattering more than any single machine. Election-day reports of misreads and glitches got loud play; proportionality is the discipline — at this scale, a zero-incident deployment would itself be suspicious. The system’s real test is whether failures were detectable and correctable, and the audit layer exists for exactly that. My standing position since I started watching these: paper ballots plus machine counting plus mandatory audits is a genuinely defensible architecture — if the audit is honored and the chain of custody is boring. Boring is the goal. We mostly got boring.

The information layer: this is where it’s bad

The actual battlefield was upstream of any machine, and this blog has been assembling the pieces for years: a country that lives on one platform’s feed, the engagement machinery of short-form video, and now generative tools that make fabricated audio and video cheap. The combination performed as predicted:

The deepfake era arrived on schedule. Fabricated audio clips of public figures, AI-generated endorsement content, resurrected-and-recut old footage — the provenance problem flagged after Sora is no longer a prediction; it’s a campaign tactic. Detection lost the arms race before it started; provenance (proving the real, not catching the fake) remains the unbuilt infrastructure.
The fact-checkers’ math is impossible. Coalitions like the academe-media tie-ups did heroic work, and the asymmetry held anyway: fabrication costs nothing and travels on outrage; correction costs research and travels on duty. A lie in a group chat outruns its rebuttal by days — if the rebuttal penetrates the chat at all, which family-thread dynamics suggest it doesn’t.
The vocabulary moved down-ballot. Disinformation playbooks once reserved for national races are now local-election tooling — cheaper tools, smaller markets, zero coverage. The provincial information environment is where the next decade of this fight actually happens, and nobody funds it.

What a developer does with this

The same answer as every infrastructure post here, scaled to citizenship: the fix is unglamorous and distributed. Be the person in your family chat who checks before forwarding — not the debater, just the friction. Support the boring institutions (audits, fact-check coalitions, local journalism) the way we fund maintainers: because load-bearing and underfunded is a dangerous combination. And build, where we can, for provenance — signed media, verifiable sources, systems that make truth cheaper instead of louder.

Democracy, it turns out, is also a system that fails at its trust boundaries. The ballots got counted; the feeds remain contested. See you in 2028 — the machines will be newer, and the fakes will be better. The audit, as always, is us.