AFOS Analytics - Open-source civic intelligence for the 2026 electoral cycle

Hi PH community, I built AFOS Analytics because I wanted to read political news without it getting averaged through a black box. Polls capture what voters say they'll do. Prediction markets capture what informed bettors are willing to back with real money. Each one is imperfect on its own — together they give a sharper read on where political reality actually sits. Coverage: Brazil 2026 plus 14 other monitored elections. Three signals running side by side — Polymarket prices, the 17+ polling institutes registered with Brazil's electoral court (TSE), and 400+ press sources auto-categorized. Here's the built-in guardrail: when markets and polls converge within ~3pp, the signal is robust. When they diverge by 5pp or more, that's where the alpha tends to sit — one source has caught something the other hasn't priced in yet. The platform flags those divergences explicitly instead of smoothing them into an average. Two concrete cases from this past week. The presidential gap on Polymarket collapsed from +6.15pp to +1.10pp in 24 hours (May 10), then snapped back to +4.85pp two days later (May 12). Both moves preceded poll publications by 24–72 hours. STF impeachment probability fell from 15.5% to 10.5% on May 12, the same day the opposition filed a new impeach request — the market priced three alternative institutional pathways (Senate TSE swearing-in, PEC Anistia, judicial coordination on Lei da Dosimetria) as making impeachment a secondary route. Why open-source: this kind of civic infrastructure shouldn't be a black box. Apache 2.0, fork-friendly, every claim links back to its primary source. Daily editorial syntheses (AFOS Daily) publish in three languages at /daily, with auditable links per claim. Stack: Next.js 14, Prisma, Neon Postgres, Upstash Redis, Vercel. Live: https://afos-analytics.com Repo: https://github.com/AFOS-Analytic... Happy to answer questions about the cross-reference methodology, the daily editorial guardrails, or the open-source approach. Honest critique welcome.