Validating the method on a known result, the 2024 US election (and where the market was wrong)
Body: AFOS runs live on Brazil's 2026 election. To check whether a "markets × polls × press" method actually works,
you need an election with a known result, so we ran our cross-referencing engine on the 2024 US presidential
election (Harris × Trump) and added it to our AFOS Analytics collection on Harvard Dataverse.
The two big Polymarket markets disagreed on the eve:
• Winner market (electoral college, ~US$ 3.7B, the largest election market in history): Trump ~56% → right.
• Popular-vote market: Harris ~74% → wrong (Trump won the popular vote too, 49.8% × 48.3%).
Inside: national poll toplines (FiveThirtyEight, CC BY 4.0, via Wayback), daily win-probability for both markets,
the market×poll divergence series, a Wayback-archived press timeline, two odds charts, a DATA_DICTIONARY and
CITATION. Harvard DOI: 10.7910/DVN/3DJCW5.
The divergence is the signal, and we keep the market's miss, not just its hits.