Voice saw this coming. Our numbers were more accurate than prediction markets in every major New York City congressional race. But our numbers don’t just predict outcomes, they tell us why those outcomes are happening. They listen to voters. How Voice called the NYC primaries. And what prediction markets missed. Last night, NYC voters swept Mamdani's endorsed slate into Congress. Voice had the insurgents winning. Here's what the sentiment data showed all along: NY-13: The ICE donations Darializa Avila Chevalier vs. "King of Dominican politics" Rep. Adriano Espaillat. When Mamdani endorsed Chevalier on May 29, her Voice rating jumped to 70%. Hochul's endorsement of Espaillat the next day flipped his rating to 55%, the sharpest single-day reversal we saw across any race all cycle. Then came the ICE contractor donations story on June 9. Espaillat's share collapsed to 28%. Chevalier hit 72%. His numbers never fully recovered. Espaillat famously leaned on his Dominican ethnicity in this race. But, as this drop off illustrates, his immigration record undermined his appeal to immigrant communities. Polymarket had Espaillat at 66% to win heading into election day. Voice had Chevalier taking 49% of the vote. Result: Chevalier won with 49.4% of the vote. NY-12: A tight one Polymarket had Lasher at 73% chance to win. Voice had Lasher at 40%, Bores at 38%—essentially a coin toss. Result: Lasher 39%, Bores 35%. NY-07: The funding integrity race Rep. Velázquez retired and endorsed Brooklyn Borough President Reynoso. Mamdani backed Assemblymember Claire Valdez. The PAC-money story (Jun 5) and AI-donor story (Jun 8) put consistent pressure on both Valdez on donor and message consistency. Valdez's two-way share actually ticked up from 55.4% to 59.4% across those moments. Polymarket had Valdez at 81% chance to win. Voice had her at 54% percent of the vote. Result: Valdez won with 58.1%. The pattern: Prediction markets weight incumbency, fundraising, and establishment endorsements. They weigh the old metrics. Voice measures how constituent sentiment reacts to narrative. We tell—and track—election stories.