How prediction markets become answers.

Prediction markets are the most honest signal in the modern world. They have to be: every contract is backed by real capital, every opinion carries a holding cost, and no false narrative survives a settlement date.

Between Kalshi and Polymarket, billions of dollars flow across tens of thousands of active contracts — colliding with an unstructured firehose of global news, algorithmic social noise, and fragmented analysis. The signal is real; the reading of it is the hard part.

Rankd reads two venues: Kalshi (a CFTC-regulated exchange) and Polymarket (an on-chain order book). Every scan filters both universes down to the top 300 markets by 24-hour volume — liquid, executable, and worth analyzing. Rankd holds no funds, executes no trades, and connects to no venue account; acting is always yours, on the venue.

Filter — the math. Deterministic screens run first: volume, spread, depth, and time-to-resolution. Markets that fail the executability gate never reach an AI. Math decides what is worth analyzing; models never decide what is true.

Analyze — the swarm. For markets that pass, live research gathers evidence — sourced, timestamped, and cited. A synthesis model reads the evidence bundle and only the evidence bundle: it never sees the market price, so the analysis cannot anchor to the crowd it is meant to check.

Score — the output. The 0–100 Rankd Score is pure code: a geometric composition of Edge, Confidence, Executability, and Timing. Every number shown traces to an evidence item or a component computation — an explainability chain, not a vibe.

Every scoring run is written to an append-only prediction log — never updated, never deleted. When markets resolve, outcomes are recorded against the original predictions. Brier scores and calibration plots are computed from day one, and calibration is published: the accuracy claim is measured, or it is not made.

Thousands of events every day, synthesized into the top 20 — with the reasoning attached and the score decomposed. Rankd is decision support: it never tells you what to do, and it shows its work so you can disagree with it.