Faro researches trade signals in prediction markets like Polymarket and Kalshi. Two kinds: arbitrage, where related prices don’t add up, and forecast mismatches, where a market disagrees with independent forecasters. Every signal is verified, backtested and paper-traded, and shown with its edge, its cost and its uncertainty.
Demo screens run on sample data. No account, no orders — Faro is a research tool.
Faro supports a decision, not a trade: its output is a decision with its evidence — what the opportunity is, how sure we are, what it costs and what could go wrong. Each question gets a gate with thresholds fixed in advance.
Can we buy every outcome of a Polymarket event for less than it is guaranteed to pay, after fees, at real depth?
Arbitrage · deterministicDo Metaculus, Manifold and ForecastBench disagree with market prices often and strongly enough to pay after costs?
Forecast mismatch · statisticalDo Kalshi and Polymarket price the same event differently by more than it costs to trade both sides?
Cross-venue · rules can differThe first question again, on Kalshi's own multi-outcome events.
Arbitrage · deterministicA signal is only emitted when a pre-registered condition is met — minimum edge after fees and uncertainty, a verified payoff, enough depth, fresh data. Fewer, better signals, each one you can trace and trust.
Third-party data is cross-checked against an independent view before we rely on it, and when two views disagree the conservative one wins. An arbitrage is only called one after every settlement state is enumerated; a forecast that disagrees with the market is a hypothesis, labelled as such.
Nothing here promises a return, and nothing is optimised toward one. Theoretical spread, executable profit after depth and latency, realised P&L, fees and idle time are always separate numbers, each traceable to a data snapshot, a config hash and a commit.
Polymarket and Kalshi markets, order books and resolutions, plus independent forecasts from ForecastBench, Manifold and Metaculus — all stored as they were known at the time, never as they turned out.
Multi-outcome events whose prices do not add up are found by a scan over quotes, then checked against real depth, the venue's own fee schedule and latency. A spread that cannot be executed is not profit.
Every run records its data snapshot, config and commit. Three fill scenarios are always reported. A hypothesis keeps an untouched out-of-sample window and has to pass a gate before anything moves forward.
Signals are paper-traded on live public data with simulated fills, limits and a ledger, so their value is measured before anyone trusts them. What survives can be exported — as a probability, an edge or an alert — to wherever it is useful, including markets Faro itself never trades.
We matched outside forecasts to the Polymarket and Kalshi markets that ask the same question, read every match by hand, and scored them against the market's own price. Manifold's crowd was worse than the market, clearly so on the questions it wrote for itself.
1 November 2026What Faro is, how it runs on one modest server, and what happened when we replayed every order-book change on Polymarket for a full week: arbitrage that is real and persistent, but worth about $20 a day.
31 October 2026A plain-language tour of prediction markets — how a 0-to-$1 contract turns into a probability, how Kalshi and Polymarket grew from curiosities into $50-billion-a-month venues, and why regulators from Rome to Ohio are pushing back.
Faro is a self-hosted research platform built by Klaratech, for its own decisions first. Prediction markets are restricted in much of the world, including where Faro is built, so it treats them as data. Ingestion, arbitrage verification, the simulator, the ledger and this dashboard exist. The first question is answered: a full week of Polymarket order books showed mispricings that are real and persistent, but thin. Forecast mismatches are next. Live execution, if it ever comes, would only carry out a decision Faro already supports.