by Predictive Labs
Public launch · September 2026
One unified book across every prediction-market venue. Order books, fee schedules, resolution rules, and oracle sources from Polymarket, Kalshi, and the venues alongside them, all normalised. The same event priced differently on two books surfaces as a single ranked line, sized for capital efficiency, scored against your own calibration history.
illustrative · same event, two venues
"Will the Fed cut rates by July?"
| Venue | YES | Depth |
|---|---|---|
| Polymarket | 0.42 | $1.2M |
| Kalshi | 0.39 | $0.8M |
2026 prediction-market volume · year to date
Tracking toward Bernstein's $240B 2026 projection · CNBC, 14 April 2026.
2025 cleared $64 billion across prediction-market venues, with daily volume holding through geopolitical, monetary-policy, weather, and sports event-driven contracts. Resolution rules tested in real disputes, oracle disputes adjudicated on-chain, settlement arbitration that did not exist in 2020. The asset class is being priced like one.
Every venue ships its own thin dashboard, none reads the others. The diagnostic stack equities desks take for granted (Brier scoring, Quarter-Kelly sizing, calibration drift by confidence bucket, Amihud-rolling exit-cost) has no equivalent for prediction markets. Practitioners rebuild it in spreadsheets each quarter. Most abandon the build by the second month.
Funds, prop traders, and market makers sizing positions in the tens to hundreds of thousands across Polymarket and Kalshi today on improvised tooling. The audience that has the spreadsheet and would prefer not to.
Same event, different venue, different price. nijinn ranks spreads net of fees, slippage, and resolution-rule divergence. Annualised return per arb, not headline gap. Five arb types classified on entry.
Read methodology →Log P_user before you look at the market. nijinn scores edge against consensus, sizes the trade with Quarter Kelly, tracks calibration drift bucket by bucket. Six sources of mispricing flagged on entry.
Read methodology →Fair value across 4 methods (mid, weighted-mid, microprice, trade-flow VWAP). P&L decomposed into spread, rebate, inventory, adverse selection. Toxicity alerts when the flow turns informed.
Read methodology →Brier, BSS, calibration curve, MaxDD with Triple-Penance recovery, VaR, Sharpe, Sortino. Amihud-rolling for exit cost. The category's first answer to 'am I good at this' and 'can I get out at fair value'.
Read methodology →Hedge tariff exposure, weather risk, policy outcomes, geopolitical events. Notional sized on actual correlation between contract and underlying, not 1:1. Right size is the deliverable.
Read methodology →Signal-to-noise is a 1948 information-theory term, Shannon-era. Nate Silver carried the frame into prediction discipline with The Signal and the Noise. Silver describes the act: find the signal in the noise. nijinn does the act: turn noise into signal. The verb is the difference.
Five Pro surfaces, each one a transformation. Scattered prices into ranked spreads. Raw confidence into calibrated edge. Individual trades into measured skill.
Predictive Labs is building the intelligence layer for prediction markets, the native asset class of event-driven finance, a $64 billion sector in 2025. The Singapore-based company applies AI and graph analytics to turn fragmented venue data into canonical event-level intelligence, serving traders, funds, developers, and AI agents at institutional scale. Founded by Johann Evrard. Backed by Coinsilium Group (AQSE: COIN).
Visit corporate site →A live workspace opens to a closed cohort a few weeks before the wider release.
Time until first access
24 July 2026 · 12:00 UTC
Public launch follows in September 2026.