market-neutral · model-free · on Canon

Enforce the math the market forgot

An AI-driven arbitrage engine for prediction markets. It never predicts who wins — it harvests the instants when a market's own prices break the logic that connects them.

P(champion) ≤ P(conference) ≤ P(series)  ·  Σ P = 1  ·  YES + NO ≥ 1

··· live markets 2 venues · lattice nodes 0 forecasts
live lattice · node size = implied probability
beyond prediction

A different kind of edge

Model-free

No forecasts, no alpha. The edge is pure internal price inconsistency — the market contradicting itself.

Market-neutral

Every trade is a hedged basket with a locked, non-negative payoff regardless of the game result.

Multi-venue

Watches the same market on Polymarket and Kalshi and arbs disagreements between independent books.

AI-classified

A free-tier LLM maps any market into the lattice, so the same engine generalizes to any sport or event.

600+live markets scanned
30teams in the lattice
2venues, fee-aware
43passing tests
how it works

Three layers of coherence

L0 · WITHIN-MARKET

Complementary

If ask(YES) + ask(NO) < 1 on one market, buy both sides for a guaranteed dollar.

L1 · SINGLE EVENT

Dutch book

The 30 title outcomes are exhaustive. If Σ ask(YES) < 1, buy them all for under a dollar.

L2 · CROSS-EVENT

Implication

Champion ⊑ conference. When P(champion) > P(conference), short the dear leg, buy the cheap one.

Why it's risk-free: if a team's championship price a exceeds its conference price b, buy NO(champion) at (1−a) + YES(conference) at b. The minimum payout in every outcome is $1 (the "wins title but not conference" case is impossible) for a cost of (1−a)+b — a guaranteed a − b, with no view on the game.
the frontier

Same market, two venues

Polymarket and Kalshi each run the 2026 NBA Champion market on independent order books. When they disagree beyond fees, that gap is a locked cross-venue arbitrage — shown live below.

arbiter // live dashboard

Cross-venue (live)

Coherence

Open the full dashboard