Settled prediction-market arbitrage results

Correction, 23 September 2026 — the headline figures above have been restated. Polymarket charges takers rate × price × (1 − price) per share, and since 30 March 2026 that fee has applied to nearly every category (politics and finance 4%, sports and economics 5%, crypto 7%; geopolitics free). Our scanner carried each market's fee flag but never charged it. Every position in this record has a Polymarket leg and was entered after 30 March (the earliest on 14 July 2026), so the fee applied to all of them — and the figures we had published left it out. The headline now charges it at 4%, the lowest rate that applies to most of these markets. Positions in fee-free geopolitics markets did better than shown; crypto markets did worse. It is fixed for every new position.

"As published" is the figure this page showed before the correction, kept so the size of the error is visible. What it means in plain terms: before the fee, the locked-in record looked like a small, steady edge; after it, the single-venue baskets lose money and only the cross-venue pairs stay positive — and those have broken twice in the settled record.

Check this record yourself. Every stock and commodity signal is sealed when it is issued: each seal is a SHA-256 of the previous seal plus the signal's own terms, so changing or deleting any past signal breaks every seal after it. The chain head, the rules version and the method are public at /api/stocks/attest — no login. Prediction-market positions below are settled by the venues themselves, and the losers are listed with the winners.

No annualised figure is published here. Multiplying a short hold up to a year assumes a year of rows like it, and the board finds a handful a day — what limits this is how many positions exist, not how much money you have.

What the locked-in record is made of:

• Single-venue baskets — 342 settled, 0 broke, −0.56% per position after the Polymarket fee. There is no edge here to protect: the fee alone is larger than the spread these positions captured. Every leg is a Polymarket taker order, so every leg pays the fee.
• Cross-venue pairs — 58 settled, 2 broke, +1.23% per position after the Polymarket fee. One break in 82 would wipe that out, and 58 settled positions only narrow the true break rate to 1 in 9 at best (95% confidence) — which is not yet enough to call the edge proven. These pay unless the two venues resolve the same question differently — the failure that has actually occurred.

How a cross-venue hedge works

Two venues price the same event independently. When they disagree far enough, you buy YES on one and NO on the other for a combined cost under $1.00. If both venues settle the question the same way, exactly one leg pays $1.00 whichever way the world goes, and the profit is locked in the moment both legs are bought.

Every position below actually closed and was resolved by the venues themselves - settled outcomes, not projections or simulations. Figures are stated per $1,000 committed, net of fees. Of 283 recommended positions, 259 held and 24 broke; the total above includes both.

A hedge breaks when the two markets turn out not to be the same event, or when the two venues resolve the same fact from different sources — then both legs can lose. The matching engine flags every cause we have identified, but broken hedges have continued to occur. That is why the board's default view shows only locked-in pairs (low match risk at entry): 398 of 400 have held so far. That is a track record, not a guarantee. Past resolutions do not predict future ones, and nothing here is financial advice.