Surf The Singularity
Back to Blog
Surf the Singularity #029
June 2026

Five Ways a Market Fakes an Edge

Five prediction-market screens lit up with what looked like free money. Five times the green was an artifact of how I was measuring, not a hole in the market — here are the traps, and the single check that catches each one before a cent moves.

> a screen lights up: the prices of every outcome in a market group add to less than a dollar. "free money."
> don't buy. first check whether the outcomes are actually mutually exclusive and exhaustive.
> they overlap, or there's an unpriced "none of these." the arb is an accounting error, not an edge.
> real partitions sit ABOVE a dollar — that gap is the market-maker's vig, the toll you'd pay.
> trap two: my own forecast disagrees with the market price. feels like alpha.
> but my model was fed a subset of what the price already knew. the disagreement was my blind spot, not the market's.
> my error against reality: 0.485. the market's: 0.287. orthogonal to the truth, not ahead of it.
> trap three nearly got me: sell cheap longshots, collect the lottery-ticket premium. +32% a year on paper.
> then i priced the actual order book. the entry my backtest needed doesn't exist to trade against.
> trap four: an eight-from-eight win streak at a 95% price. a flawless record.
> 0.95 to the eighth is 0.66 — winning all eight is the EXPECTED outcome. eight coin flips, not a signal.
> trap five: an 87%-win bot that still lost money. every dollar of loss sat in one price tier.
> i wasn't unlucky there — i was getting picked off. the win rate hid an adversely-selected leak.
> five green lights, five structural illusions, zero dollars wired. the scanner's real output is the red light.

A liquid market is a machine for eating your information and pricing it back to you. So when one of them looks like it's handing out free money, the base rate is overwhelming: you're not staring at a hole in the market, you're staring at a hole in how you measured it. Anyone's scanner finds the green light. The skill is the boring check that throws the red one a beat later — before you fund anything. Five traps, five checks. Steal the checks.

The Receipts · By the Numbers
FREE MONEY, OR THE MARKET-MAKER'S VIG? $1.00 scanner-flagged "arbs" — legs overlap or miss a "none" outcome ~$0.45 genuine partitions — exactly one outcome can win $1.04 – $1.10
The flagged "underrounds" weren't partitions — overlapping thresholds and unpriced residuals fake the whole signal. Screen for sets where exactly one outcome must win and the sums sit at $1.04–$1.10: the vig you'd pay, not free money you'd collect.
THE PRICE YOUR BACKTEST NEEDS DOESN'T EXIST 0% +32% / yr backtest entry $0.92 ~0% fillable entry ~$0.96
The cheap longshot resolved YES only 2.4% of the time vs an 8% break-even — a real edge, on paper. But NO is just 1−YES: the $0.92 fill the +32% relied on isn't in the book. At the ~$0.96 you can actually pay, the edge evaporates.
EIGHT-FROM-EIGHT: A RECORD THAT KNOWS NOTHING break−even 95% confidence interval on this bet's true per-trade return −29% +5%
Winning 8 of 8 at a 95% price is the most likely outcome of an unremarkable bet — 0.95⁸ = 0.66. The interval on its true return runs −29% to +5%: the record is consistent with a great edge and with a terrible one. It knows nothing.

None of these were sophisticated frauds. Each was an ordinary, structural reason a measurement overstates an edge: overlap faking an arbitrage, a subset of information faking a forecast, an unfillable price faking a yield, a small sample faking a streak, an average hiding a leak. Stack them and the lesson is almost rude in its simplicity — on a liquid market, "free money" on the screen is nearly always a fact about your measurement, not the market. The scanners weren't useless. They were disconfirmation tools wearing alpha's clothing.

Appendix · Further Signal

The Verdict

None of the five was a real edge — each was a measurement error wearing alpha's face. That's the trap worth naming: an edge is the easiest thing in the world to fake by accident, and the sharper the tool that finds one, the more confidently it sells you the mistake. Verify every edge against the structure before you trust it — AI included, especially AI.