PolyEdge — the edges we tested and killed, with the numbers.

01 Live board

—

A live sharp-line divergence board.

Devigged Pinnacle fair value (Shin method) vs Polymarket's actual asks, pre-match main lines only. Taker edge = fair − (ask + fee). Maker edge = fair − (bid + 1¢), zero fee. Snapshots every 15 minutes. The historical version of this comparison failed its gate (see below) — the board runs on as live monitoring, and as a public record of how tight these markets actually are.

Game	Taker edge ▼	Side	Fair	Ask	Maker edge	Depth $
loading tape…

Edge across the board, right now

taker edge per priced leg — most of an efficient market sits below zero

Divergence tape over time

per 15-min snapshot, accumulating toward the gate sample

02 The gates

no gate, no capital

Passing is a number, not a feeling.

Each strategy is off until it clears a pre-registered statistical gate. Passing is a number, not a feeling — and a fail gets published just as loudly.

03 The graveyard

edges we tested and killed

Every edge we tested and killed.

Every number below came from our own cluster-clean tests on the full archive — honest costs, no midpoint fantasies, inference clustered by event. Click a card for the failure anatomy. A published null is a result, not a failure.

04 Dispute log

why we grade the fine print

When the rules pay against the crowd.

A Polymarket bet doesn’t pay on what happened — it pays on what the resolver decides happened. Disputed markets go to UMA, an open “optimistic oracle” where token-holders vote on the outcome over 48–72 hours, voting power weighted by tokens staked. Most of the time it matches reality. But the vote can turn on a single word in the rules, and a handful of large holders can swing it — sometimes while holding a position in the very market they’re voting on. These are real, widely-reported cases where the payout went against the obvious answer. This is exactly why the checker flags a market with no named source, or one that’s still being settled.

≈ $242M traded

“Will Zelenskyy wear a suit before July?” (2025)

He appeared at a NATO summit in a black jacket, matching trousers and collared shirt — described as a suit by numerous outlets. The market first leaned Yes, then resolved No, with UMA citing no “credible reporting consensus.” Over half the volume traded after it should have settled.

→ Paid “No” against the obvious read.

Reported by Decrypt · CoinDesk

tens of millions redistributed

Israel–Iran “ceasefire before July?” (2025)

The rules only counted a mutually-agreed, general halt — and spelled out that a unilateral pause, a humanitarian pause, or an informal agreement would not qualify. When a pause arrived that didn’t fit that exact definition, traders who bet on the plain-English “ceasefire” found the fine print decided it, not the headlines.

→ The definition, not the news, set the payout.

Reported by Calcalist · rules on Polymarket

the pattern, not a one-off

Why one clean case isn’t the point

These aren’t freak accidents — they’re what happens when a market’s wording leaves room and the money is big enough to fight over. Critics note UMA’s token market cap has at times been smaller than a single market’s volume, so the incentive to sway a vote is real. Before you bet, the only defence is reading the resolution rules and asking: could the fine print pay against the obvious answer here?

→ Read the rules before the headline.

Our checker’s “fine print” section does this scan for you on any market.

05 Method, hands-on

play with the actual models

The same code that powers the board, in your browser.

Devig any price line and cost any trade with the exact functions the desk runs.

Devig playground

American odds in → fair probabilities out, three methods. Shin is the board's default; multiplicative is the conservative floor.

Home

Draw (blank for 2-way)

Away

Method	Home	Draw	Away

What a trade really costs

Polymarket taker fee: rate × p×(1−p) per share — verified empirically against 78,525 on-chain FeeCharged events (100% reconciliation; rates sit on a per-market grid, taker-only, no maker fee). Docs were ambiguous; the chain wasn't.

contract price 0.50

—taker fee / share

1.0¢typical spread cost

—edge needed to break even

06 The tape

recording since 2026-07-05 — books aren't recoverable from the chain

Recording every fill forward, with gap detection.

The free historical tick source for Polymarket died in January 2026 (fills remain backfillable on-chain; resting order books do not). Our recorder captures every fill forward, with gap detection, alongside book snapshots — the raw material for maker-fill measurement and flow research.

Fills per 5 minutes, last 6 hours

whole-exchange trade flow (5-minute crypto markets dominate)

Total fills

—

since recorder start

Last hour

—

fills recorded

Feed lag

—

API indexing delay

Detected gaps

—

honesty counter — zero is the goal

07 House rules

what makes this trustworthy

What makes this trustworthy.

Executable prices only

Backtests mark at asks plus fees plus lockup cost — never midpoints. If a fill can't be proven possible, it doesn't count. Phantom marks are how paper records lie.

Cluster-clean inference

Hundreds of markets resolve on one real-world event. Every t-statistic is clustered by event and by resolution week — the two ways backtests fake significance.

Nulls get published

The graveyard is permanent. A strategy that fails its gate stays visibly dead — it doesn't get quietly re-tuned until it looks good.

Pre-registered gates

Thresholds are written down before the data comes in: ≥1.5%/trade net, week-clustered t ≥ 2, minimum sample sizes. Moving a goalpost is a public event.