What is the Luck Share?

The answer first

The Luck Share takes a cohort of wallets that a leaderboard presents as the best, runs the same frozen skill test on every wallet with enough resolved positions to read, corrects for the number of tests run, and reports how much of the cohort does NOT survive. On the 2026-06-09 reading of Convexly's own published top-50 Polymarket cohort, the Luck Share is 100 percent: 0 of the 35 testable wallets clear the frozen, FDR-corrected bar, and 15 of 50 had too few resolved positions to test at all.

Read that carefully, because it is easy to over-read. The claim is NOT that these wallets have no skill. The claim is that at these sample sizes their resolved records cannot be told apart from chance at the corrected bar, in either direction. That honest limit is the finding.

Worked example: the 2026-06-09 reading

The published top-50 cohort breaks down like this in the frozen scan:

• 15 of 50 had fewer than 30 resolved positions: too thin to test, not counted as testable.
• 19 of 35 testable read as not separable from chance: their bootstrap 95 percent intervals include zero.
• 15 of 35 testable were reported by that scan as too concentrated to read: the single biggest winning event is at least 60 percent of the net result, and often larger than the net result itself, because the ratio is a dollar figure divided by net PnL and is not capped at 100 percent. (The frozen scan predates the 2026-07-15 reframe; under the current machine that threshold is a descriptive style label, not a demotion.)
• 1 of 35 testable had a positive interval before correction, roughly what chance predicts across 35 tests (about 0.9 false positives expected at a 2.5 percent one-sided threshold), and it does not survive the multiple-comparisons correction. Its net PnL on the board is negative.

Cleared after correction: 0 of 35. That is a count over a closed set of 50 addresses selected once on 2026-04-19 from a 2026-04-15 snapshot and never added to, not an estimate of how often Polymarket wallets clear. The count over wallets is a census; the read inside each wallet is not, since every wallet carries a 95 percent interval over its own resolved positions and the correction controls errors across these 35 tests only. The zero does not rest on the concentration screen or on an independence assumption: dropping the concentration screen and correcting across all 35 tests still clears none, and neither does a correction that stays valid when wallets trade the same events.

Why the number exists

Most prediction-market trackers rank wallets by raw profit. At the sample sizes real wallets have, profit does not separate skill from variance: a fat-tailed market hands large realized PnL to some wallets by chance alone. The Luck Share is the inverse of a leaderboard. Instead of asking "who made the most?" it asks "how much of the cohort survives a corrected skill test?" and publishes the answer even when the answer is zero. Publishing the zero is what makes the bar worth anything when a record eventually clears it.

What the Luck Share does NOT claim

It does not claim any individual wallet is lucky. It does not claim the cohort has no skill. It does not forecast future performance: a past read is not a forecast. And it is not a verdict on the 15 concentration-heavy wallets, whose issue is different: their records are dominated by a single event, a composition fact that more positions would not change. The precise register for each state is fixed in the lexicon.

Where the readings live

The dated reading ledger is at /research/luck-share. The full per-wallet table behind it, with every interval and the one uncorrected positive record, is at /research/top50-skill-scan. Both are built from the same frozen artifact, so they can never disagree. Diagnostics on public on-chain records, not investment advice.

Check a specific wallet

Paste any Polymarket wallet into the free analyzer. It returns the same skill-vs-chance read with its interval, not a profit screenshot.