Quality ≥ 3, whenever it first holds
by PConsensus · Oct 25 2026 · Site export · done
Study rule: Buy the consensus side at the first poll where the consensus has 60 %+ of 5+ holders (price 0.05-0.95) AND quality.net >= 3, even if that happens after the consensus first formed. Hold to resolution. Study verdict (Oct 22 2026, skill): A- (the engine-grid version: weaker, the most robustness-tested)
- Closed trades
- 1,620
- Mean return / $1
- +3.1 %
- −0.4 % to +6.5 %
- Hit rate vs break-even
- 79 % vs 67 %
- In / out of sample
- +2.8 % / +2.8 %
- 984 / 543 trades
Buy the side 60 %+ of 5+ tracked wallets agree on, in any market, at any price (0.05–0.95) and at least 3 more quality holders (win rate ≥ 60 % over ≥ 20 settled, $100+ in the market) for it than against it, bought as soon as the rules hold, held to resolution; after a 1¢ half-spread and Polymarket's taker fee, $1 a trade, over the dataset's whole window. Over 1,620 closed trades between Mar 10 2026 and Oct 22 2026 it made 3.1 % per $1 staked (95 % interval −0.4 % to +6.5 %). It won 79 % of its trades and needed 67 % to break even at its entry prices.
The mean is positive, but either the in-sample half is not significant or the out-of-sample half disagrees. Picking the best of many variants produces results like this by chance. Past results, not advice.
Plain consensus, all markets on the same data and period: −2.4 % per $1 over 11,908 trades (−4.1 % to −0.7 %). This rule did 5.5 points better per trade, with non-overlapping intervals.
The numbers
| Per $1 staked, after costs | All | In-sample (to Aug 10 2026) | Out-of-sample (after) |
|---|---|---|---|
| Closed trades (events) | 1,620 (1,020) | 984 (542) | 543 (456) |
| Mean return per trade | +3.1 % | +2.8 % | +2.8 % |
| 95 % interval | −0.4 % to +6.5 % | −1.1 % to +6.7 % | −3.9 % to +9.4 % |
| Hit rate | 79.2 % | 81.5 % | 73.9 % |
| Break-even hit rate | 67.1 % | 68.4 % | 64.0 % |
| P/L at $1 stakes | +$50 | +$27 | +$15 |
| Largest drawdown | $19 | $19 | $13 |
| Most stakes open at once | 311 | 297 | 77 |
| Average entry price | 0.76 | 0.78 | 0.71 |
| Average hold | 20.7 d | 26.0 d | 3.5 d |
| p (mean > 0, one-sided) | 0.038 | 0.081 | 0.208 |
Period Mar 10 2026 – Oct 22 2026; markets ending before Oct 15 2026 count; split at Aug 10 2026. 13 positions still open at the end of the data are excluded. Interval: cluster bootstrap (events resampled).
Cumulative P/L at $1 stakes
By topic (the 10 most traded)
| Topic | Trades | Hit | Mean / $1 | P/L |
|---|---|---|---|---|
| Geopolitics | 533 | 80 % | +0.6 % | +$3 |
| Soccer | 130 | 74 % | +12.3 % | +$16 |
| Crypto | 125 | 85 % | +7.0 % | +$9 |
| Finance | 122 | 82 % | +2.7 % | +$3 |
| Esports | 116 | 75 % | −0.3 % | −$0 |
| Elections | 114 | 79 % | +2.0 % | +$2 |
| Politics | 77 | 83 % | −7.2 % | −$6 |
| Tech | 58 | 83 % | −0.2 % | −$0 |
| Other | 55 | 91 % | +14.4 % | +$8 |
| Tennis | 53 | 77 % | +21.2 % | +$11 |
By entry price
| Price band | Trades | Hit | Mean / $1 | P/L |
|---|---|---|---|---|
| 0.05-0.30 | 56 | 23 % | −3.7 % | −$2 |
| 0.30-0.50 | 134 | 49 % | +13.7 % | +$18 |
| 0.50-0.70 | 298 | 64 % | +3.1 % | +$9 |
| 0.70-0.95 | 1,072 | 89 % | +2.6 % | +$28 |
| 0.95-1.00 | 60 | 92 % | −4.6 % | −$3 |
By month of entry
| Month | Trades | Hit | Mean / $1 | P/L |
|---|---|---|---|---|
| 2026-03 | 323 | — | +2.5 % | +$8 |
| 2026-04 | 271 | — | +1.1 % | +$3 |
| 2026-05 | 152 | — | +7.9 % | +$12 |
| 2026-06 | 160 | — | +3.0 % | +$5 |
| 2026-07 | 146 | — | +7.2 % | +$11 |
| 2026-08 | 92 | — | −16.0 % | −$15 |
| 2026-09 | 46 | — | +0.8 % | +$0 |
| 2026-10 | 430 | — | +6.1 % | +$26 |
Best trades
- Will there be at least 1900 measles cases in the U.S. by April 30, 2026?No at 0.16 · entered Apr 7 2026 · won+465 %
- Braga: Zsombor Piros vs Elmer MoellerElmer Moeller at 0.17 · entered Oct 8 2026 · won+434 %
- Trump approval Up or Down this week?Down at 0.19 · entered Mar 20 2026 · won+385 %
- U.S. anti-cartel operation outside of the U.S. by July 31?No at 0.20 · entered Jul 29 2026 · won+358 %
- Suzhou: Tatiana Prozorova vs Kyoka OkamuraKyoka Okamura at 0.22 · entered Oct 7 2026 · won+319 %
Worst trades
- Japan Open Tennis Championships: Carlos Alcaraz vs Jiri LeheckaJiri Lehecka at 0.43 · entered Oct 5 2026 · lost−100 %
- Spread: Atalanta BC (-1.5)Venezia FC at 0.59 · entered Oct 12 2026 · lost−100 %
- Will Tisza win the national list vote in the 2026 Hungarian Parliamentary election by 9%+?No at 0.55 · entered Apr 9 2026 · lost−100 %
- Spread: Packers (-3.5)Bears at 0.67 · entered Oct 11 2026 · lost−100 %
- FC Bayern Munchen vs. Virtus BolognaFC Bayern Munchen at 0.76 · entered Oct 8 2026 · lost−100 %
Read before believing it
- Trying many variants finds lucky ones: the more rules you test, the more of them will look good by chance. A single Promising run is a hypothesis for a paper bot, not a proven edge.
- Returns are per $1 staked, equal stakes, prices from the tracked wallets' own fills (median error about half a cent, more in fast markets), entries and exits at hour boundaries. The live order book will differ.
- The in-sample half is the first 70 % of the period by time, the out-of-sample half the last 30 %; a trade belongs to a half when it was entered in it and its market ended before the half closed.
- Past results, not advice.
- Only the fills of wallets the site tracked, from the day it started tracking them: curated wallets from Mar 10 2026, the leaderboard set from Oct 2 2026. A position a wallet opened before it was tracked is invisible.
- Holders are rebuilt from on-chain fills; the live site also reads Polymarket's position answers, so the live consensus can differ (85–92 % side agreement on tradable markets).
Quality ≥ 3, whenever it first holds
Study rule: Buy the consensus side at the first poll where the consensus has 60 %+ of 5+ holders (price 0.05-0.95) AND quality.net >= 3, even if that happens after the consensus first formed. Hold to resolution. Study verdict (Oct 22 2026, skill): A- (the engine-grid version: weaker, the most robustness-tested)
Past results, not advice. Numbers are per $1 staked after the run's costs on the tracked wallets' own fills; the live order book will differ. Clone this run and change one thing.