Smart Money Consensus — What It Means When Five Whales Agree

The one-whale problem

Top-50 whales are selected on historical PnL. That means, by definition, they are winners so far — but it doesn't mean any individual trade is right. Even an 80th-percentile trader is wrong 35–40% of the time on individual bets. Tailing one whale is noisy because you inherit their individual variance.

Why overlap changes the math

If Whale A (uncorrelated to Whale B) bets YES on Market X with 65% accuracy, and Whale B (uncorrelated to A) independently bets YES on Market X with 65% accuracy, the joint probability that both are wrong is roughly (35% × 35%) = 12%. Add a third uncorrelated whale, and joint-wrong drops to ~4%.

That math has an important caveat: “uncorrelated”. In practice whales in the same category (e.g. politics specialists) are correlated — they all read the same polls. Three politics whales on the same market is less of a signal than three whales from three different categories on the same market. We don't currently weight for this (transparent known limitation); it's the next methodology iteration.

How we actually compute it

Every five minutes, we:

1. Pull the top 50 wallets by all-time PnL from the Polymarket leaderboard API.
2. Fetch every wallet's currently-held positions (filtering out resolved markets — redeemable: true or endDate < now).
3. Group positions by (market_id, outcome). For multi-outcome markets (candidate lists), each named outcome gets its own group.
4. Keep only groups with 2+ whales.
5. Rank by whale count descending. Tiebreak by total $ at risk across the whales.

The implementation lives in /app/lib/services/consensus.js. It's one file, deliberately small, and we publish it as-is in the methodology page.

What you see on the dashboard

Each consensus card shows: the market question, the dominant outcome (e.g. YES / a specific candidate name), the number of whales agreeing, the % of the top-50 cohort that represents, the total $ stacked across those whales, the current live $ value, and a drill-down showing every whale's individual $ / entry price / unrealized PnL.

Cards are color-graded by cross-platform confluence: green when the matched Kalshi market's flow leans the same way, amber when it diverges, gray when Kalshi sample is too thin to compare.

The honest caveats

• Cohort selection bias. The top-50 are historical winners. Next year's top-50 will be a different set of wallets. Our consensus is the consensus of last year's winners, which is a noisy predictor of who will be right next year.
• Correlated bets. As above, category-specialists often agree because they read the same news.
• Sample size inside a market. 3 whales is not a lot. 5+ is much stronger. We surface the count prominently so you can weight accordingly.
• Price drift after whale entry. Whales got in at an average of 58¢; by the time you see the consensus on our page, the price may be 64¢. That 6¢ tax eats some of the edge — always look at the weighted average entry vs. current.

Live track record

We publish the cohort's aggregate numbers at /track-record. Today's snapshot: +3.5% aggregate ROI on $6.55B lifetime volume, median whale ROI +8.6%, open-position win rate 63.2%.

We are also snapshotting daily consensus data to a database from June 2025 forward. Once we have 90+ days of snapshots with resolved markets, we'll publish a dedicated consensus hit rate and a calibration curve — does a 60¢-entry consensus actually resolve 60% of the time, or higher?