Whale Watch: large prediction-market trades, in context

What Whale Watch is

Prediction markets publish their trades. When someone puts thousands of dollars on one side of a game, that's visible. Whale Watch collects large fills from the public Kalshi and Polymarket trade feeds and asks a practical question: can you still get a price as good as the whale got, and what would that price be worth if the whale's average price were fair?

Coming soon Whale Watch is in build. The signal math already runs in the engine; the screen and the running record come next.

Which fills count

A trade is a whale when its notional (price × contracts) is at least the threshold, $5,000 by default. For one outcome, take the best retail net price still available, d*, across the venues you can use, after fees. Its break-even probability is b* = 1 ÷ d*.

Fills at better prices than you can get don't count. They show where the market was, not a price you can match.

Worked example

Three large fills on one side: 20,000 contracts at 55¢ ($11,000), 15,000 at 56¢ ($8,400) and 10,000 at 52¢ ($5,200). The best retail price still available is −115 (decimal 1.8696), so b* = 53.49%.

The whale fills are before fees: they're the prices paid on the exchange. Your side is your net price after fees.

Fill by fill, and both sides

Now suppose the other side also drew a large fill: 12,000 contracts at 46¢, $5,520. The two-sided bar compares the whale money on each side: $24,600 against $5,520, a split of 81.7% to 18.3%. One-sided money is a clearer signal than money on both sides; a near-even split means the large traders disagree.

Signals are grouped into size buckets ($1k+, $5k+, $25k+, $100k+) so the record can show whether bigger money has meant anything, once there are enough graded signals to say.

What the screen will show

• A card per play with the whale dollars behind it, the edge, the matchup and the best venue.
• The individual fills, with a break-even divider separating fills that count from fills that don't.
• A "Where to bet" table of every venue on both sides, best first, net of fees.
• A running record: every signal logged once at first sight and graded later, wins and losses, by size bucket.

How to use a whale signal

1. Look at the counting dollars, not the total. In the example, $24,600 traded but only $19,400 was at prices you can still match.
2. Check both sides. Large money on both sides of the same market is a disagreement, not a signal.
3. Compare with the market fair. If the +EV finder's consensus fair agrees with the whale's average price, the edge is better supported than if only the whale's price says so.
4. Use your net price. The retail edge is computed on the best net price after fees at a venue you can use.
5. Size modestly. A whale's conviction isn't your edge estimate. Use your normal Kelly fraction or less.

What Whale Watch won't tell you

It won't tell you who the trader is or why they traded: we strip trader identity fields from the data and show only the trades. It won't tell you about sportsbook money: books don't publish their handle. And it can't tell you whether large prediction-market traders are right more often than the market. That's an open question, and the graded record exists to answer it with data rather than a story.

Pitfalls

• A big bet can be a hedge. The trader may be offsetting another position. Size is context, not a reason by itself.
• The whale's price isn't proof of fair. Treating the VWAP as fair is an assumption. Compare it with the market's fair price on the +EV finder.
• Thin exchange markets. The best retail price may be good for only a small amount.
• Stale prices. By the time you see a fill, the market may have moved toward it.
• Pregame only. We only track trades before the start.