Reversal Index

The baseline, and why it proves nothing

Before committing to anything forward-looking we scored 83 positions that had already resolved. Whale-backed outcomes landed 66% of the time. The market was quoting 59% when they bought. That is an edge of 7.1 points — about 6 extra wins over 83 bets.

A 66% hit rate reads well, and it would be easy to put it on a chart and stop there. It does not hold up. The 95% confidence interval runs from 56% to 76%, and the market's own 59% sits inside it. At p = 0.16, chance explains the gap comfortably.

This is the question our first backtest could not answer: back then only three markets were both genuinely contested and heavily whale-funded, which is no sample at all. It is 83 now, and the answer is still the same one.

There is a second, worse problem. We chose the selection bars after looking at this data. Criteria fitted to a sample will always flatter it, so no backtest run this way can settle the question, however large it gets. That is the reason the live ledger exists at all — it is the same measurement with the fitting removed.

Method

• What counts as whale flow. Individual fills of $10k or more, from the on-chain trade tape.
• What makes a position. A market needs at least $100k of such fills across 3 or more wallets, with at least 65% of bought notional on one side.
• It has to be a real call. The notional-weighted entry price must sit between 15¢ and 85¢. Buying at 97¢ is not a prediction, and including it would inflate the hit rate without adding information.
• Grading. We have no official resolution feed, so a position is scored when one outcome's last print reaches 97¢. Markets that go quiet below that stay open instead of being scored against whichever side happened to be ahead. This is a proxy for resolution, not an oracle.
• Nothing is revised. Once recorded, a position's market, side, entry price and date are frozen; the only fields that can ever change are the outcome and the date it landed.
• What this is not. Not profit and loss — it ignores position sizing, fees and exits. Not advice. Wallets are identified by address only; no attempt is made to say who is behind one.

Data as of 2026-09-28. Related: whale activity by market · whale index