The method

About Sharpe

Sharpe reads a wallet’s public prediction-market history on Polymarket and Hyperliquid HIP-4, computes a deterministic reputation score, and renders a shareable card. The score is the product: a fixed, open formula over public on-chain data that anyone can recompute.

§ 1

What the score measures

Most leaderboards just count wins. That rewards buying heavy favorites and tells you nothing about skill. We measure edge: were you right, and were you right when the market priced your side as unlikely?

For each resolved prediction we take c, the price you paid for the side you took (its implied probability, 0–1), and w, whether that side won (1 or 0). Your raw edge is the average of (w − c) — profit per $1 share. Buy a winner the crowd priced at 15¢ and you earn a big positive; pay 90¢ for a loser and you’re punished.

Only markets you held to resolution count. A position you opened and closed before the market settled has no outcome to grade, so it’s left out — as are markets that haven’t resolved yet. That’s why your resolved-call count can be lower than the “markets traded” figure on your Polymarket or Hyperliquid profile: those tally every market you touched, while we score only the ones that actually settled while you were holding a side.

§ 2

Why a small sample can’t fake it

Three lucky calls shouldn’t outrank a 180-of-250 grinder. So instead of dividing your total edge by your number of predictions N, we divide by N + 15. Those 15 pseudo-predictions shrink tiny samples toward zero until you’ve made enough real calls to prove the edge.

On top of that, anyone with fewer than 10 resolved predictions is Unranked — no score until the sample is real.

The headline
score = clamp( round( 50 + 250 × adjEdge ), 0, 100 )

We also show calibration (Brier) as a secondary signal — lower is better — but it isn’t the headline, because pure calibration penalizes the skillful move of buying a cheap winner.

§ 3

The tiers

UnrankedN < 10

Not enough resolved predictions to tell skill from luck. We need at least 10 settled markets before a score means anything — three good calls don't make an oracle. Keep making calls and the rank unlocks once the sample is real.

Oracle85–100

The top tier. Calls that are consistently right — and right when the market said they shouldn't be. An Oracle doesn't just win; they buy the outcomes the crowd underprices and watch them resolve true. Over a real sample, that's the signature of genuine edge, not a hot streak. Rare, and earned.

Sharp70–84

A clear, repeatable edge over the market. Sharps land on the right side more often than the prices imply, and get paid for it across enough markets that luck isn't the explanation. Not flawless — reliably ahead of consensus. The kind of track record other traders quietly follow.

Solid55–69

Above the crowd, with a modest but real edge. Solid predictors are right a little more often than the odds suggest — enough to come out ahead over time, not enough to retire on. An honest track record that beats coin-flipping, with room to climb.

Coin Flipper45–54

No demonstrable edge — results that look indistinguishable from guessing. The market priced these outcomes about as well as the predictor did, so the wins and losses roughly cancel. Not bad, just not better than chance. The score isn't an insult; it's a mirror.

Degen0–44

On the wrong side more often than not — paying above fair value and watching it resolve the other way. Over a real sample, a Degen's calls lose to the market's own pricing: conviction outrunning accuracy. The upside is there's nowhere to go but up, and every Oracle has a few of these markets buried in their history.

§ 4

Trust & safety

  • You stay in control of your wallet. You connect a self-custodial browser wallet (MetaMask, Rabby, and others) through RainbowKit. The app never sees or stores your private keys and cannot move your funds.
  • We only ever ask you to sign one thing: an exact USDC transfer to pay a small fee. Never an approval, never a permit, never an arbitrary contract call. A direct transfer grants no spending rights — there is no drainer surface.
  • We read only public on-chain data from an address you supply. “Verified” means anyone can recompute the score from the same public data — not that we proved you own the address.
  • Payment is verified server-side. We confirm the transfer on-chain before unlocking generation; we never trust a client’s claim that it paid.
§ 5

Disclaimers

Not financial advice. For entertainment and informational purposes only. Scores are derived from public on-chain data and may be incomplete or delayed.

Sharpe is not affiliated with or endorsed by Polymarket, Hyperliquid, or the Hyper Foundation. All trademarks belong to their respective owners. We never custody funds or keys.