How the Skill Score works (v0.2)
Every wallet is scored from its last 180 days of closed and resolved positions, sampled in time order (not by profit), so the sample isn't cherry-picked.
| Part | Weight | What it measures |
|---|---|---|
| Edge | 40% | Half: t-stat of per-position returns, trimmed at the 5th/95th percentile so one moonshot can't carry it. Half: calibration t-stat on resolved markets — did they buy below the real chance of winning? |
| Robustness | 20% | ROI after removing the 3 best positions; full marks at 20%+. |
| Breadth | 15% | Share of traded categories that are profitable (on a category board: share of profitable events). |
| Consistency | 15% | Share of active months in profit. |
| Sample | 10% | Number of markets, full marks at 1,000. |
Not ranked
- Not enough history: fewer than 20 markets.
- One-hit: one position is over 50% of gross profit.
- Market maker: rebates and rewards over 35% of profit, turnover over 150× profit, or over 30% of markets held on both sides.
Beat the odds
The average of outcome minus entry price on resolved markets, in points. Positions sold before resolution count at half weight and are measured at their exit price. A trader who buys at 60¢ and wins 70% of the time beats the odds by 10 points. Since v0.2, nobody with negative beat-the-odds (on 10+ resolved positions) is shown above 3 stars.
Proof: how cohorts are picked
The Proof scoreboard opens two hypothetical $1,000 paper portfolios every Monday 00:00 UTC: one follows the top 10 traders by Skill Score (wallets flagged by the gates above are excluded), the other the top 10 by raw monthly PnL.
- Recent activity required (both arms): a trader is eligible only if they made at least one buy in the 14 days before the cohort's Monday 00:00 UTC. The check only looks at trades before that moment, so nothing that happens after the cohort opens can affect who is picked.
- Backfill: an inactive trader is skipped and the next-ranked eligible trader takes the slot, so each arm still follows 10 traders whenever 10 are eligible.
- Buys are copied for 7 days, 5 minutes late with 1.5% slippage and sized in proportion to the trader's own allocation; sells are mirrored; markets settle at resolution; open positions are marked daily. Fees and liquidity limits are not modelled.
Stars = 1 + floor(score / 20), capped at 5. Scores refresh at least every 6 hours.
Hypothetical analytics on public data. Past results don't predict future results. Not financial advice.