§C · strategy validation
MirrorTopN backtest
For every closed position from a wallet who was already a top-50 leader at the time they entered, we simulate a mirror trade at the same prices with 50 bps of slippage and a $ 1,000 notional cap. The Wilson lower bound on the mirror's hit rate is the strategy's scientific gate: only when it clears 50% does the signals dashboard mark anything BUY.
Strategy gateOPENWilson lower > 50%
Hit rate51.0%50,414 W · 48,464 L · 4 flat
Wilson interval (95%)50.7–51.3%α=0.05 on n=98,882
Mirror P&L-$448.1Kavg per-trade -4.46%
Sharpe proxy-0.03σ 162.8%
Last run2026-06-12 10:46 UTCtop 50 · 30d · 50 bps
Top winners
Top losers
Methodology & caveats
- Eligibility: a closed position counts only if the wallet appeared in the top-50 by 30d PnL at some captured leaderboard snapshot ≤ the position's
opened_at. Otherwise we couldn't have known to mirror them. - Mirror entry = leader's avg_cost × (1 + 50 bps); exit = avg_exit_price × (1 − 50 bps). Prices clamped to [0,1].
- Mirror size = min($1,000 notional ÷ entry, leader's actual qty). We can't mirror more than the leader actually traded.
- Selection bias: the leader pool is whoever's top-N right now. A walk-forward backtest that re-picks the pool at each historical date is the honest next step (needs more leaderboard history first).
- Survivorship bias: "top-N by 30d PnL" selects for recent good outcomes. Some of that is skill; some is variance. The Wilson lower bound is conservative on purpose.
- No drawdown gate: a strategy can be majority-winning and still unfollowable if drawdowns are deep. The cumulative curve above is the honest picture of what holding through this strategy would have felt like.