traders-v1 contract

Boards

• pnl:7d / pnl:30d / pnl:all — top 50 by upstream Data API leaderboard PnL for that window.
• sharpe:30d / sharpe:all— top 50 by daily unannualized Sharpe among that window's top-250 PnL candidates. Framed as consistency, not scale-invariant skill.
• There is no 7d Sharpe board: a 7-day daily series yields at most 7 deltas, at the min_observations: 7 floor, so rankings would be noise.

Candidate universe

Each release unions the three top-250 PnL lists (≤750 wallets, typically ~500–650 after dedup). Every candidate gets a profile page. Sharpe boards therefore have an explicit selection-bias caveat: they rank consistency inside the PnL elite, not the full Polymarket population.

Formulas

• sharpe_daily(window) = mean(daily Δ) / sample_std(daily Δ, ddof=1) over user-pnl deltas inside the window. Requires ≥7 non-null deltas and nonzero variance; otherwise null.
• roi_realized = sum(realizedPnl) / sum(avgPrice × totalBought) from closed positions in-window.
• win_rate= count(pnl>0) / count(pnl≠0); flat closes excluded.
• max_drawdown, underwater_days, and lifetime series PnL are always computed on the full equity series.

Tie-breaks for board order: metric value desc → window PnL desc → address asc. Releases are intended to be byte-reproducible from the same wallet artifacts.

Completeness

Every metric carries source, window, methodology version, and completeness (complete / partial / unavailable). Partial or unavailable values render as N/A in the UI, never as zero.

Refresh cadence

A daily ingestion Worker job discovers candidates, fetches positions / closed positions / user-pnl under per-host rate limits, builds boards, runs a publish gate (≥95% profiles, exactly 50 rows per board, schema validation), then atomically flips traders/releases/active.json and warms lean KV board slices. Failed gates leave the previous release live.

Deferred (v2)

Trades-tape metrics (trade count, avg ticket, notional concentration), archetypes that depend on them, 7d Sharpe via hourly fidelity, ClickHouse release history, and per-category boards.