How we count signals.

Most platforms only show their winners. We publish the full scoreboard — hit rate, median return, worst performer — because a promise without receipts is marketing.

• Signal types: only new_pair and volume_anomaly signals — they capture a token price at fire time in metadata_json.price_usd. Whale transfers and news events don't have a price at capture and aren't scored.
• Dedupe: first catch per token wins. If a token triggers three volume anomalies in the same week, only the earliest entry price counts. No cherry-picking the best entry.
• Liquidity floor:pairs with less than $5K of current liquidity are excluded. Dead pairs distort both winners and losers — a token that fell 99% to $100 of liquidity isn't a meaningful miss, just an abandoned pair.
• Age floor:signals younger than one hour are excluded. Too soon to know; a pair minted minutes ago isn't a scored outcome.
• Return source: current price from DexScreener divided by the price_usd we stored when the signal fired. No backfills, no optimistic rounding.
• Hit definition:a signal “hit” if its current return is ≥ 2×. A “strong hit” is ≥ 5×. Median return is the 50th-percentile multiple across all scored signals — a much harder number to fake than “we caught a 100×.”

What we don't do

• · We don't delete losing signals from the database.
• · We don't pick the highest-return moment post-hoc — current price only.
• · We don't exclude signals just because the token bled out. Bled-out tokens with any liquidity still count in the median.
• · We don't show P&L as if you could have bought at the exact signal price with infinite liquidity. It's a reference point, not a trade.