paste a CA. we replay every new high, wait four hours, and check whether the floor held.
Data. GeckoTerminal hourly OHLCV (up to 1000 candles ≈ 41 days) on the token's deepest pool, plus DexScreener live pair stats (price, buy/sell transaction counts, volume, mcap). Both public and free — every verdict's PROOF section links to the raw sources, replay it yourself.
Rules, exact constants. Breakout = 1h close ≥2% above the running window high. 4h later: still ≥85% of breakout = layer HELD; ≤60% = ROUND-TRIP. Score = levels ×.35 + buy pressure ×.25 + green-day streak ×.15 + volume trend ×.10 + retention ×.15. EXIT LIQ = a round-trip within 12h and nothing held since. Breakouts and holds are judged on closes, not wicks — painting a candle is cheap, holding a price for 4 hours against real sellers is not.
Known limits, before you find them. DEX data only — that's why majors and stables are refused (NOT A TRENCH ASSET), never mis-scored. "ATH" means the 41-day window high. Buy pressure counts transactions, not size — wash-tradeable, which is why it only carries .25 weight and why fake volume with zero held levels flags as PVP CHURN instead of scoring well. Single deepest pool, not cross-pool aggregate. No holder data yet (v2). Receipts measure forward returns from the confirmation close (breakout +4h) — the first moment the signal actually existed. No lookahead.
Falsification. The thresholds are priors, not curve-fits, and they're being calibrated against forward-return buckets in public. Every verdict is committed to a public git ledger with tamper-evident timestamps — wrong ones included: github.com/scarlettzhangxh/exitliq. If the favorable buckets don't beat the risk buckets, this product deserves to die. That's the deal.