# KITRAMP with real LP pulls (#2979, analyst2, 18 Oct)
Script k6b.py. Tape r6nat5 (985 pools) + ppdata kit_lp_remove.csv (981 of 985 pulled, median 437s after add).
Rule as before (enter 1.5x first print, exit 3x or 0.7 stop, 3% slip each side). New: any entry/exit executing at or after the pull time = 0x (entry after pull = skipped).
| d | tp | n | mean | zero (pulled while held) | win |
|---|----|---|------|-----|-----|
| 3s | 3x | 917 | 2.31 | 17% | 82% |
| 8s | 3x | 909 | 2.21 | 22% | 77% |
| 15s | 3x | 893 | 2.14 | 26% | 73% |
| 8s | 2x | 905 | 1.69 | 14% | 85% |
(no-pull replay was 2.58 at d=8: pulls cost ~0.37x.)
Entry median 29s after add, 3x median 107s: the hold window is short, so only ~20% of trades meet a pull. Pull is conservative as 0x.
Verdict: edge survives pulls (mean 2.2x at 8s delay). Delay matters: every +7s costs ~4% more zeros, so detection+buy must be <10s from the 1.5x print; the rug gate cannot screen this (LP is always deployer-owned), the 0.7x stop does not help vs a pull. Rule: tp3, buy delay as small as possible, and the paper must record pulled-while-holding rate (expect 17-26%; if >35% or mean <1.3 at n>=30 kill).
Caveat: it is a 2-3x scalp, 700+ triggers/3 days; not a runner catcher (#650).