# F-CLV-STAGE pull-history gate replay (research, 17 Oct) - task #2790
Data: entries of F-CLV-STAGE (a2 mi14 harness, CLV 1.3-2, n=350, pulled entry = 0x, base mean 1.57 lo80 1.29, 85 runners), lpburn.db pulls (>=30% pool ETH), deployer = tx.from of pool-creation tx (rug6/fetch+f2.py, 2866 pools, all 350 resolved), gate.py.
Gate evaluated at entry-time (age 300 s cutoff, only history before it).
| gate | blocked n | of them pulled | runners lost | kept mean (lo80) |
|---|---|---|---|---|
| none | 0 | 0 | 0 | 1.57 (1.29) |
| deployer has >=1 earlier pulled pool | 17 | 11 | 4 | 1.64 (1.34) |
| deployer has any earlier pool | 32 | 14 | 4 | 1.63 (1.34) |
| global pause: >=20 pulls in last 2 h | 150 | 76 | 31 | 1.29 (1.01) WORSE |
| global pause >=10 / >=40 | 273 / 28 | 138 / 13 | 55 / 6 | 1.79 (1.19) / 1.45 |
## Verdict
- 91% of entries (318/350) are on a deployer's first pool; deployer history catches 11 of 171 pulls (6%). Same as the wlink finding: one-shot deployers.
- Deployer pull-history gate is cheap and safe (loses 4 runners of 85, lifts mean 1.57 -> 1.64) but does NOT fix the 49% pull rate. Add as a minor rug_gate component only.
- A "pause after N pulls in 2 h" gate does not work here: pull waves are the normal background (median 18 in 2 h), blocking drops winners as much as rugs (>=20 kept mean falls to 1.29).
- Still untested (no data): LP NFT owner / funder of deployer. Funder walk remains blocked (free RPC). The remaining lever is pre-buy exit speed (pull median 44 s after first outside buy), not a history gate.
- Recommendation: F-CLV-STAGE keeps its worst-case 1.57 edge; stay on paper/shadow; rug_gate = deployer-prior-pull block only.