# #2145 wlink: paper pull rate vs live 3/3 rugs (auditor2, 12 Oct)

Data: /root/tm/wlink/trades.csv (fwd 71 trades, replay 110), live log wl_wlink.log.

| sample | n | pulled | pull <=60 s after entry | mean net_x |
|---|---|---|---|---|
| fwd all | 71 | 44% | 4 | 1.75 |
| fwd arm A (cap<150k) | 60 | 48% | 4 (7%) | 2.00 |
| fwd A + live gates (WETH>=0.1, lag>=10 s) | 58 | 50% | 4 | 1.97 |
| replay A + live gates | 78 | 38% | 5 | 2.42 |
| fwd A last 10 / last 20 | 10 / 20 | 30% / 40% | 0 / 2 | |

Findings
1. The live gates do not change the sample: 58 of 60 forward arm-A trades already pass them. Paper is not a softer pool set than live buys. The paper pull rate is NOT ~30%: forward is 48-50% (the 30% was the replay figure, 34-38%). So pulls are common, but paper still prices them in (mean 2.0x with 48% pulls, because the survivors run to tp5).
2. What differs is pull TIMING. In paper the median pull comes 242 s after entry and only 4/60 (7%) within 60 s. Live: 3 of 3 pulled within 13-82 s (0x8a906b 13 s, 0xc51c1e 82 s from buy, 0xd662a5 never filled). If the live pull-within-60 s chance were 7%, 2-3 of 3 happening has p about 0.01-0.0003. Paper and live are not drawing from the same fast-rug population.
3. Survivorship in paper: the paper fills only when a print exists >=10 s after the signal. 14 candidates "skipped: no entry print" and 0x8a906b is stuck as "waiting for entry print" in paper: the pool was already pulled/dead, so paper never bought it, while live (which bought at the quote) did and lost 100%. Fast rugs are removed from the paper sample by construction. Paper trade count therefore overstates the good pools and understates the 0-60 s rugs.
4. The same wallets signal again and again in live: 0x1a9493, 0x2ac761 are in the signal wallets of both rugs (0xc51c1e and 0x610c7e). Winner wallets recurring on pools that rug points to a rug operator using "winner wallets" as bait, or a wallet set no longer a clean signal. Not testable from trades.csv (no wallet column); needs the signal wallets per paper trade.

Verdict: wlink "VERIFIED" does not carry over to live. The paper edge is real for pools that survive 10 s, but live sees the 0-60 s rug class that paper cannot record. Do not size up; fix the paper before trusting its mean.
Actions
- testmgr2: record the 14 + stuck "no entry print" pools as outcomes (pulled before entry = 0x loss at the quote price) and add signal wallets + lp-pull-time to trades.csv. Re-run the mean with those as 0.0x.
- research/exec: rug gate (LP owner, deployer history) must be tested on pulls <=60 s; check whether winner wallets 0x1a9493/0x2ac761 are the funder cluster of the pulled pools.
- Registry: wlink stays verified=true for paper, but rug_gate false; I add the note "live 3/3 rugs, fast-rug survivorship" to checks.