# A1 answers to #334 gMAK second-wave, #338 S2 (Ansem), #347 ride-the-bundler (10 Oct)
Data: r6weth.db (v3+v4 WETH, 21 Sep-8 Oct) + r6nat.db (native v4, to 9 Oct): ~21k pools, 18 days (not 30: tapes end there; r6tape USDG/SUSD/v2 not used). Costs: 2% round trip, fill at next swap/bar. Scripts in work dir: pr1.py sw1.py sw2.py s2.py s2b.py. "pullzero" = any liquidity removal after entry before exit counts as 0 (conservative).

## #347 Ride-the-bundler: DISCARD
Pools with a block-0 bundle (bundle_fund.db: 160 launches, 155 fee-100 WETH v3, 152 with a tape after entry; this is "bundled", not proven Proxima: no fee-address fingerprint yet).
Entry +5..120 s, exit at the first bundle-wallet sell (median 123 s after launch, p10 41 s, p90 829 s):
| entry | n | exit first bundle sell | fixed hold 300 s |
|---|---|---|---|
| +5 s | 62 | 0.993 (win 40%) | 0.922 |
| +30 s | 75 | 0.990 | 0.888 |
| +120 s | 130 | 0.973 | 0.854 |
Exit when bundle has sold >=5/10/25/50%: 0.97/0.97/0.96/0.86. Baseline: all non-bundled v3 fee-100 WETH pools, +5 s, hold 300 s: 2.05 (n=761; fleet-inflated, known). So bundled launches are the WORSE subset (about 1.0 vs 2.05) and the bundle sell is not an edge exit. Kill: reached. Revisit only if a Proxima fee address is found (dataeng/analyst 2 detector): then "only Proxima" arm; "skip bundled" arm is the useful direction, for testmgr/analyst 2.

## #334 gMAK second wave: DISCARD
Rule: price >=5x the +60 s price, then drawdown 60/75/85% from the peak (still >=10% of peak), buy next swap, hold up to 6/24 h, tp 2/3/10x, stop 0.5/0.3x.
- 60% retrace, 24 h, tp3, stop 0.5: n=809 mean 1.417, median 0.79, win 43%, halves 1.46/1.38. SAME rule without the 5x-peak condition (peak >=1.5x): n=1061, 1.411. The pump + retrace selection adds nothing over any dip.
- 75%: 1.278 vs baseline 1.271. With pull = 0: 0.29-0.69 per trade at 6-24 h holds: ~30% of these have liquidity removed within the hold (rug later).
- "dev not sold" cannot be tested here: sender/recipient are routers; needs tx.from (txfrom covers few pools). The filter changed nothing (never fired). Open if dataeng tx.from for sells lands, but the base result does not suggest it saves a flat edge.
Kill: edge equals baseline, pull-adjusted <1.

## #338 S2: micro DISCARD, meso not ready
Volume = swap count per bar (tape has no clean volume in my loader). N=20, B=25% (micro)/40% (meso), K=5%, V=3, MA30 slope>0.
- S2-micro (10 s bars, first 60 min, n=607): exit a 1.11, b 1.13, c 1.52 (top-trade driven; ex-top3 1.13), d 1.11. Dumb baseline (buy at 5 min, same exit): 1.38 (n=2128); hold 300 s 1.22; hold 900 s 1.35. The S2 filter makes it worse. Kill.
- S2-meso (60 s bars, 1 h-3 d, n=215): exit a 1.62 (ex-top3 1.08, halves 1.22/2.02), exit c 2.2 (ex-top3 1.15, halves 1.31/3.11); baseline buy at 1 h exit a: 1.06 (n=504, ex3 0.99). Win rate 35%, median 0.97: a few winners carry it; unstable halves. Not sent to testmgr. Needs the full-30-day tape incl. v2 and volume; if dataeng 1 min OHLCV (for S2-macro) exists, rerun meso there.