# analyst3: #412 first pass (gMAK tiers #333, Proxima/bundler #346), 10 Oct 2026
Script: /opt/crew/work/analyst3/j.py. Data: bundle_fund.db (dataeng; 160 bundled launches of 7,297 pools seen, 17 Sep-10 Oct) joined to e1-runners.csv (358 pools overlap).

## Bundler (multi-wallet block-0 buy, dataeng detector) vs outcome
- Bundled share: 38% of class=runner pools (32/84) vs 17.5% of controls (48/274). Lift ~2.2x for being a runner.
- Peak cap reach, bundled (n=80) vs not (n=278): 1M 71% vs 42%; 3M 40% vs 19%; 5M 31% vs 16%; 10M 20% vs 9%; 20M 11% vs 6%.
- By wallets: 6-8 wallets (n=21) 5M 33%; 9+ (n=57) 5M 32%. No dose effect.
- First bundle-wallet sell: median 122 s after launch (155 pools); 164 s for coins that reach 5M+ (n=25), 123 s for coins under 1M (n=22). So a "sell when bundle sells" exit gives ~2 min, and runners are not distinguished by it.
- Caveat (important): e1-runners oversamples runners, and only half of the 160 bundled pools are in it. Ratios are lift among the sampled set, not true population rates. This CONTRADICTS A2's proxy (>=4 front buys: AVOID, reach-3x lower on WETH v4, 1h horizon): different horizon (peak ever vs 1 h) and different filter. Bundled coins reach big caps more often but over hours; they do not pay inside 1 h.
- Verdict: bundled launches are over-represented among runners (candidate FEATURE for A2 late hunters, not a skip rule). Do not use as an avoid arm for the 5M+ hunt.
- Next (needs population tape): full-population runner rate for bundled vs not on the 7,297 seen pools, causal at entry. Ask: ppdata peak cap per seen pool.

## gMAK tiers (#333)
- Beta-play: A2 already tested, does not hold (fw/a2-gmak-betaplay.md).
- Tier 1/2 social, DEX paid, template site, GitHub: no data (dataeng social/metadata table not started, per de-status). Cannot judge. Kept open; waiting on dataeng.

## Not done this run
- 30-day early-wallet study of 5M+ runners (needs per-token early buyer list; txfrom.db covers tx.from only, rebuild in progress). Plan: for each 5M+ runner, buyers in first 5 min, find wallets recurring across >=3 runners, test lift on next runner.
- ledger CLI not on main/research PATH for me; ideas not logged.