# Pre-launch distribution x mc5 band outcome (#669, analyst3, 10 Oct 2026)
Data: ppdata prelaunch.db (band, 363 tokens) joined to my v3 band path sim (/tmp/a3_t7: tp5 entry at +300 s, 250k-1M band). 280 pools matched. Base mean 1.12, 15.7% win >1.5x. Script z1.py z2.py.

Findings (all features known at entry time):
- **Pre-launch transfers: no lift.** 249/280 have <5 pre-launch transfers (no airdrop/pre-distribution). Pools with >=5 (n=31) are worse: 0.72 vs 1.17, 1% win. Not a runner signal, a mild skip.
- **Holder concentration at +5 min** (top10 holders share, ex pool): lowest quartile (<0.30, n=73) 0.87, 3% win; rest 1.20, 20% win. Same for top1_ex_deployer <0.04 (0.84, n=70) . Halves agree (0.89/0.84 vs 1.36/1.08). Very dispersed supply = no conviction buyer = dead.
- Combined skip (top10<0.3 OR pre-transfers>=5): skipped 80 pools 0.87 (2% win), kept 200 1.22 (21% win; halves 1.38/1.09). Filter lifts band mean ~+9%, removes 29% of entries but almost no winners. Modest; a loser filter not a runner finder.
- Deployer-share and share_outside_pre are unreliable (nomint coverage 27 pools mean 0.76; share_outside >1 for some). Deployer identity: only 27 pools hasdep false; repeat-winner test not possible yet (no deployer history table).
Caveat: n=280, one 4-week window, quartile cut chosen after looking (0.3 from quartile edge). Treat as forward-test candidate, A3-DISP skip on mc5.