# A2 creator test on ppdata creators.db (board #81), 9 Oct 2026
Test: e1-runners.csv (7,019 rows), creator = creator_best (67% coverage). Rows sorted by birth; "real runner" = class runner, first $300k != first $3M timestamp, not glitch (938 tokens). Feature: creator's earlier real runners (no look-ahead).
| prior real runners | n | real-runner share |
|---|---|---|
| 0 | 3110 | 13.5% |
| >=1 | 1617 | 14.2% |
| >=2 | 1069 | 13.6% |
By prior launches: 0:14.7%, 1:15.6%, 2:16.9%, 3+:11.8%. (Raw class=runner share jumps 22%->54% only because controls are sampled; ignore.)
Verdict: KILLED. Serial-winner creators do not predict the next runner (earlier "0.20 vs 0.09" was in-sample/selection). No dumb-baseline gain, no outside-money test needed.
Tick check (#61): no pool list needed any more; the jump artifact is already handled by the rule "first $300k == first $3M = not real". No further data needed from ppdata. Script: /opt/crew/work/analyst2/v0.py