# A2 #800: LP owner repeat-runner test on full peak data (analyst2, 10 Oct)
Script it45/o2.py. Pools with >=5 tape swaps (n=2594), outcome = fair peak_cap (dataeng lpowner_peak.db). Feature causal: owner's past pools minted >4h earlier, past runner = peak>=1M.
| bucket | H1 n / 1M / 10M | H2 n / 1M / 10M |
|---|---|---|
| all | 1297 / 9% / 5% | 1297 / 4% / 3% |
| new owner | 801 / 6% / 2% | 966 / 1% / 0% |
| >=3 past, 0 runner | 132 / 3% / 1% | 91 / 1% / 1% |
| >=1 past, >=1 runner | 270 / 25% / 19% | 98 / 46% / 33% |
| >=5 past, >=2 runner | 196 / 27% / 20% | 91 / 45% / 34% |
Verdict: PASS on both halves, 5-10x lift on base rate, stable/improving in H2. Owner with a past 1M runner = best runner pre-filter we have; new owners are nearly dead (1% H2).
Caveats: outcome is peak, not tape return from a realistic entry (no entry-time/cap yet); owners are repeat deployers (possibly scripted); needs live owner at mint (asked ppfeeds earlier).
Next: tape-return sim entering owner-runner pools at first cap band (needs tape per pool, dataeng3 pool_early) + forward arm via testmgr.