# v4:native symbol-repeat replay (#4063, research, 26 Oct)
Source: radar.jsonl, ev=cross th=300k, venue v4:native, not backfill, depth_usd>0, cap_first<450k. n=2061 pools. Hit = also crosses 1M later, excluding phantoms (1M within 60 s of 300k; 976 such tokens dropped).
| group | n | reach 1M | rate |
|---|---|---|---|
| all | 2061 | 262 | 12.7% |
| first use of symbol | 860 | 89 | 10.3% |
| symbol seen before | 1201 | 173 | 14.4% |
| seen before AND an earlier same-symbol token reached 1M | 528 | 108 | 20.5% |
| seen before, no earlier runner | 673 | 65 | 9.7% |
Halves by time (rep-after-runner): early 56/223 (25%), late 52/305 (17%).
Read: lift is ~2x over the base rate for the "ticker already ran" copy-cat group, stable in both halves. It is a hit-rate only: no entry fill, exit, rug or outside-seller model, so it is NOT a P&L. Older v3 version (#256, #100) was killed for artifact; this one is on v4:native which mc5 does not read.
Next (if analysts want it): price replay on the 528 with tp/stop at 300-450k entry, using radar cap5 and swap path. Rug risk unchecked (v4:native pulls).