# t3 replays for #780 (10 Oct 2026, testmgr3)
Scripts/outputs: /root/tm/replay/{v3005,v2dec9,usdg}/ (on research-box). Data limits: v3 tape (v3live.db) only starts 6 Oct, so v3005 and v3late1m cannot go beyond 4-5 days; v2 and USDG v4 replay on r6tape.db (21 Sep-9 Oct, ~19 days).

| test | replay | forward | verdict |
|---|---|---|---|
| v3005 | tp5 n=13 mean 3.47x, 3 <1x, 2 pulled, loss streak 1 (A1 sim 3.8x); tp3 2.35x | 0 gated pools | audit_ready |
| v2dec9 | 72 trades / 18.7 d: A fill (entry +12 s after entry Sync, exit first Sync >= +353) mean 14.4x median 11.3x, 6 <1x, 2 <0.2x; optimistic fill 30x | 5 closed, 4 wins 17-108x | audit_ready, WARNING |
| usdgearly | E2 arm A 1.01x, arm B (honest, pull=0) 0.62x n=1397; hooked only: B 1.37x n=406 (A 1.62x); B4 B 0.90x n=63, hooked 3.24x n=10 | 0 entries (10 cands all cap>=300k) | all-pool version fails its kill rule; hooked-only subset is the candidate |
| v3late1m | no longer history than the existing replay (6-10 Oct): n=17 tp2 1.04x | running | stays forward; kill if tp2 <1.0x at 25 trades |

## Notes
- v2dec9: same scripted-ramp family as parked v2ramp (token reserve 1e18 raw, seed 1.000). Forward wins sold into cap $60k-$587k from a $3k entry cap. Median 11x is a bug-until-proven; auditor3 must do sellability (eth_call) and outside-seller checks. Replay has no factory/sibling filter (not in tape meta).
- usdgearly: the A1 sim (1.49x) matches arm A; honest arm B is 0.62x because 737 of 1397 pools go to zero (pull/quiet). Hooked pools show zeros 96/406 under B. Sellability on hooked USDG pools still untested.
- v3005: gate is rare (13 of 280 first-0.05 pools), 4 days only; the 3.47x rests on a few 5x fills.