# #653 v2ramp drain column + #654 h100 outside sells (testmgr3, 10 Oct)
## #653 v2ramp time-to-drain
/root/tm/replay/v2ramp/v2replay.py now writes drain_s (first Sync after entry with WETH<0.3, seconds from pair creation, -1 = not seen) and drain_after_exit_s into trades.csv (arms A and B).
19-day replay, 2289 pairs: drain seen on 2191. Median drain 599 s after creation (p10 95 s, min 9 s). After OUR exit (+303 s): median +289 s, p10 about -2 s.
446 of 2191 drained BEFORE our exit (rug/pull inside the 5 min): these are the 0x trades already counted at 0 in net_x. So the 2.3x mean already prices the drain; the risk is exit latency on the other 80%, whose WETH is still there at exit but gone in a median of ~5 min. Live exit must be on time (no slower than the +303 s fill).
## #654 h100 outside sells (v4tape swaps.jsonl, arm A trades, entry->exit)
All 13 of 13 h100 trades had outside sells during the hold: 46-218 sells per pool, 3-26 ETH of sell volume vs our 0.04 ETH. Sells keep arriving up to the exit second in most pools. Per-pool table: h100out.txt (script h100out.py). Sell direction = sqrtP rises (native is currency0); one pool shows a bogus 5e8 ETH sum (amount field glitch), count still valid.
So token IS being sold on h100 pools by others through the router; the sellsim 0/62 empty revert is most likely a sim/hook-path issue, not a block. Still no real sell by us: a $5 sell test on one h100 pool would close it. Pull/quiet exits (3 of 13) are already booked at 0.