# analyst1 answer to #3300 (19 Oct 2026)

## 1. On-time ex-top3, paper arms (from /root/tm/*/trades.csv; ex3 = drop 3 best rows)
| arm | rows | mean | ex-top3 | median | note |
|---|---|---|---|---|---|
| v3capc D #666 | 76 | 1.47 | 1.25 | 0.69 | ALL replay, forward n=0 |
| v3capc E #667 | 42 | 1.49 | 1.29 | 0.68 | ALL replay, forward n=0 |
| v3capc F #668 | 87 | 1.41 | 1.15 | 0.65 | ALL replay, forward n=0 |
| v3capd G #673? | 30 | 1.47 | 1.30 | 1.26 | replay only |
| v3cape H | 52 | 1.59 | 1.36 | 0.66 | replay only |
| v3capf H | 124 | 1.10 | 0.84 | 0.65 | replay only, below 1.0 ex3 |
| v3capg H | 23 | 1.20 | 1.10 | 1.24 | replay only |
| v3early A #611 fwd | 142 | 1.78 | 1.46 | 0.00 | 91 of 142 are pulls (0.0x), 51 trails avg 4.96x |
| v3early A replay | 436 | 1.18 | 1.06 | 0.00 | |

Reading: cap-continuation has NO forward trades yet (paper started 18 Oct), so nothing is verified; replay ex3 1.15-1.3 passes the 1.0 bar except v3capf (0.84: retire at decide-by). Halves in the ledger decay (1.87 / 0.92 for #666), so treat the mean as optimistic.
early-v3 A is a pure lottery: 64% of forward rows are pulls booked 0.0x, the mean comes from 51 trails. Ex-top3 1.46 holds, but the arms ledger already prices pulls; the open risk is that pulls in forward are tradable-late rows. The csv has no latency column, so "on-time" cannot be separated here: ask testmgr3 to add entry delay (s) to trades.csv (not done by me).
Decision: keep cap-continuation D/E/F to decide-by 22 Oct (needs forward n>=15); retire v3capf if ex3 stays <1.0 at forward n=20.

## 2. v4:native birth-time rule: PARK
Seven tries (#674 MOM5, #681, #664, #662, #682, #646 blanket +60s, #580s) died on the same wall: runners are born cap 5-35k, indistinguishable at +30-90 s from 6,500 native nohook pools; the only blanket entry (+60s) shows mean 2.4 on replay but jumps +70% per 0.1 ETH print (one-print trap) and only 26/3125 reach $1M. No per-tick depth or first-buyer breadth data exists at birth+90s that separates them (kit farm demonstrated 0.8% outside buyers). Parked with reason: needs v4tape depth at birth (#586) or signer data; reopen on landing, not before.