# testmgr2 daily sweep, 10 Oct 2026 (08:50 UTC)
Forward trades.csv per test (24h entries / main-arm n, mean x, zeros). All six tests alive, written within the last 3.5 h except d0fd.
| test | last entry | 24h entries | main arm A |
|---|---|---|---|
| h3k | 1.9 h | 148 | n27, 1.63x, 9 zeros |
| h100 | 3.5 h | 68 | n13, 2.24x, 3 zeros |
| h100s | 3.5 h | 14 | n2, 3.56x (Z ctrl n8 2.11x) |
| busy60 | 2.3 h | 154 | n7, 1.66x; control Z n63 0.83x, B n22 0.53x |
| h3n | 0.6 h | 87 | n15, 8.68x, 0 zeros (too good: audit hand-rebuild needed) |
| d0fd | 15.3 h | 3 | n1, 4.68x (bursty launcher, 1-4/day, expected 3/day: no alarm) |
- Missed entries: h3n and h100s cands.csv log every skip with a reason (reserve<0.3/0.4/0.5, late>2.5s, hook has earlier pool); no technical skips seen.
- Pulls: forward and replay scripts count "quiet" (pool stops swapping) as 0, so silent pulls are already counted as total loss (conservative). A ModifyLiquidity-aware tape is built: /root/tm/replay/mk_mods.py -> tape/mods.jsonl (v4 kind=10 events, 21 Sep-9 Oct).
- Not done this run: 30-day replays for h100s/h3n/busy60/d0fd (h100s needs slot1 eth_getStorageAt per hook, h3n/d0fd read live v4tape liq.jsonl that only starts 9 Oct). Next run.
- busy60: rule arm A has only n7; control Z 0.83x < 1 means the busy gate carries all the edge; on-time n>=30 still open.
- Sellability for all six waits on Ben's $5 sell test.