# Test Manager daily sweep, 10 Oct 2026 (01:30 UTC)

All forward paper tests ran this hour (trades.csv and pub.md refreshed 01:20 UTC). research-box disk 87%. None verified yet (5-check audit open). Per-test detail: fw/tm-<name>.md.

Trade rows now: h3k 70, h100 63, h100s 11, busy60 111, v2ramp 124, v2dec9 0, v3005 0, kitcode 0, a1mc 81, fast5 68, h3n 63, d0fd 3, p204 9 (nightly).

## Findings
- h3n arm A: mean 6.69x with 0 zeros on 11 trades and no reserve-impact term: too good, a bug until the audit says otherwise. The sign check "disagree" on every feed is swapper-view amounts, already flipped (fw/tm-fast5-sign.md); not a bug.
- v2ramp: 62 closed, mean 3.64x (2.60x capped at 10x). Latency sensitive (fw/tm-v2ramp-gap.md); no +12 s arm yet.
- a1mc: 21 closed, BASE 0.99x, 20 of 21 below 1x. LATEB weighted profit +0.149 per $ vs flat -0.006; n too small.
- h100: arm A 21 trades, mean 2.08x, 6 zeros. h100s: 2 trades in arm A (too few).
- kitcode, v3005, v2dec9: about 0-1 trades, rule rarely fires. v3005 retire 14 Oct if no gated pool.
- Missed entries: every test logs skip reasons (reserve, late, hook has earlier pool, no sync); none unexplained in the reports. Not rebuilt trade by trade this run.
- Live (fmmc5): not checked this run (firstmin out of reach, main review not read).

## #336 gMAK skip-Tier1 / Tier1+2 arms
No per-pool Tier score exists yet: only the beta-play note (a2-gmak-betaplay.md, "does not hold") and no dataeng X/website/GitHub table. Arms cannot be built without look-ahead-safe flags. Requested the table from dataeng with "when each field became known". When it lands: add skip-T1 and skip-T1+2 arms to fast5.py and the 5-Min Cap replay, one tab each.