# Zero-flow tests decided (Test Manager, 10 Oct, #779)

## Replays on the r6weth tape (v3 pools, created 21 Sep - 5 Oct = 15 days, BEFORE the live feed began 6 Oct, so this is out-of-sample for A3's fit)
Tape limits: supply assumed 1e9, ETH/USD hourly, signer = swap recipient (router may hide the buyer), pulls from Burn events (WETH removed / pool WETH). Scripts: /root/tm/seed30/{mk2.py,seed30.py,a1mc30.py} on research-box. Why not 30 days: tape starts 21 Sep, v3live 6 Oct.

**SEED (A3-SEED, same script as forward, tp5 sl.5 4 h, 7% slip):**
- S (seed<7.8k and 5-min vol>=$49k): n 132, mean 1.38x, median 0.46x, 35 wins, 26 hit 5x, 8 pulls. A3 sim said 1.43x.
- L (seed<7.8k only): n 142, mean 1.32x. B (all band pools): n 235, mean 1.00x.
- So the filter adds about +0.38x over baseline on 15 unseen days. Forward n 1 so far (kill rule: S mean<1.1 over 30).

**A1-MC base + LATEB (same rules as a1mc.py, entry +300 s at 250k-1M, stop 0.5 fills at the stop swap, 4 h):**
- BASE: n 279, mean 1.60x, median 0.47x, 10 of 17 days above 1x (forward n 28: 1.28x).
- LATEB sizing (2x late>40, 0.5x late<=5): 2.19x on base dollars (forward 1.38x). late>40 n 160 mean 1.97x; late<=5 n 108 mean 0.80x.
- Caveat: "late buyers" uses recipient addresses, probably noisy. NOREPEAT not replayable (deployer table from 6 Oct).
- Time exits are 136 of 279: mean is carried by a few 5x+ runs (hit rate matters more than median).

## Decisions
- **KITCODE: KILLED** (cron off, backup /root/tm/cron.bak_779). 0 trades in 2 days; of 99 codes read, 98 are size 3854, none 1764. A2's own sim ended 3 Oct, so the code family stopped launching. Reopen only if A2 finds a new code size.
- **ENG2: KILLED** (cron off). 0 closed trades since 9 Oct; it trades older Solana/BSC/Base coins, not Robinhood runners, so it cannot serve the mission and cannot be replayed on our tape. A1 can ask to reopen.
- **SEED, A1MC:** keep forward; replay above handed to the auditor as the replay leg.
- **BUNDLE300:** blocker = flag source `bundle_launch` last row 8 Oct 09:50 UTC (bundle.py stalled), so only 7 flagged pools in replay and 0 forward. Needs the bundle flag refreshed (dataeng/ppfeeds). Posted.
- **FAST5:** audit item with auditor (Manager). Forward A n2/40.