A1 late buyers / no repeat

Waiting for: TOPB11 arm uses swaps.signer (router); A3 definition is swap recipient, which v3live.db swaps lacks. Needs recipient column (asked dataeng2); until then TOPB11 arm is invalid, ignore it. Other arms: forward n 33 of 60; owner testmgr4

Still open:

• Audited 10 Oct: too small, mean = 2 winners (BASE 0.43x without them); re-audit at n>=60
• Not audited: fills vs real depth, kit overlap
• Deployer table from 6 Oct only
• Replay leg audited 10 Oct: no look-ahead, seed robust (1.25x minus top5), a1mc BASE 0.95x minus top10; still forward n28, sellability open: fw/audit-seed-a1mc-replay.md
• 11 Oct FIX: TOPB11 identity must be tx.from not router; fwd n13 medians<=0; fw/audit-wide-a1mc-11oct.md

Test Manager forward test on research-box. Kill rule: LATEB weighted profit/$ not above flat after 60 trades or late>40 mean <1.3; repeat mean >=1.3 after 40 coins | ENS6: bucket 4-6 minus 0-1 mean x <0.5 at n>=150, or halves opposite sign.. Live only after 2 profitable days without bugs; only Ben switches.

Each step is a trade with exit rule LATEB, numbered in time order.

# A1-LATEBUYERS / A1-NOREPEAT paper arms (Test Manager) - 12 Oct 08:20 UTC

Rule (fw/a1-iter14.md, #251 #252): mc5-style entry on v3live (+300 s, mcap 250k-1M, stop 0.5x filled at the real swap price, 4 h, pull = 0, 1%+2% each side). Pools created from 9 Oct 00:00 UTC (out-of-sample).
- Closed: 51; waiting for +300 s: 3; skipped {'cap out of band': 222, 'no entry swap': 586}; no deployer row yet 0
- BASE flat: n 51, mean 0.89x, <1x 49
- LATEB (2x late>40, 0.5x late<=5), return on base $: n 51, mean 0.70x, <1x 49; profit per $ weighted -0.177 vs flat -0.114. Kill: weighted not above flat after 60 trades, or late>40 mean < 1.3.
- late>40 group: n 38, mean 0.71x, <1x 37; late<=5 group: n 7, mean 0.54x, <1x 7
- NOREPEAT (deployer new): n 38, mean 1.10x, <1x 36; repeat deployers skipped: n 13, mean 0.25x, <1x 13. Kill: repeat mean >= 1.3 after 40 coins.
- Not tested: real fills vs v3live pool depth, kit/bundle overlap with live mc5, deployer table covers only from 6 Oct (repeat window shorter than 7 d for early pools). Only Ben switches live or sizes (D-030).
- STOP01H8 (#2003, A1 rug model: stop 0.1x, pull exit 0.05x, hold 8 h) vs BASE on the same pools: arm n 50, mean 1.47x, <1x 46; base n 50, mean 0.90x, <1x 48; ex-top3 arm n 47, mean 0.17x, <1x 46, base n 47, mean 0.40x, <1x 47. Pairs open until 8 h. Decide at n>=40 closed (A1: not robust in time; lottery on whales). Kill: arm mean not above base by 0.15 at n 40 or ex-top3 below base.
- DUPNAME tag (A1-DUPNAME, #267, tag only): dup n 32, mean 0.76x, <1x 31; non-dup n 19, mean 1.11x, <1x 18; name unknown 0. Kill: dup mean not above non-dup by 0.15 after 60 coins per tag. Name window uses v3live pools only (v4/v2 pools not seen), so dups may be undercounted.
- A1-ENS6 (#276, 6-flag score, entry/stop/exit unchanged; arm ENS6 = 2x at score>=4, 0.5x at <=2, score column on all arms): score 0-1 n 4, mean 0.24x, <1x 4; 2-3 n 7, mean 0.60x, <1x 7; 4-6 n 40, mean 1.00x, <1x 38; hit5 4-6 2; profit per $ tiered -0.047 vs flat -0.114. Report at n>=150. Kill: bucket 4-6 minus 0-1 mean x < 0.5, or halves differ in sign. Flag 6 from launch-block buyers in v3live (v3 pools only, 24 h window starts 8 Oct 00:00 UTC); thresholds in-sample (A1).
- top-share median 0.26, nbuy median 177
- NPRE10 (#91, keep pools with >10 swaps before +300 s): keep n 44, mean 0.94x, <1x 42; skipped n 7, mean 0.54x, <1x 7. Replay: keep 1.38 vs skipped 0.90. Kill: keep not above skipped by 0.2 after 60 coins.
- TOPB11 (#98, top buyer share of 0-300 s buy ETH <0.11, >=7 buyers): keep n 1, mean 0.97x, <1x 1; rest n 50, mean 0.88x, <1x 48. Replay tp5: 1.58 n146 vs 0.99. Kill: keep not above rest by 0.2 after 60 coins. (Arms use a1mc exit rule, not tp5; compare to BASE.)