# fmmc5 live rule replayed over 13 Sep to 8 Oct 2026 (A1, for Ben)

Replay = the exact live rule (v3 WETH, cap $250k-$1M at first swap +300 s, entry +306 s, stop 0.5x, 4 h time stop, pull exit, 1% fee each way) on the golden tape, data to 8 Oct 20:07 BKK. 201 closed trades in 26 days (13 Sep to 8 Oct), 1 still open. P&L at $10 a trade = $10 x (x - 1), x already net of fees and gas (this reproduces the 25 live trades: mean 2.31x, +$328). Scripts q1.py, q2.py in /opt/crew/work/analyst1 on research-box. Data after 8 Oct 20:07 is not in the tape yet; live trades after that are in fw/mc5live.txt.

Caveats, plainly: (a) replay stop fills are 0.49x, live stop fills 0.22-0.52x (live 0.30 / 0.22 / 0.01 on gaps and rugs), so the replay is about 0.05-0.1x per losing trade too kind; (b) the "max 3 open" limit is not applied (4 trades were open at the same time at most, so negligible); (c) 3 of 11 live-match coins matched so far, no mismatch; (d) 5 of the 22 trading days have 1-2 trades, so day-level numbers are noisy.

## 1. Day by day (BKK date of entry, $10 a trade)
| day | trades | >=2x | >=10x | losers (<1x) | worst streak that day | P&L $ | best x |
|---|---|---|---|---|---|---|---|
| 2026-09-13 | 1 | 0 | 0 | 1 | 1 | -10.0 | 0.0 |
| 2026-09-14 | 2 | 0 | 0 | 2 | 2 | -14.1 | 0.6 |
| 2026-09-15 | 2 | 1 | 0 | 1 | 1 | 7.2 | 2.1 |
| 2026-09-18 | 1 | 0 | 0 | 1 | 1 | -10.0 | 0.0 |
| 2026-09-21 | 1 | 0 | 0 | 1 | 1 | -6.2 | 0.4 |
| 2026-09-22 | 5 | 3 | 2 | 2 | 2 | 298.7 | 17.6 |
| 2026-09-23 | 6 | 2 | 2 | 4 | 3 | 221.2 | 15.5 |
| 2026-09-24 | 11 | 2 | 0 | 8 | 4 | 49.9 | 7.5 |
| 2026-09-25 | 22 | 4 | 3 | 18 | 11 | 342.1 | 16.9 |
| 2026-09-26 | 17 | 3 | 0 | 12 | 3 | -3.5 | 4.0 |
| 2026-09-27 | 6 | 1 | 1 | 5 | 4 | 102.5 | 13.9 |
| 2026-09-28 | 11 | 0 | 0 | 10 | 10 | -54.2 | 1.9 |
| 2026-09-29 | 9 | 1 | 0 | 8 | 6 | -23.3 | 4.3 |
| 2026-09-30 | 7 | 2 | 1 | 5 | 3 | 144.6 | 16.6 |
| 2026-10-01 | 8 | 1 | 0 | 7 | 6 | -15.9 | 3.5 |
| 2026-10-02 | 15 | 2 | 1 | 12 | 8 | 282.2 | 27.9 |
| 2026-10-03 | 10 | 3 | 0 | 7 | 5 | 52.0 | 7.5 |
| 2026-10-04 | 26 | 3 | 1 | 23 | 11 | 32.6 | 14.2 |
| 2026-10-05 | 14 | 2 | 0 | 12 | 8 | 17.5 | 8.3 |
| 2026-10-06 | 7 | 3 | 3 | 4 | 3 | 818.4 | 42.0 |
| 2026-10-07 | 9 | 1 | 1 | 8 | 6 | 118.7 | 17.6 |
| 2026-10-08 | 11 | 1 | 1 | 10 | 6 | 119.7 | 19.5 |

## 2. Totals
- 201 trades, 40 ended >= 1.0x (20%), 35 reached >= 2x (17%), 16 reached >= 10x (8%). Mean 2.23x, median 0.48x (the stop).
- Total +$2,470 at $10 (about +$95 per calendar day, +$12 per trade). Remove the top 2 trades (42x and 28x): still +$1,791. So it is more than two lucky coins.
- Best day 6 Oct +$818 (three 10x+ runners, 42x). Worst day 28 Sep -$54 (11 trades, 0 winners). Worst calendar day is small because the loss per trade is capped near -$5.
- Longest losing streak 14 (twice: 25-26 Sep and 3-4 Oct), then 12 (28-29 Sep), 11 (4 Oct), then four of 8. So streaks of 10+ happened 4 times in 26 days.
- Biggest drawdown (peak to trough of the running total): $126, that is 12.6 tickets. Worst 30-trade stretch: -$98; only 5% of all 30-trade windows were negative.

## 3. How unusual was 8 Oct?
Live had 15 losses in a row (counted from the first live trade on 8 Oct 13:12). The replay's longest streak is 14, with three more at 11-12, so 15 is at the top of what we saw but not out of range. Bootstrap (resample the 201 results 5,000 times): P(at least one streak >= 15 in 201 trades) = 78%. Analyst 2 got 79% with a simulated 20% hit rate. Verdict: 8 Oct was a normal bad stretch for a rule that wins 1 time in 5, not a freak day. Expect a 10+ streak about every 6 days.
Replay for 8 Oct itself shows +$120 only because of the 19.5x coin that live missed (HTTP 429 at entry, fixed 8 Oct 15:23). With that coin live would not have had 15 in a row.

## 4. Market mood
Detailed in Analyst 2's cross-check (fw/a2-mc5-mood.md), I agree after checking the numbers: BTC down > 1% over 24 h: 35 trades, 11% >= 2x, mean 1.34x; flat 20%, 2.38x; up 18%, 2.53x. Gap is 1.3 standard errors: not significant, ETH not monotone. Day level: worst days 28 Sep (BTC -1.7%) and 8 Oct (-2.7%), but best day 6 Oct had BTC -0.3% and 22 Sep +3.8%. No on/off radar worth building from BTC/ETH. The only candidate (pause when BTC 24 h < -2%) would only have affected 24 Sep and 8 Oct (13 trades), saving little. Not recommended. A launch-count "chain mood" series is still untested; I have asked nothing yet because daily trade count (1-26) is driven by the cap filter, not mood.

## 5. Plain verdict
- Edge looks real: positive on 14 of 22 days with trades, P(month < 0) 0.02% in the bootstrap, still positive without the top 2 (P(<0) 0.2%). But the profit is 17% winners and a few 20-40x coins, so it comes in lumps.
- $10 is the right size for now. Live has 25 trades, replay 201: the live proof is still thin and live stops fill worse than the replay.
- To go back to $50: (1) 100+ live trades with mean >= 1.5x after fees, and mean still > 1.0x with the top 2 removed; (2) live stop fills within 0.05x of replay (execution gap closed, 429 and rug fills reviewed); (3) bankroll for the drawdown: typical month worst drawdown 144 (p50) to 224 (p90) at $10, so $50 means $720-1,120 down at times, 315 x 5 = $1,575 at p99. Only Ben switches size (D-030).