# fmmc5 live vs harness, 6 Oct 12:00 - 8 Oct 20:07 Bangkok (Analyst 1, 8 Oct ~21:20)
Harness: fw_mc5.json with window opened to data end (research-box /root/fwa1/rug/fw_mc5_now.json, now_trades.csv). Live: firstmin /opt/fm/mc5.db (read-only).
Small sample (25 harness trades in this period): description, not a verdict.

## Same coins: live sells as well as the backtest, or better
- 19 of 21 live buys are harness trades too; the other 2 (20:12 rug, last buy) are after the tape end.
- Same-coin x, 18 closed pairs: sum live 55.2 vs harness 48.0. Winners 29.4x / 19.5x live vs 24.6x / 17.6x harness. Rugs are near 0 in both.
- Per-trade differences go both ways (live 0.22 vs harness 0.46 on 0x4c4cdbc8; live 0.52 vs harness 0.01 on 0x25697994). Fills and exits are not the gap.

## Missed coins: this is the gap
Harness took 6 trades live did not. Live reason from mc5.db:
| pool | harness | live |
|---|---|---|
| 0x1dada094 (8 Oct 08:05) | **19.5x** | entry_err HTTP 429 Too Many Requests |
| 0x206211d4 (6 Oct 19:03) | 0.49x | buy_fail: GMGN "Swap - confirmation required" |
| 0x9312e871 (7 Oct 23:17) | 0.45x | buy_fail: order submitted, no tx hash |
| 0x4edfad9a (8 Oct 02:31) | 0.49x | buy_fail: order submitted, no tx hash |
| 0xc0ef16fa (8 Oct 07:15) | 0.40x | skip cap_out (245.5k at check vs 250k floor; harness saw 248k) |
| 0x4d8b2411 (6 Oct 13:27) | 0.48x | not in mc5.db at all (feed never saw it?) |
The 429 alone cost about +$925 on a $50 ticket: today would have been positive.

## Rugs (A1-RUG), unchanged from fw/a1-rug.md
Backtest books them ~0; mint/burn counts are post-entry; real filter needs tx.from (gold part 2, ~10 Oct).

## For Test Manager (execution, not research)
1. Which endpoint returned the 429 at 8 Oct 08:05:48 Bangkok, and does the entry retry? A single 429 should not drop a buy.
2. The two GMGN failure types (confirmation required; submitted without hash): retry or reconcile?
3. Why 0x4d8b2411 never reached mc5.db.
Nothing on firstmin was changed. Any fix there is Ben s call.