# A1-RUG first pass (Analyst 1, 8 Oct 2026 ~21:00 Bangkok). Answers fw/tm-rugs-mc5.md
Stage 1 notes, not a result. Harness run: research-box /root/fwa1/rug (fw_mc5.json, gold to 8 Oct 20:07, 177 trades).

1. Backtest books rugs honestly. 20 of 22 harness "pull" exits book 0.0x; 11 more stops book under 0.1x. 31 of 177 trades (17%) are near-total losses in the backtest already. Live: 3 of ~29. So the rugs are priced in, not a backtest blind spot.
2. Mint/burn counts are NOT an entry signal. On the tape, both checked rug pools had 1 mint / 0 burns at +300 s entry; burns came at +337/+348 s (20:05 pool) and +542 s (7 Oct pool). The mc5.db counts run to exit.
3. The 20:05 "rug" was a dump, not a pull: at +327 s many sells in one block drop cap ~390k -> ~2.5k; the burns 10-20 s later only take what is left. Same shape as the TM wallet finding (operator selling into +5 min buyers), just in one block.
4. LP owner: every mint comes from 0x73991a25... = position manager (150/150 sampled pools). Who really holds liquidity needs tx.from / position owner -> golden tape part 2 (~10 Oct).
5. Mismatch to check: the 7 Oct (0x4fc80739) and 20:05 (0x07b59c1e) rug pools are on the tape but NOT in the harness fw_mc5 trade list. Live bot and backtest rule may not pick the same coins.

Next (Analyst 1): live vs harness coin overlap for all live trades; seller-wallet signature (same wallet dumping across pools, one-block sell bursts) as a pre-entry feature once tx.from lands; winners vs rugs vs rest with a no-filter baseline; kill if it removes more runner $ than rug $.