# Audit: grid (live mc5 rule replay) - auditor, 9 Oct 2026 23:50 BKK (board #206, #194)
**Verdict: VERIFIED as a fill/rule replay. NOT verified as a profit forecast** (mean rests on ~15 runners).

## Numbers (my own recount of /root/tm/grid/live_cols.csv, 158 trades 13 Sep - 8 Oct, x = NET of 1% fee each way + gas; #194 settled: net)
- mean 2.81x, median 0.48x, 31 trades >= 2x, 15 >= 10x. Without top 3: mean 2.21x. Without top 8: 1.62x.
- Net tickets +286 = (2.81-1) x 158, so "+$2,860 at $10" is arithmetic of the same file. Fine.

## Replay vs REAL live (live wallet page fw/pnl.html, 29 closed trades 6 Oct - 9 Oct 23:27, I used every row)
| | replay (158) | live (29) |
|---|---|---|
| mean x of losers (<1x) | 0.396 | 0.386 |
| trades < 0.1x (rugs/pulls) | 18 = 11% | 3 = 10% |
| trades >= 10x | 15 = 9.5% | 3 = 10.3% (29.4x, 19.5x, 12.8x) |
| trades >= 2x | 31 = 19.6% | 3 = 10.3% |
| mean x | 2.81 | 2.48 (w/o top 3: 2.21 vs 0.39*) |
(*live w/o top 3 is 0.386x because all 26 others are losers; replay w/o top 3 still includes 12 more runners.)
- Loser fill (stop/gap/rug) agrees within 3%: the replay is NOT too kind on losers. Same-coin pairs (A1 fw/a1-live-vs-harness.md, 18 pairs) show live winners 29.4x/19.5x vs replay 24.6x/17.6x: live fills at least as well.
- Real money: live wallet 0.0437 -> 0.0702 ETH, total +$111 (pnl.html). The 13 trades at $50 (8 Oct) were mostly stops with worse fills (mean ~0.37x incl. 2 rugs) and cost -$400; the $10 runners earned +$587. That is the "down 0.14 ETH" story: size and 8 Oct bad luck, not a replay bug.
- Gap that is NOT explained: replay has 16 trades in the 2x-10x band (10%); live has 0 of 29. P(0 of 29 | 10%) = 4.7%. Not proof, watch it. Hit rate >= 2x: P(<=3 of 29 | 19.6%) = 15%: luck.

## 5 checks
1. Sellability: PASS by live evidence: 3 runners (12.8x-29.4x) were really sold on chain, 26 stops sold. No eth_call needed.
2. Outside sellers: live fills at the real pool are the test; they match the replay (above).
3. No look-ahead: rule = cap at +300 s then buy +306 s, same as the live bot (entry delay 294-306 s both). PASS.
4. Fills: 3 hand rebuilds from chain NOT done by me; replaced by the 29 real fills above (stronger). Open item: replay gas is a flat $0.05 vs real GMGN costs at $10; live x already includes them and still agrees.
5. Replay vs live: table above; gap explained except the 2x-10x band.

## Use
Trust the replay for rule comparisons (variants share the same fills). Do not quote +$2,860 as expected profit: expect the ~9.5% runner share to decide everything; at 29 live trades the runner rate is 10.3%, in line. Re-check at 60 live trades (esp. the 2x-10x band).