# A1.19 More trades a day: other checkpoints and bands (FW Analyst 1, 6 Oct 2026, 15:35 Bangkok)

**Verdict: FAIL. No new cell passes the H1 pick.** Every cell's NEW coins (coins the base does not already hold) lost money on H1, so none reached the H2 test, and the combined book equals the base.
- On H2 three cells look positive: +5 min $1M to $2M, +10 min $500k to $2M, +10 min $1M to $3M. But H1 has only 3 to 8 new coins per cell, and all lost.
- So they are **watch cells for paper logging, not rules.**
- Most of what +10 / +15 min finds is coins the base already holds: overlap 63 of 83 and 57 of 78 on H2.

## Setup
- 816 v3 WETH pools with a market cap of $100k+ at +5 min and 4.5 h+ of data. Swaps and Mint/Burn were pulled to 12 h: 1,083,790 more swaps, 1,444 Mints, 1,297 Burns.
- Every cell: $100, real-depth fill at the checkpoint, stop 0.5x, **time stop 4 h after entry**, pull = 0 (Burn of 50%+ of minted liquidity units), one position per token (the first signal wins).
- $ a day is divided by all days in the half (H1 13 days incl. quiet 13 to 21 Sep, H2 12 days), so the base numbers differ a little from A1.18 (which used trading days and dropped 6 Oct).
- Coins under $100k at +5 min that reach $500k+ by +10 / +15 min are not in the data (no Mint/Burn pulled). There are probably a few.

## Per cell ($100; NEW = coins not in base)
| cell | half | all trades (a day) | mean / median | $ a day | overlap with base | NEW coins: n, mean, $ a day |
|---|---|---|---|---|---|---|
| **base +5 min $250k to $1M** | H1 | 42 (3.2) | 2.66x / 0.45x | $536 | n/a | n/a |
| base | H2 | 158 (13.2) | 2.05x / 0.48x | $1,386 | n/a | n/a |
| +5 min $1M to $2M | H1 | 3 (0.2) | 0.21x / 0.20x | -$18 | 0 | 3, 0.21x, -$18 |
| +5 min $1M to $2M | H2 | 20 (1.7) | 1.71x / 0.48x | $119 | 1 | 19, **1.80x, +$127** |
| +10 min $500k to $2M | H1 | 22 (1.7) | 2.09x / 0.47x | $185 | 18 | 4, 0.22x, -$24 |
| +10 min $500k to $2M | H2 | 83 (6.9) | 1.83x / 0.47x | $575 | 63 | 20, **1.46x, +$77** |
| +10 min $1M to $3M | H1 | 10 (0.8) | 0.32x / 0.32x | -$53 | 2 | 8, 0.34x, -$41 |
| +10 min $1M to $3M | H2 | 31 (2.6) | 1.46x / 0.49x | $119 | 7 | 24, **1.31x, +$62** |
| +15 min $500k to $2M | H1 | 29 (2.2) | 1.61x / 0.45x | $137 | 20 | 9, 0.30x, -$49 |
| +15 min $500k to $2M | H2 | 78 (6.5) | 1.84x / 0.49x | $549 | 57 | 21, 0.80x, -$34 |
| +15 min $1M to $3M | H1 | 12 (0.9) | 0.72x / 0.43x | -$26 | 4 | 8, 0.42x, -$36 |
| +15 min $1M to $3M | H2 | 36 (3.0) | 1.27x / 0.50x | $82 | 11 | 25, 0.72x, -$57 |

## Pick and test
- H1 pick rule: a cell's new coins must make money on H1. **No cell passed.**
- Combined book = base: H1 $536 a day, worst 3 days -$457; H2 $1,386 a day, worst 3 days +$400.
- The +15 min cells lose on new coins in both halves: **dead.**

## What I propose
- **Keep the rule as is:** +5 min, $250k to $1M, stop 0.5x, hold 4 h.
- Price Predict can log **+5 min $1M to $2M** and **+10 min $500k to $2M** as paper-only watch cells (no money), to collect enough new coins for an honest H1/H2 retest after about a week.

Code: a1code/v3cells.py, v3ext.py (research-box /root/fwa1).