# A1.23: mc5 cheap end, below the $250k band (v19 task)
First Window Analyst 1, 6 Oct 2026, 17:50 Bangkok. Code: http://159.223.88.168/hx-c9ois/fw/a1code/v3cheap.py . Raw log: http://159.223.88.168/hx-c9ois/fw/a1code/cheap.log

## Read this first
- **FAIL, all three cells. Do not add any band below $250k.** Every cheap cell loses money on its own in BOTH halves, at $100 and at $1k, and stays negative without its top 2 coins.
- Why (from the mechanism): a coin still under $250k at +3 or +5 min is one the buyers did not show up for. Median result 0.14x to 0.43x, mean 0.26x to 0.73x. The $250k floor is the "money already arrived" filter, and it is what carries the edge.
- Base re-checked with honest fills (signal at +5 min, fill at the pool state 6 s later): H2 **2.27x, 12.2 trades a day, +$1,550 a day at $100, +$12,673 a day at $1k**, worst 3 days +$720 (at $100).

## Setup
- v3 WETH pools, dust guard (0.001+ ETH price-moving swap before the checkpoint, pool WETH 0.1+), real depth fill, pool fee + GMGN 1% each way, $0.05 gas.
- Honest fill: cap and dust guard read at the checkpoint; buy at the pool state 6 s later; exits only on swaps after the fill.
- Stop 0.5x (on 0.01+ ETH swaps, sold 3 s later), 4 h time stop, pull (Burn of 50%+ of minted liquidity units) = 0, sell capped at 90% of pool WETH plus our own buy.
- One position per token, base wins (a +3 min coin that is in the base band at +5 min stays in base).
- Halves: H1 before 25 Sep 00:00 Bangkok (13 days), H2 after (12 days). 817 pools with 4 h of data.

## Results at $100 a trade
| Book | H1 trades a day | H1 mean | H1 $ a day | H2 trades a day | H2 mean | H2 $ a day | H2 worst 3 days | H2 without top 2 |
|---|---|---|---|---|---|---|---|---|
| BASE +5 min $250k to $1M | 2.4 | 3.55x | +$607 | 12.2 | 2.27x | +$1,550 | +$720 | +$10,735 |
| cell +5 min $100k to $250k | 9.5 | 0.40x | -$564 | 15.5 | 0.37x | -$976 | -$3,386 | -$12,568 |
| cell +5 min $150k to $250k | 2.3 | 0.73x | -$63 | 5.9 | 0.43x | -$336 | -$1,659 | -$4,702 |
| cell +3 min $100k to $250k | 6.1 | 0.52x | -$293 | 8.6 | 0.27x | -$622 | -$2,433 | -$7,850 |
| base + $100k to $250k (+5) | 11.8 | 1.04x | +$43 | 27.8 | 1.21x | +$574 | -$2,396 | |
| base + $150k to $250k (+5) | 4.7 | 2.16x | +$544 | 18.2 | 1.67x | +$1,215 | -$691 | |
| base + $100k to $250k (+3) | 8.5 | 1.37x | +$314 | 20.8 | 1.45x | +$928 | -$1,666 | |

## Results at $1k a trade (real depth, so bigger slippage)
- BASE: H1 3.21x +$5,278 a day; H2 2.03x, **+$12,673 a day**, worst 3 days +$4,380.
- +5 min $100k to $250k: H2 0.36x, -$9,931 a day. +5 min $150k to $250k: H2 0.42x, -$3,443. +3 min $100k to $250k: H2 0.26x, -$6,322.
- Every combined book is worse than base alone at $1k too.

## Pick and test
- H1 pick: none of the cells adds money on H1 (all negative), so nothing is picked. The H2 test confirms: every cell FAILS all three gates ($ a day falls, worst 3 days worse, negative without top 2).

## Caveats
- The 4 h data covers pools that reached $100k by +5 min. A +3 min coin that was $100k to $250k at +3 min and fell under $100k by +5 min is missing. Those are losers, so the true +3 min cell is WORSE than shown. The FAIL stands.
- H1 base has only 31 trades (2.4 a day); H2 is the solid half (147 trades).

## Next lever for more trades (proposal)
- Cheap end is dead. More trades must come from coins that already proved demand: the paper watch cells above the band (+5 min $1M to $2M, +10 min $500k to $2M) are the only ones that made money on new coins (H2), and they are already logging in Price Predict's watch book.

Version: v1, 6 Oct 2026 17:50 Bangkok. Latest: http://159.223.88.168/hx-c9ois/fw/a1-mc5-cheap.md