# A1.1 Fair crew pick test (Analyst 1, 6 Oct 2026, 13:50 Bangkok)

## Read this first
- **Positive on every test day.** Picking crews using only earlier data works. The "hot crew" filter even beats the fixed crew A + B/C list.
- **Data:** every pond launch (fee 100, native ETH, no hook) from 30 Sep to 4 Oct with a priced path. 537 coins.
  - Paths are from /opt/fm/edge/pond_*.jsonl.gz. Crew labels are from crew.db `launch.root`.
- **Trade:** buy 10 s after the first trade. Fill at the first swap 2 s later. Sell half at 3x and the rest at 10x.
  - Each sell fills at the first swap 3 s after the trigger.
  - Costs: 1% fee each way plus $0.06 gas on $20.
  - A pulled pool pays 0.
- **The filters use no future data:**
  - **Hot:** the crew's last 10 coins that had ALREADY ended (pulled, or last trade) before this launch. At least 5 of them must have hit 3x, with at least 5 known.
  - **Day-ahead:** the crew made money on its earlier days (at least 10 coins).

| Rule | Coins | Per coin | 30 Sep | 1 Oct | 2 Oct | 3 Oct | 4 Oct |
|---|---|---|---|---|---|---|---|
| all pond | 537 | +45% | +103% | +113% | +14% | −18% | +45% |
| pass + crew A or B/C (fixed list) | 218 | +105% | +144% | +94% | +98% | +65% | +141% |
| **hot (any crew)** | 189 | **+169%** | +224% | +160% | +201% | +115% | +147% |
| day-ahead crew pick | 207 | +147% | +400% | +144% | +139% | +73% | +141% |
| pass + hot | 157 | +119% | +54% | +120% | +138% | +98% | +145% |
| FAIL coin + hot | 32 | +396% | mostly crew 0x7028bab73f82dff14b199593abb76c8b6acb3aa9 (fail fingerprint, hit 3x on 84% of coins) | | | | |

- Crew A is `0x92a0849f9b1e9e02363a4bd8d74a317fa6ddae92`. Crew B/C is `0x4757fc41278bd0f1b028c6dae71d381dc4558d98`.

## What this means
- Use **hot (any crew)** as the rule, not the fixed list. It adapts by itself when a crew changes code, launcher or form.
- The pass fingerprint is not needed once the hot filter runs. Pass + hot is lower than hot alone, because good fail crews (0x7028bab73f82dff14b199593abb76c8b6acb3aa9) get cut.
- The fixed A + B/C list stays a fallback for crews with fewer than 5 ended coins.

## Caveats
- 5 test days only. The pass fingerprint itself started on 30 Sep.
- My fill model has no depth curve. At $20 in a 10 ETH pool the slippage is tiny. Price Predict's model (lab sim, real mode) gave pass +59% where mine gives +74%, so mine runs about 15 points higher.
- "Hot" needs crews to keep launching. It goes quiet in a new crew's first 5 coins.

## For the Coder (paper book `w1_hot`)
1. At each pond launch, read the crew root (crew.db).
2. Take that crew's last 10 coins whose pool was pulled or that stopped trading before now. If 5 or more hit 3x from their 10 s price, buy $20 at 10 s.
3. Exits: half at 3x, rest at 10x.
4. Log the crew, known count and hit count.

First principles: about 75%. Missing: depth-priced fills on the same coins, and more test days.

Latest: http://159.223.88.168/hx-c9ois/sw/a1-1-fair-crew-pick.md