# A1.8 Walk-forward check of the pull-time sell rule (Analyst 1, 6 Oct 2026, 13:15 Bangkok)

## Read this first
- **The rule holds walk-forward, and it doesn't hang on 90 s and 200 s.**
  - Each crew's median pull time comes only from its coins that ended before the new coin launched (at least 5), so it is recomputed with every coin.
  - All 6 cut-point pairs are positive on every day. At 10 s they land between +211% and +236%. At 35 s, between +129% and +136%.
- **But the cut points barely matter, because almost every hot crew pulls after 200 s.** So the rule mostly means "all at 10x".
  - Typical walk-forward median pull: A 263 s, B/C 414 s, `0x0427e497f8075524fdc6a73ed6dfe1be958a2a56` 246 s, `0x7028bab73f82dff14b199593abb76c8b6acb3aa9` 3,983 s, `0x2dacdf0f44d07234832799a10a7cc5c8eff0223a` 4,194 s.
  - Only `0xca24b6f1b00a7d336ef2984239daa836f17cb6c1` (48 s) and `0x0ec66d31aab7e387b2f4cda075e39f7a6a0f541b` (60 s) get the "all at 2x" plan.
- **Big catch from A1.6 and A1.7:** this assumes our buy is ignored. In A and B/C it may not be, and then "all at 10x" turns into −95% while "all at 2x" stays at +35%.
  - So the final rule is: **A and B/C sell all at 2x; everyone else uses the pull-time rule.**

## Results (depth fills, $20, fees in)
| Plan | Hot + tag 10 s (159 coins, 30 Sep to 4 Oct) | Hot 35 s (165 coins) | All hot coins 10 s (200 coins, 26 Sep to 4 Oct) |
|---|---|---|---|
| **Walk-forward, cuts 90 s / 200 s** | **+228%**, win 43%, 5/5 days | +129%, 5/5 | +222%, 8/8 |
| cuts 60 / 150 | +236%, 5/5 | +136%, 5/5 | +228%, 8/8 |
| cuts 60 / 200 | +228%, 5/5 | +129%, 5/5 | +222%, 8/8 |
| cuts 90 / 300 | +217%, win 54%, 5/5 | +134%, 5/5 | +215%, 8/8 |
| cuts 120 / 300 | +217%, 5/5 | +134%, 5/5 | +215%, 8/8 |
| cuts 150 / 400 | +211%, win 58%, 5/5 | +129%, 5/5 | +211%, 8/8 |
| Fixed: all at 10x | +251%, win 35% | +135% | +240% |
| Fixed: half 3x, rest 10x | +188%, win 70% | +99% | +177% |
| Fixed: all at 2x | +96%, win 89% | +44% | +84% |

- Cutting earlier (60 / 150) helps a little, by sending the fast pullers to 2x sooner. Cutting later (150 / 400) costs up to 54 points on 4 Oct.
- Fixed "all at 10x" beats every cut point because the hot list is mostly slow pullers. It is the riskiest plan, though (win rate 35%).

## Method
- Hot filter and launch tag as in A1.1 and A1.4. Depth fills as in A1.7.
- Median pull is counted in seconds from the pool's first block, using only that crew's coins that ended (pulled, or traded for the last time) before this coin's first block, with at least 5 of them. With fewer, the plan defaults to half 3x / rest 10x.
- Plans: under cut 1, all at 2x. Between the cuts, half at 3x and the rest at 10x. Over cut 2, all at 10x.

First principles score: 85%. The rule is now fully walk-forward and tested at 6 cut points. Missing: only 5 to 8 days, the hot list is close to one plan type, and the A/BC trigger risk (A1.6) can flip the best plan.

Scripts: /root/a1/a17/a18.py, grid.py (review container). Latest: http://159.223.88.168/hx-c9ois/sw/a1-8-walk-forward-sell.md