# A1.17 Size tilt hunt (Analyst 1, 6 Oct 2026, 16:20 Bangkok)

## Pre-registration (6 Oct 2026 16:25 Bangkok, fixed, no changes later)
- **Rule:** "fast gate" = the $1M gate comes within 60 minutes of pool creation. Fast coins get a 1.5x ticket, slower coins 0.5x, rescaled to equal total stake vs the flat $1k book.
- **Test set:** half C only = wide-rule gates from 7 Oct 2026 00:00 Bangkok. Run once, when half C has 30+ wide trades. Same cut, no other cuts tried.
- **Pass:** $ a day up vs the flat book AND worst day no worse. Anything else = fail, and the tilt is dropped.

## Read this first
- **Nothing ships. No gate feature passes both tests: $ a day up in both halves AND worst day no worse.**
- **The closest is "fast to the gate".** Coins that reach the $1M gate within 60 minutes of launch get 1.5x, the slower ones 0.5x, with the same total stake. With one fixed cut, it raises $ a day in both halves:
  - Half A: +$6,656 a day vs +$5,412 base.
  - Half B: +$4,634 vs +$4,142.
  - But the worst day in half A goes from +$223 to −$1,687, so it fails your rule.
  - Walk forward, the cut moves (68 minutes picked on A, 35 on B). Gains then: +$164 on B, +$1,166 on A.
  - Worth a paper tag (`min_to_gate`), not money yet.
- **"Gate before 08:00 Bangkok" behaves the same way:** up in both halves (+$7,154 / +$4,459) but worse worst day in A (−$448). It's likely the same effect as fast gates: coins that gate before 08:00 took a median 33 minutes to get there, vs 63 for the rest. Paper only.
- **Farm-funded and Kit B flip between halves**, so they can't be tilts.
  - Walk forward with equal stake: farm −$835 a day on B and −$290 on A.
  - Kit B (picked on A) +$338 on B. Picked on B, the rule gives nothing in A, because A has no Kit B coins.
- **Everything else loses or does nothing** walk forward: deployer funding size, pre-gate buyer count, ETH in pool at gate, Kit B launches in the last 24 h.

| Feature (rule picked on one half, tested on the other, equal total stake, $1k base) | Test B gain | Test A gain | Worst day still OK? |
|---|---|---|---|
| minutes launch to gate (fast gets 1.5x) | +$164 | +$1,166 | no (A −$590 vs +$223) |
| gate hour (early gets 1.5x) | +$379 | +$670 | no (A −$499) |
| Kit B | +$338 | $0 (no Kit B in A) | n/a |
| Kit B launches in last 24 h | +$329 | $0 (no Kit B in A) | n/a |
| farm-funded | −$835 | −$290 | no |
| Kit A | −$349 | −$251 | no |
| deployer ETH at launch | $0 | −$301 | |
| pre-gate buyer count | −$227 | −$241 | no |
| ETH in pool at gate | −$972 | −$755 | |

- First principles: a size tilt only pays if the feature changes the odds of the 5x vs the 0.3x stop. With 34 to 36 trades per half, a split leaves about 17 trades per class, where one 5x more or less (about +$4,000) swings the result. So only a split that holds in both halves with the same cut can be trusted. None does yet.

## Method
- **Trades:** Analyst 2's 70 wide-rule trades (sw-a2-16-wide-trades.csv), x at $500 with the 0.3x stop.
- **Size:** each ticket's x is re-priced from $500 to its own size using the pool's real WETH at the gate block as the constant-product reserve (buy impact R ÷ (R + X), sell impact at R × √x). That is conservative, because v3 virtual depth is larger.
- **Tilt:** 1.5x on the better class, 0.5x on the worse, then rescaled so the total stake equals the flat book. Half A = 21 to 27 Sep gates, half B = 28 Sep to 6 Oct. Rules were picked on one half (best of the 33%, 50% and 67% cuts and direction) and tested on the other.
- **Features at the gate:** farm-funded (A1.16), kit (A1.14), deployer ETH just before the launch block, unique pre-gate buyers, minutes from pool creation to the gate block, WETH held by the pool at the gate block, gate hour Bangkok, and Kit B launches in the 24 h before the gate.

## Asks for Supervisor
- No size tilt goes live. Coder: log `min_to_gate` and `gate_hour` on every live trade as paper tags. Re-test when half C (from 7 Oct) has 30+ trades, using the fixed 60-minute cut decided now.

First principles score: 80%. Every feature is known at the gate and read from chain, and the test is walk forward with equal stake. Missing: Analyst 2's own $1k tick-walk fills per trade (I re-priced from $500 using pool WETH), and a third half to settle the fast-gate cut.

Scripts: /root/a1/a117/feat.py, tilt.py, tilt_norm.py (review container). Latest: http://159.223.88.168/hx-c9ois/sw/a1-17-size-tilt.md