# Gamma-wall migration — pre-registration (frozen 2026-09-09 13:55, before any policy run)

## Established facts (from the hourly archive, 33 sessions)
- The QQQ strike book changes **once per session**: the 09:10 snapshot equals the prior day's dated book (r = 1.00), the 10:10 snapshot equals the same day's dated book (r = 1.00), and 10:10 → 16:45 the book is frozen. Wall changes after 10:10 are spot-relative selection only, not book changes.
- Therefore the migration a trader sees at 10:10 on session D is exactly `history[D−1] → history[D]`, reconstructable for 730 sessions (2023-10-10 → 2026-09-08).

## Measures (computed at 10:10 ET on D)
Old walls: D−1 book at the 09:10 spot (prev close × NQ 09:10 / NQ prior close). New walls: D book at the 10:10 spot (prev close × NQ 10:10 / NQ prior close). Walls/flip/tilt via `computeGamma` unchanged.
- Δcw, Δpw = new − old, as % of spot; Δmid = (Δcw+Δpw)/2.
- Δtilt = tilt_new − tilt_old. Regime from the new tilt.
- Flip cross: sign(spot − flip) changed between old and new.

## Policies (Part A fill rules: entry = next 1m open after the 10:10 signal, 0.25 slip, stop = max(12, 2×ATR3 of 5m bars), target 1.5R, cap 60 min / 15:59)
- **M1 wall drift**: Δcw and Δpw have the same nonzero sign → trade in that direction.
- **M2 tilt shift**: Δtilt ≥ +0.08 → long; ≤ −0.08 → short.
- **M3 flip cross**: flip migrated across spot → side = sign(spot − flip_new).
Descriptive (no verdict): NQ 10:10→16:00 return by Δmid tercile; P10 cage held with new vs old walls; share of sessions where each wall moved.

## Decision rule
Per policy: mean R > 0 with one-sided p < 0.0167 (Bonferroni over 3), positive in both halves (split 2025-03-15), n ≥ 100 triggers. Otherwise REFUTED. Quad-witching sessions excluded. No variants, no threshold tuning; a failed policy is closed.