# Gamma-wall migration — results · 2026-09-09 13:10

Pre-registration: `2026-09-09-wall-migration-prereg.md`. Sessions 693 (2023-10-11 → 2026-09-04), quad-witching excluded.

## Descriptive
- Wall moved at the 10:10 refresh: call 49%, put 53%; both same direction 22%; |Δtilt| ≥ 0.08 51%; flip crossed spot 16%.
- Cage held (RTH close inside walls): new walls 72% vs old walls 68%.
- NQ 10:10→16:00 return by Δmid tercile: down -0.094% (up-share 52%, n=231), flat +0.018% (up-share 58%, n=231), up +0.099% (up-share 57%, n=231).

## Policies (Part A fill rules from 10:10)
- **M1_wall_drift**: triggers 154 (long 69/short 85), W/L/T 14/38/102, R/trigger **+0.050** (p=0.224), halves -0.041 (n=61) / +0.110 (n=93), by side long +0.043 / short +0.056, $ at 1 NQ +27,883 → **REFUTED**.
- **M2_tilt_shift**: triggers 349 (long 179/short 170), W/L/T 30/83/236, R/trigger **+0.065** (p=0.071), halves +0.069 (n=158) / +0.061 (n=191), by side long +0.129 / short -0.002, $ at 1 NQ +81,313 → **REFUTED**.
- **M3_flip_cross**: triggers 113 (long 59/short 54), W/L/T 18/31/64, R/trigger **+0.115** (p=0.099), halves +0.107 (n=56) / +0.124 (n=57), by side long +0.324 / short -0.113, $ at 1 NQ +23,495 → **REFUTED**.
## Interpretation
- The refresh is a single daily event landing between 09:10 and 10:10 (most likely the day's OI posting). Half the sessions move at least one wall; only 22 % move both walls the same way.
- Migration carries a **weak** directional lean into the close: Δmid top tercile +0.10 % vs bottom −0.09 % (up-share 57 % vs 52 %), about 0.2 % spread ≈ 55–60 NQ pts of average drift, not a trade by itself.
- All three policies are positive in both halves but none clears p < 0.0167; most triggers time out at the 60-minute cap (M2: 236 of 349), i.e. the lean is slow, not a burst. M2 tilt-shift long side +0.129 R and M3 flip-cross long side +0.324 R are the only cells worth remembering; the short sides are ≤ 0.
- **Verdict: REFUTED as mechanical 10:10 entries; suggestive as a context factor.** Per the pre-registration these policies are closed on this sample. A legitimate re-test is a *forward* ledger only (new data, new pre-registration): e.g. the 10:10 refresh surfaced on the DESK as "walls migrated ▲/▼, tilt Δ" with M2/M3 shadow-graded, verdict at ≥100 forward triggers.
- Cage with the refreshed walls holds 72 % (vs 68 % with the pre-open walls) — the refreshed book is the better range prior for the rest of the session; the app's gamma panel already re-reads walls hourly, so this is already the live behaviour.