# Pre-registration — mechanical crown-play policies (NQ) · 2026-09-09 **Question.** Is there a fixed, mechanical rule mapping the 09:20 ET state to a play that has positive expectancy under the desk's own fill rules — i.e. a crown play that can be backtested and then updated from its own grades? ; this is its replacement candidate set. **Frozen before the simulator reads a bar.** Script `fp-backtest/crown_policies.py`. Data: NQ 1-minute (GLBX .c.0, 2012-01-03 → 2026-09-04), VXN daily. Excluded: quad-witching Fridays, sessions < 300 RTH bars, no prior RTH day within 4 calendar days, sessions lacking 21 prior RTH days for ATR20. ## Fill rules (the grader's, unchanged) Signal = a **5-minute bar close** through the level (bars aggregated from 1-minute, ET-aligned). Fill = open of the next 1-minute bar ± 0.25 pt slippage against the trade. Stop = **2.0 × ATR(3) of 5-minute bars** at signal, floored at 12 pt. Target = entry ± 1.5 × stop distance. Exit on stop (assume stop when a bar touches both), on target, at **60 minutes after fill** (bar close), or at 15:59 ET. Trigger window 09:35–12:00 ET; one play per session per policy; first signal wins. R = (exit − entry)/stop distance, signed. Dollars at 1 NQ = 20 × points, no commissions (≈ $4.5 round-turn would subtract ~0.006R). ## Inputs available at 09:20 (all from the archive) Prior RTH close, prior RTH VAH/POC/VAL (70 % volume area, 5-pt bins), overnight high/low (18:00 → 09:29), ATR20 = mean prior-20 RTH ranges, gap = (09:30 open − prior close)/ATR20, ON range/ATR20 with the 2026-09-08 quintile edges (0.422, 0.55, 0.687, 0.9), ON-last position vs prior VA. ## Policies - **P1 Expansion breakout** — only when ON range ∈ Q4–Q5: long on the first 5m close above the overnight high; short on the first 5m close below the overnight low. - **P2 Contained fade** — only when ON range ∈ Q1–Q2: after a 5m close outside the prior value area, the next 5m close back inside it → trade toward POC (long from below VAL, short from above VAH). - **P3 Gap fade** — gap ≤ −0.033×ATR20 (Q1–Q2): long on the first 5m close above the prior close; gap ≥ +0.157 (Q4–Q5): short on the first 5m close below it. - **P4 Value continuation** — ON last above prior VAH: long on the first 5m close back above VAH after a 5m close below it (a held pullback); mirror below VAL. - **P5 Unconditional ON breakout** — P1 without the day-type condition (control for the conditioning). - **P6 Opening-range breakout** — first 15 minutes' high/low, 5m close beyond (control; previously refuted). - **P7 Regime switch** — P1 on Q4–Q5, P2 on Q1–Q2, no play on Q3. ## Reporting Per policy: sessions eligible, triggers, W / L / time-out, expectancy R per trigger and per eligible session, hit rate, total and per-year P&L at 1 NQ, max drawdown in R, split 2012–2018 / 2019–2026 and by year. ## Decision rule (frozen) A policy is **VALIDATED** only if: expectancy ≥ +0.10 R per trigger in BOTH halves, ≥ 100 triggers overall, and per-year expectancy positive in at least 10 of 15 years. Validated policies become bandit arms (Phase 2); the crown play is then the arm with the best posterior for the day's regime. If none validates, the crown play is retired as a directional call (becomes "primary scenario") and the sizing line moves to the opportunity watch. Caveat recorded: the quintile edges were estimated on this same archive on 2026-09-08 (weak leakage, five thresholds); the year-by-year and half splits are the robustness check. ## PART B (pre-registered 2026-09-09, after Part A's null) — geometry grid Same seven policies, same signals and fills; only the exit geometry varies: target ∈ {1.0R, 1.5R, 2.0R} × time cap ∈ {60, 120, 240 min} × stop ∈ {1.5, 2.0, 3.0} × ATR(3). 189 combinations. Decision rule as Part A (≥ +0.10 R/trigger in both halves, ≥ 100 triggers, ≥ 10/15 positive years) AND the extreme-cell test at α = 0.05/189 = 0.00026 (t-test of mean R vs 0). Anything short of that is REFUTED; ≥ +0.05 R both halves is "suggestive" and may only enter a forward shadow, never the live crown. ## PART C (pre-registered 2026-09-09) — gamma-wall policies, FORWARD SHADOW The gamma archive (fp-data/gex, QQQ 09:10 ET snapshots) starts 2026-07-29 — too short to validate anything; this part defines the policies and starts their ledger. Walls per the app's `computeGamma` (nearest strike with ≥50 % of peak call/put GEX within ±6 % of spot, 0.3 % ATM buffer; flip = cumulative-net minimum; tilt = net/gross, |tilt| < 0.08 = neutral), computed from the 09:10 QQQ snapshot, scaled to NQ by (NQ 09:30 open ÷ QQQ 09:10 spot). Fill rules as Part A. - **P8 long-γ fade**: tilt ≥ +0.08 → after a 5m close above the call wall (below the put wall), the next 5m close back inside → short (long) toward the flip; target 1.5R. - **P9 short-γ break**: tilt ≤ −0.08 → long on a 5m close above the call wall, short on a 5m close below the put wall. - **P10 cage test**: any regime — did price close the RTH session inside [put wall, call wall]? (descriptive; the live grader's "cage held"). Decision rule: same as Part A, evaluated only once ≥ 100 triggers per policy exist (≈ 6–9 months). Until then the ledger is reported as n-too-small; nothing from Part C may drive the live crown. ## PART D (pre-registered 2026-09-09) — first-touch wall fade, live-graded forward shadow Descriptive prior (31 sessions with recorded walls, 24 first touches): fade ≥15pt before break ≥15pt 67 % (put wall 71 %, call wall 60 %); first touches in 09:30–10:00 faded 54 %, afternoon touches 6/6. Hypothesis, not a result. **Shadow rule (in the live grader, `lib/scoreBrief.ts` → `score.wallFade`):** for each of the call wall and put wall recorded at 09:20 from the NDX+QQQ book, if the wall sits on the correct side of the 09:30 open (call above, put below) — a LIMIT fade at the wall on its first touch during RTH (fill at the wall or the bar's open, whichever is worse), stop = max(12 pt, 0.5 × expected 5-minute range) beyond the wall, target = 1× risk toward the cage, exit at stop / target / session close. Grading is mechanical and identical every day; regime and tilt are stored with the record. **Decision rule:** evaluate at ≥ 100 touched walls. VALIDATED if hit rate ≥ 57 % at 1R (breakeven 50 % + costs + slippage margin) in both halves of the ledger and per-regime sign agrees with the long-γ dampening mechanism (long-γ ≥ short-γ). Until then: shadow only, shown on the RECORD tab; nothing feeds the live crown.