# PRE-REGISTRATION — what the desk can actually publish: a strategy suite

**Frozen 2026-09-16 10:15 ET, BEFORE any result below was computed.** Runner: `run_product_strategies.py`.

> **ANSWERED 2026-09-16** by `2026-09-16-product-strategies-results.md`. **29 arms ran (a declared deviation from the 27 frozen here), zero validated.** No day-session entry survives, no overnight timing or calendar refinement beats the plain hold, gap continuation is a coin flip, and three large t-statistics (ZN ×2, 6E) proved to be the round-turn cost rather than drift. The pre-committed fallback in "What I will publish" therefore governs the product decision.

## The question

The desk's published pick family is refuted on its own tape (5,507 trades, ≈0 gross, negative net; `<data-root>/fp-data/studies/2026-09-15-why-the-tape-loses.md`), as are the opening-range family, crown policies, wall and cage rules, flow direction and gamma-conditioned late-day momentum. The Desk Read therefore needs **something to say that survives evidence**, and members who day-trade futures need an honest answer about what the data supports.

This suite tests every candidate the desk could publish from the data on this drive, under one frozen protocol. It is deliberately weighted toward *simple, mechanical, published-in-advance* rules, because those are the only kind a subscription can honestly sell.

## Data, costs and conventions

- Bars: `<data-root>/fp-data/glbx/<ROOT>/1m`, 2010→2026 (event-conditioned arms start 2016, where `fp-context` begins).
- Roots and round-turn friction (2 ticks slippage + $4 commission, the desk's standing assumption): **NQ** $14, **ES** $29, **YM** $14, **RTY** $14, **GC** $24, **CL** $24, **ZN** $35.25, **6E** $16.50. Full-size contracts; micros are one tenth of NQ/ES/YM/RTY/GC/CL and are noted where relevant.
- Quad-witching dates excluded from any return crossing 09:30 ET.
- All results are **net dollars per contract**, long unless stated.
- Splits: price-only arms **train 2010–2018 / confirm 2019–2026**; event arms **train 2016–2021 / confirm 2022–2026**.

## Families and hypotheses (frozen)

**A — Overnight hold, per root (8 tests).** Long the 16:00 ET close, flat the 09:30 ET open. Which of the eight roots pays after friction? *Prior: NQ and GC yes, ES marginal, unknown elsewhere.*

**B — Overnight timing grid, NQ and GC (4 tests, NQ primary).** Entry at **16:00** or **18:00** (Globex reopen); exit at **09:30** or **10:00**. Does the edge live in the Asia/London session or in the US open?

**C — Calendar filters on the overnight (3 tests).** (1) **Turn of month**: the last session of a month plus the first three. (2) **Weekday**: Friday→Monday nights excluded. (3) **Month-of-year**: November–April versus May–October.

**D — The day session (10 tests).** (1) RTH long, 09:30→16:00, per root (8 tests) — the control that says whether *being long during the day* pays at all. (2) RTH return on **CPI/PCE/NFP mornings** versus other days. (3) RTH return on **FOMC decision days** versus other days. Plus a descriptive hour-by-hour profile of NQ (no test attached).

**E — Gap follow-through by event class (2 tests).** After an **AMC mega-cap earnings night** and after a **quiet night**: does the first 30 minutes continue the overnight gap or fade it? Measured as the 09:30→10:00 return signed by the gap's direction.

Twenty-seven inferential tests. **Bonferroni threshold p < 0.00185.**

## Method

Per arm: n, mean net $, t-statistic, win rate, bootstrap 95% interval (10,000 resamples), and a placebo of 1,000 equal-sized random draws from the same root's population. Train/confirm means and sign agreement. An arm is:

- **Validated** — p below the corrected threshold, placebo p < 0.05, signs agree across halves.
- **Suggestive** — p < 0.05 only. Not tradeable, not publishable as a signal.
- **Refuted** — otherwise.

## What I will publish, decided in advance

To stop the result from being reverse-engineered into a story:

1. Anything **validated** goes into the Desk Read as a dated, mechanical rule with its own sample size and net expectancy, carrying the CFTC 4.41 statement.
2. Anything **suggestive** goes on the ledger as an open question, explicitly not a signal.
3. If the day-session families (D, E) come back null — which is my expectation — the honest product answer is that **the desk cannot sell intraday entries**, and members who day-trade get *risk and timing guidance* (which hours are dead, which days to stand down, how wide the distribution actually is) instead of trade calls. That is written here before the numbers exist.

## Prior expectation

The overnight family survives; the day family does not. If that is what comes back, the recommendation is to reposition the product around the overnight and event calendar plus honest day-session risk guidance, not around entries.