Does adding a QQQ 0DTE fly diversify the SPX fly?

ID: S049 Slug: spx-qqq-0dte-fly-diversify-v1 Failed at: Exploration — closed unfrozen, never counted against the pillar budget Run date: registered 2026-07-14 · closed 2026-07-20 Outcome: FAILED Headline metric: the +26% Sharpe-lift is void — it was measured against an SPX sleeve that does not exist Fail reason: The study asked whether a QQQ 0DTE fly diversifies the SPX fly. It measured the lift against S045, whose apparent edge was a settlement-proxy artifact. With the SPX sleeve corrected to a loss and the QQQ sleeve already measured at roughly zero, there is no book to diversify and no lift to claim.

What we tested

One SPX ATM fly plus five QQQ ATM flies, sized on a 63-day window, both with a 25% profit-take. The metric of interest was per-trade Sharpe of the combined book versus the SPX book alone, plus the correlation between sleeves and the behaviour of QQQ on SPX's worst days.

What we found

This study had already survived two corrections before the one that closed it, and both are worth keeping in the record:

Correction 1 (2026-07-15) — duplicate minutes. The raw QQQ dataset carried collapsed-minute duplicates: multiple sub-minute ticks stamped to the same minute, about 90% of them at different prices. The engine picked one arbitrarily, which banked profit-take exits that never happened. Cleaned to VWAP 1-minute bars, the QQQ standalone edge roughly halved and the headline lift fell from +69% to +27%.

Correction 2 (2026-07-16) — an impossible session. A P&L-integrity pass found one QQQ session printing +$688 per contract on a structure whose payoff ceiling is about $225. A settlement artifact had survived the VWAP clean. The engine now clips every session to the fly's payoff ceiling. The lift survived this one; the drawdown-reduction claim did not, and was retracted.

The close (2026-07-20) — no base to build on. The SPX sleeve was invalidated. The comparison had nothing left to stand on.

Forward evidence: effectively none. The QQQ fly ran one live session (2026-07-17) before the family was closed, so there is no forward record on this side. The SPX sleeve's forward paper is reported under S045 — positive over 44 sessions, and shown there to be statistically indistinguishable from the corrected expectation.

What we learned

A ratio inherits every weakness of both terms. "Adding B improves A by 26%" is a claim about A as much as about B. When A turned out to be an artifact, the study did not need re-running with better numbers — it needed closing.

Three independent data faults in one study. Duplicate minutes, an impossible print past the payoff ceiling, and an inherited settlement bug — each found by a different check, none by the strategy logic. The structural guard added here (clip every session to the payoff ceiling) is the durable output of this study: defined-risk structures have a mathematical maximum, and anything beyond it is a data fault, never a result.

Closing unfrozen was the right call. This was an ONGOING exploration, never frozen and never counted against the 0DTE pillar's trial budget. The exploration/confirmation split did its job: a study could be pursued optimistically, corrected twice, and then closed without spending a trial or publishing a claim.

What this doesn't tell us yet

Whether QQQ 0DTE flies diversify anything. The question is untested, not answered: the comparison needed a working SPX book as its baseline, and there isn't one. Re-asking it would require a structure that survives correct settlement first.

Whether the sleeve correlation of roughly +0.09 means anything. It was measured between two books that are now both understood to be without edge, so it describes the covariance of two noise processes.

What happens next

Nothing. Closed as an exploration — never frozen, never counted against the 0DTE pillar's trial budget.