SPX 0DTE at-the-money iron fly

The design, in plain terms

Pulling the managed near-ATM condor's shorts all the way to ATM (an iron fly), wings held at the fixed 0.4%, captures more intraday variance premium per session — a real in-sample improvement on mean, Sharpe and average-tail (CVaR), at the cost of a deeper single-worst session — and it is more robust to exit friction than the 0.2% condor. No conditioner, no expected move (EM) wing (falsified), one management lever (uniform 25% PT), held to cash settlement.

0DTE theta is richest at the money. Selling the ATM straddle (fly) collects the most premium; the uniform 25% PT banks a bigger dollar target fast, so the book is a tall cluster of quick small winners — hence the higher mean/Sharpe and the better CVaR (positions come off before the afternoon). The structural cost is gamma: an ATM short goes in-the-money faster on a trend day, so the rare deep loss (capped by the wing) is a touch deeper than the 0.2% condor's. The monotonic 0.2%→0.1%→0 gradient is the signature of a real theta mechanism, not a fit. Exit-friction robustness follows from the same bigger credit.

How it earned a forward slot

(a) The fly holds a positive forward expectancy and beats the S044 condor on mean and CVaR; (b) its worst-session runs deeper than the condor's forward (the gamma cost is real); (c) the Sharpe deflates hard from 3.66 toward ~1.5–2.0 forward (capped-upside illusion), and possibly lower once real ATM PT-exit fills (not mid) are paid; (d) no EM-scaled wing would beat the fixed 0.4% wing forward (the overfit tell); (e) a FAILED — the deeper tail plus real fills erase the edge — is fully possible and publishable.

FROZEN 2026-07-12 (owner "freeze & publish") — the base trade, the single management lever, and the Stage-3 forward gate are locked; a change to any is a new identifier, not an edit here. gates.md section added + ledger cells counted (cumulative 6 → 11). Graduates to forward paper* as the live gate (OOS already corroborated on the owner-directed standing split, §0b — discounted for the 2nd program touch). [CALIBRATE] fields (the forward tail budget) remain owner judgement calls. Sibling of S044; seeded from the Doc Severson / Theta Profits intake (one-way feed). English per repo convention. Location:

Forward record so far

Forward paper-trading against the frozen gate is under way; the record here is machine-generated from the live track and updates on its own. It is interim — no result is claimed until the gate's window completes.

The live forward track has not yet accumulated a reportable window — the frozen promotion gate below states what it must show.

Conditions & caveats

No load-bearing caveats beyond the gate below.

The promotion gate

DSR > 0 after pillar deflation (met: DSR ≈ 0.98 @25 trials); the fly beats the S044 condor on Sharpe and CVaR on common sessions (met) and survives the charged-PT-exit robustness (met). (shorts on the ATM strike, ≈0.50Δ) · 0.4% wing · 09:35 entry · uniform 25% PT · hold to cash settlement · NO active loss-stop (the wing is the defined-risk cap). - Risk posture — sized for staying power, NOT a dollar stop. A hard max-loss budget is a portfolio-level stop, and the stop study (§0c) showed stops are net-harmful for this short-gamma edge (they cut on the bottom and forfeit the recovery that is the edge). So the risk control is sizing, not a cut-off: allocate so that ~2–3× the observed max equity drawdown (obs max DD ≈ $5,316/contract, ~2.5 months underwater; budget ≈ $11–16k/contract — and the true worst is likely deeper, short sample) is a small, survivable slice. Then ride the drawdowns; a drawdown alone is never an exit. - Sizing — DOCUMENTED (decided 2026-07-12, already coded in spx_0dte_managed_runner.managed_contracts; the open [CALIBRATE] was resolved then, gate SHAPE above unchanged): anti-martingale from each cell's OWN realised cushion — 1 contract until ≥60 resolved sessions AND positive cushion, then +1 per $22.4k banked (scaled) net P&L, hard cap 10, steps back down to floor 1 (sizing, not a stop). The per-contract P&L stays the gate basis; contracts is a stored overlay. No active dollar stop (§0c); per-session defined risk = wing − credit (~$0.5–2.7k/contract). The ≥60-session scale threshold matches the edge-review trigger below. Forward paper is LIVE (flag ZERODTE_MANAGED_ENABLED, since 2026-07-12); first managed session 2026-07-13 09:35 ET. - Reserve raised 2026-07-13 (owner "optie 1", after the all-data drawdown study): the per-contract DD reserve was $6k (2× → $12k step) when only the in-sample max DD ($5,316) was known. The full-data study DD is $11,180 — ~2× that, and out-of-sample (2025, still ongoing). Reserve → $11.2k, keeping the 2× buffer → $22.4k/added contract. So scaling now banks ~2× the true observed drawdown per lot, not ~1×. Gate SHAPE unchanged; this is a sizing-parameter recalibration, not a base-trade change. - Edge-review trigger (pre-registered — the discipline on the right axis, not on loss depth): review/pause only if, after ≥ 60 forward sessions, the net expectancy after honest (asymmetric, real) fills is ≤ 0, OR the fly fails to beat the S044 condor on tail-aware terms (CVaR/worst per session actually traded). The trigger is edge broken, never drawdown deep. - Full gate: min 100 sessions, CVaR ≤ the S044 condor benchmark, positive expectancy net of asymmetric fills, deflated for pillar count. Track alongside S044 (same apparatus) — like-for-like fly-vs-condor. - Fail = FAILED, recorded openly** (worst sessions shown). A forward null — the fly's in-sample edge is the flattered-Sharpe illusion and the deeper tail dominates once fills are real — is a clean, publishable result.