ID: S047
Slug: spx-0dte-atm-fly-scalp-v1
Failed at: Stage 1 (in-sample, TRAIN)
Run date: 2026-07-12
Outcome: FAILED
Headline metric: every forced-exit cell net-negative in-sample; Deflated Sharpe Ratio (DSR) ≈ 0.00 across the board (only hold-to-settlement is positive)
Fail reason: Doc's actual scalp management — a tight profit-take plus a ±expected-move price stop plus a 10:35 "first-hour" time-exit — is mechanically loss-making on the S045 fly. Every cell with any forced exit posts a negative annualised Sharpe and a worse tail than simply holding to settlement; the only positive configuration is holding to settlement, where the wing is the cap.
What we tested
A locked exit-management knob on the frozen S045 base (short call + short put on the ATM strike, parity-centered, 0.4% wing, 09:35 entry, 1 contract): Doc's actual scalp management — a tight profit-take of {5%, 10%} of credit (25% carried as the S045 reference), a price stop at ±expected-move (proxied as a stop at 90% of the per-session max-loss), and a 10:35 "done within the first hour" time-exit. Each exit was tested separately and combined, against hold-to-settlement. Forced exits pay a full §3a second cross; the profit-take fills at ≈ mid. In-sample TRAIN only, 444 sessions.
What we found
Every forced-exit cell is net-negative; only hold-to-settlement is positive:
| profit-take (PT) | exit rule | mean $ | ann SR | worst $ | CVaR₅ $ | win % |
|---|---|---|---|---|---|---|
| 5% | hold-settle (ref) | +18 | 1.52 | −805 | −379 | 89% |
| 10% | hold-settle (ref) | +63 | 3.69 | −850 | −662 | 94% |
| 5% | ±expected move (EM) stop (90%) | −27 | −1.68 | −948 | −644 | 82% |
| 10% | ±EM stop (90%) | −34 | −1.45 | −1021 | −720 | 77% |
| 5% | stop + 10:35 (Doc's full) | −37 | −2.58 | −899 | −540 | 73% |
| 10% | stop + 10:35 (Doc's full) | −66 | −3.73 | −899 | −547 | 45% |
Deflated Sharpe ≈ 0.00 across every scalp cell. Two clean sub-findings: the stop makes it worse, not better — the 5% profit-take is +$18 held-to-settle but −$27 with the ±EM stop, and the tail worsens too (worst −$948 vs −$805; CVaR₅ −$644 vs −$379); it fires on 11–22% of sessions and locks in a ~90%-max-loss on days the short-gamma fly would have recovered. And the time-exit is also harmful (the same forfeited-theta lesson as S044), with Doc's full combo the worst of all.
What we learned
Doc's entire scalp management — tight profit-take + ±EM stop + first-hour exit — is mechanically loss-making on this fly; the only configuration that works is holding to settlement, where the wing is the cap. His live survival is most likely his discretionary regime-skipping (sitting out corrective / trending tapes), not the stop — which a mechanical test can't replicate and which mechanically hurts here.
What this doesn't tell us yet
Because the likely true source of the edge is discretionary (skipping bad tapes), this falsifies the mechanical scalp, not a disciplined discretionary version of it — untestable here, and claimed neither way. It is in-sample TRAIN only, on the assembled 0DTE surface, not forward marks. Separately, a by-product lead surfaced that this fail does not close: a tighter profit-take held to settlement (10% + hold-settle) posts annualised Sharpe 3.69 ≈ S045's 3.68 with a materially smaller tail (CVaR₅ −$662 vs −$891; worst −$850 vs −$1,791) — a profit-take-level refinement of S045, not tested to a gate here.
What happens next
No Stage 2, no vault touch, no forward slot: a setup decisively negative in-sample does not advance. The honest lead is the 10%-profit-take-held-to-settlement tail-lighter refinement of S045 — a base-trade parameter change that, if pursued, becomes its own new identifier, not this one. The intake net: S045 (the fly structure) promoted, S047 (the scalp management) failed — one of each, both in the open.
For the specialist — methodology details (click to expand)
- Base: frozen S045 ATM fly (parity-centered short call + short put, 0.4% wing, 09:35 entry, 1 contract); only the exit management is under test.
- Exits: tight profit-take {5%, 10%} (25% = S045 reference); ±expected move (EM) stop proxied at 90% of per-session max-loss; 10:35 time-exit; tested separately + combined vs hold-to-settle. Forced exits pay a full §3a second cross; the profit-take fills at ≈ mid (patient limit).
- Gate: Stage-1 in-sample required positive tail-aware expectancy net of §3a and Deflated Sharpe > 0 after pillar deflation — not met (DSR ≈ 0.00 every cell).
- Data: 444 parity-centered 09:35 fly sessions, TRAIN, §3a spread.
FAILED at Stage 1, recorded openly (CLAUDE.md: "failed" is the pipeline term; the fail record is the marketing surface). Pre-registered at the owner's request 2026-07-12, run the same day. Sibling of S045; seeded from the Doc Severson / Theta Profits intake.