# Results — backtest of claimed "profitable prop trader" strategies

**Pre-registration:** `PREREG-2026-09-16.md` (sha256 `4e921928…cfe32`, frozen before any run). **Code:** `engine.mjs` (tape), `sim.mjs` (prop accounts). **Outputs:** `out/tape-summary.json`, `out/prop-summary.json`, `out/prop-size-sensitivity.json`, `out/trades-{NQ,ES}.json`.
**Data:** GLBX 1m NQ/ES `.v.0`, IS 2019-01-01→2023-12-31, OOS 2024-01-01→2026-09-11. Micros, $1.50 RT + 1 tick each way (news stop entries +4 ticks). 13 claimed strategies × 2 roots = 26 cells, each with a seeded coin-flip twin (same times, entries, stop/target geometry).

## Headline
1. **Tape: 0 of 26 cells VALIDATED.** Nothing clears t ≥ 1.2 net; five cells are decisively negative.
2. **Prop overlays (frozen H-P1): not supported.** The "rule-harvest" overlays do not raise payout odds on zero-edge streams except the micro→mini ladder on Apex (more size), and the ladder *lowers* them on Topstep (the consistency rule punishes big days).
3. **Prop business test (frozen H-P2): holds.** No stream — strategy or control — has EV per attempt > 0 in both halves at list fees under the pre-registered overlays; none even at the cheapest fees.
4. **Post-hoc size sensitivity (addendum, not a pre-registered verdict):** at 5–20 micros the simulator reproduces Topstep's published ~16.8% pass rate — and **strategies and coin-flip controls are indistinguishable** on pass rate, payout rate and EV. Of the 4 cells (of 520) with EV > 0 in both halves at list fees, **3 are coin-flip controls.** Where the prop game pays, it pays the account contract's option value, not the chart read.

## 1. Tape verdicts (1 micro, net of costs)
| arm | NQ verdict | NQ IS / OOS net $/trade | NQ full net t (gross t) | ES verdict | ES IS / OOS net | ES full net t (gross t) |
|---|---|---|---|---|---|---|
| ORB15-2X (15-min ORB, 2× range) | INCONCLUSIVE | +3.94 / +7.47 | 1.01 (1.30) | INCONCLUSIVE | +0.23 / +0.41 | 0.11 (0.68) |
| ORB15-RUN ("let it run") | REFUTED | +6.99 / −10.94 | 0.16 (0.50) | REFUTED | +0.54 / −1.68 | −0.10 (0.52) |
| IB-EXT | REFUTED | +7.63 / −0.25 | 0.98 (1.28) | REFUTED | −1.15 / +0.55 | −0.22 (0.37) |
| IB-FADE | REFUTED | −3.94 / +11.20 | 0.29 (0.65) | REFUTED | −7.42 / −3.35 | **−3.21** (−2.41) |
| RANGE-FADE | REFUTED | −4.71 / −3.01 | −1.53 (−0.97) | REFUTED | −3.03 / −7.77 | **−3.60** (−2.45) |
| ICT Silver Bullet NY AM 2R | INCONCLUSIVE | +0.85 / +3.59 | 0.53 (0.97) | REFUTED | −3.13 / −4.44 | −1.87 (−1.08) |
| ICT Silver Bullet NY AM 3R | INCONCLUSIVE | +1.83 / +4.80 | 0.73 (1.11) | REFUTED | −3.01 / −7.44 | −2.05 (−1.37) |
| ICT Silver Bullet London | REFUTED | −0.31 / −1.33 | −0.28 (0.37) | REFUTED | −4.65 / −5.39 | **−3.39** (−2.35) |
| VWAP ±2σ fade (claimed tight stop) | REFUTED | −1.83 / −3.25 | −2.48 (−0.87) | REFUTED | −4.95 / −5.19 | **−10.78** (−7.56) |
| VWAP trend pullback | INSUFFICIENT | +11.47 / +11.82 (n 145/98) | 1.17 (1.32) | INSUFFICIENT | −9.57 / −5.28 | −1.88 (−1.53) |
| EMA 9/21 | REFUTED | +5.87 / −4.69 | 0.60 (1.03) | REFUTED | −2.52 / −2.22 | −1.25 (−0.47) |
| News straddle (CPI/NFP/FOMC) | INSUFFICIENT | −35.21 / −43.62 | **−6.51** (−6.26) | INSUFFICIENT | −28.85 / −38.39 | **−9.44** (−9.01) |
| News post-spike fade | INSUFFICIENT | +4.65 / +26.77 (n 61/37) | 1.06 (1.18) | INSUFFICIENT | −4.18 / −0.98 | −0.46 (−0.22) |

Reading the INCONCLUSIVE cells honestly: NQ ORB15-2X nets +$5.18/trade but its coin-flip twin nets **+$4.00** (NQ rose through 2019–2026, so any geometry with a 2× target collects drift); NQ ICT NY AM beats its twin by ~$5 at t 0.5. Neither is distinguishable from its control. The news straddle loses even gross: it wins 17–21% at 2:1 when 33% is required; excluding the 11.6% of events where both stops fill inside one minute (booked worst-case) it is still about −$31/trade on NQ. The VWAP fade with the stop the blogs quote (12 NQ / 4 ES ticks beyond the tag) is stopped out 71–73% of the time — gross t −7.6 on ES, so it is not a commission artefact.

Outliers were checked by hand and are real sessions (2025-04-09 tariff pause, April 2025 crash, 2022-02-24, 2020-03-12), not data errors.

## 2. Prop-account simulator (pre-registered overlays)
Attempts start on every trading day of a half; censored attempts (unresolved at the half's end, 2–9%) are excluded.

**H-P1 — do the rule-harvest overlays raise P(payout per attempt) on zero-edge streams?** Mean over the 13 coin-flip controls:

| firm / half / root | O0 1 micro | O1 $25 risk | O2 ladder | O3 best-day cap | O4 2-loss/+$300 stop | O5 all | controls where O3/O4/O5 > O0 |
|---|---|---|---|---|---|---|---|
| Topstep IS NQ | 1.0% | 0.0% | 0.8% | 1.0% | 0.9% | 0.9% | 2/13 |
| Topstep OOS NQ | 2.2% | 0.0% | 1.4% | 2.2% | 2.0% | 1.4% | 3/13 |
| Topstep IS ES | 0.3% | 0.0% | 0.3% | 0.3% | 0.3% | 0.3% | 1/13 |
| Topstep OOS ES | 0.0% | 0.0% | 0.5% | 0.0% | 0.0% | 0.7% | 1/13 |
| Apex IS NQ | 0.1% | 0.0% | 0.9% | 0.1% | 0.1% | 0.9% | 3/13 |
| Apex OOS NQ | 0.1% | 0.0% | 2.2% | 0.1% | 0.1% | 2.1% | 2/13 |
| Apex IS ES | 0.0% | 0.0% | 0.3% | 0.0% | 0.0% | 0.3% | 1/13 |
| Apex OOS ES | 0.0% | 0.0% | 0.0% | 0.0% | 0.0% | 0.0% | 0/13 |

**Not supported.** The daily caps (O3, O4) change nothing measurable; sizing to $25 risk (O1) almost never permits a trade; only the ladder (O2) moves the needle, and only on Apex — it is more size, not more discipline.

**H-P2 — any stream with EV per attempt > 0 in both halves?** At list fees (Topstep + $149 activation; Apex $167/period + $140 activation): **none.** At the cheapest fees (Topstep $0 activation; Apex $35/period, $0 activation): **none.** At the catalog's 1-micro sizing a 50K account rarely reaches +$3,000 inside 60 days, so the fee is paid and the payout rarely arrives (EV ≈ −$140 Topstep, −$100 to −$500 Apex per attempt).

## 3. Addendum — post-hoc size sensitivity (fixed 1/3/5/10/20 micros; NOT a pre-registered verdict)
Mean over the 13 strategies vs their 13 coin-flip controls:

| firm | root / half | pass % at 5 micros strat / ctrl | pass % at 20 micros | payout per attempt at 5 micros | EV list fees at 20 micros |
|---|---|---|---|---|---|
| Topstep | NQ IS | 11.1 / 9.9 | 9.0 / 9.6 | 4.1 / 3.1% | −$50 / −$51 |
| Topstep | NQ OOS | 12.5 / 10.9 | 8.6 / 7.8 | 3.3 / 3.2% | −$43 / −$43 |
| Apex | NQ IS | 13.6 / 12.4 | 19.3 / 20.4 | 1.3 / 1.2% | −$221 / −$234 |
| Apex | NQ OOS | 14.5 / 15.2 | 22.0 / 21.6 | 0.7 / 1.0% | −$207 / −$200 |
| Apex | ES OOS | 9.2 / 10.5 | 13.2 / 16.2 | 0.5 / 1.9% | −$221 / −$235 |

- Pass rates at realistic size (9–22%) bracket Topstep's published 2025 figure of **16.8%**, so the simulator is calibrated where it matters.
- **Strategy and control columns are the same numbers.** Size decides who passes; the chart read does not.
- Of 520 cells (26 streams × 5 sizes × 2 firms × … both halves), **20** have EV > 0 in both halves at the cheapest fees and **4** at list fees — and **3 of those 4 are coin-flip controls** (NQ ORB-geometry random direction on Topstep at 5 and 10 micros; NQ ORB-RUN geometry at 5 micros). The only strategy among them is NQ ICT NY AM 3R at 20 micros on Topstep (+$99 IS / +$44 OOS per attempt), which its own tape verdict marks INCONCLUSIVE at t 0.73. That is the multiple-testing rate, not a discovery.
- Mechanism: an evaluation is a **barrier option** — a fixed fee buys capped downside (the trailing drawdown) and a volatile upside (target → payout). Size is volatility; volatility is option value. That is why "copy 20 accounts" and "ladder to minis" appear in every playbook: they buy more option, not more edge.

## Implementation notes (clarifications, not rule changes)
- **VWAP-FADE trend filter:** "first-hour range > 1.5× its 20-day mean" would look ahead for signals before 10:30. Signals 10:00–10:29 use the running range since 09:30; later signals use the completed first hour.
- **News straddle:** when both stop orders are touched inside one 1-minute bar, the trade is booked worst-case (filled, then stopped) — 11.6% of NQ events; excluding them leaves the verdict unchanged.
- **ICT Silver Bullet** is a discretionary pattern; this is one explicit, reproducible reading (sweep → 5-bar-fractal MSS within 20 min → 3-bar FVG → 50% entry). Other readings exist; the catalog's "77% at 1:2" is not reproduced by this one (NQ 36% win, ES 32%).
- **Firm rules** are approximations of 2026-09-16 public pages (trail type, payout gates, fees); Apex's intraday trail is approximated from each trade's MFE/MAE and their order. Funded accounts restart at $50,000 after a pass.

## What a prop trader can take from this
- None of the 13 most-sold intraday setups beats friction on NQ/ES micros, and the two that look positive on NQ are indistinguishable from random direction.
- Evaluation outcomes are driven by **size relative to the trailing drawdown**, not by the setup. The fee is the price of a lottery ticket whose odds you set with size.
- The discipline overlays (best-day caps, stop after two losses) protect the account but do not create payouts; they neither help nor hurt measurably.
- The one repeatable drift on this desk (overnight) is the one most firms prohibit.