HOME/DAILY RESEARCH
EDGE MARKET INTELLIGENCE

Research

Honest futures research for NQ, ES, and GC. Every edge here was tested on 3 years of real 1-minute data with an out-of-sample split and modeled costs — and we publish the setups that failed, not just the ones that worked.

HOW WE TEST · OUT-OF-SAMPLE OR IT DOESN'T COUNT
DATA3y real 1-minDatabento CME, 2023–26
VALIDATIONIn/out-of-samplepick on 67%, prove on 33%
COSTSSlippage modelednet, not gross
GUARDSCoverage checkssparse data can't fake a win
LEAD #2 · THE 1990 “STRETCH” BREAK · BETAPOSITIVE EVERY YEAR · ZERO TUNED PARAMETERS

Measure the 10-day average 'stretch' (how far the open sits from the nearest extreme). Place a buy stop that far above the open and a sell stop that far below it. The first side hit takes the trade, the opposite level is the stop, and the position is held to the close.

NQ+0.114RLEAD
44% win · PF 1.23 · n=719 · maxDD 14.5R · 57% months +per-year +0.04 / +0.13 / +0.13 / +0.13Positive every year; still +0.088R at 4pt round-trip slippage (median risk ≈100pt).
ES+0.062RMARGINAL
44% win · PF 1.13 · n=711 · maxDD 18.5R · 54% months +per-year +0.16 / +0.01 / +0.04 / +0.12Positive every year but crosses to negative at ~2pt slippage — cost-fragile (risk ≈21pt).
GC-0.070RREJECTED
40% win · PF 0.87 · n=719 · maxDD 55.1R · 27% months +per-year −0.20 / −0.03 / −0.04 / −0.08Negative every year. Gold does not do open-continuation — published as a failure.

⚠ SIZING — the number nobody publishes: A positive edge can still fail a prop evaluation. At $250 risk/trade the 14.5R drawdown is ≈$3,625 — it BREACHES a $2,000 trailing-drawdown account. Tradeable size is ≈$100–125/trade, or a filtered subset that trades less often.

BETA — one untuned parameterization from a 1990 book, positive every year on NQ and robust to 4× our normal slippage, but it has not yet been proven forward in the open. We publish it as a lead, not a validated edge, and the forward ledger below decides which it becomes. 3-year Databento 1-min backtest · executable fills · net of slippage (2023–2026).

WHAT WE TESTED — AND WHAT FAILED
Gap fill — base rate (NQ / ES)~86% fill at some point; ~56% / ≈breakeven as a TRADE
Prior-day H/L break + cross-asset confirm (NQ)WITHDRAWN 2026-06 — fill artifact; −0.19R executable
Overnight-range break, continuation (NQ / ES)WITHDRAWN 2026-06 — ≈breakeven at realistic slippage
Turtle Soup — fade prior-day sweep−0.08R (NQ) / −0.12R (ES)
ICT Judas Swing — fade London sweep−0.34R out-of-sample (NQ)
AM Silver Bullet — fade 9am sweepbreakeven-to-negative
CISD / “order block” as a mechanical entry−0.28R across ~72,000 signals
TTrades “Fractal Model” — HTF sweep + close-back reversalall 44 variants negative OOS (−0.06…−0.32R)
Naive opening-range / initial-balance break (1:1)≈ breakeven once fills are modeled
“ORB 1:1” — fade the side that formed firstthe 87% sweep stat is REAL; the trade is −0.30R (NQ) / −0.48R (ES)
Volume-confluence ORB breakout (the 80–90% videos)great in-sample, dies out-of-sample; thresholds disagree across instruments
Hourly mean-reversion “magic hours” (83% claim)touch rate replicates — but the do-nothing baseline is HIGHER (92%)
RTH-gap 70.5% break → full fillfilter is genuinely predictive (+22pp) yet the trade still loses on geometry
Fair-value-gap (FVG) retest continuation−0.41R (ES) / −0.46R (NQ), negative every year
2×200EMA + VWAP pullback scalppositive only in the 2026 regime; flat 2023–25 at honest fills
Crabel “stretch” break (1990) — open ± 10-day stretch, hold to close+0.114R/trade NQ · 44% win · PF 1.23 · positive every year

Browse the full research library — 44 strategies with their out-of-sample and per-year columns →
The pattern: on index futures the sweep is a continuation signal, not a reversal — every "fade the sweep" setup loses out-of-sample, and continuation only pays at major once-a-day levels. When our OWN numbers fail a re-test, we withdraw them publicly (⚠) — same standard we apply to everyone else. The one live lead (below) is in an open forward test. See the gap-fill breakdown →

THE SESSION BIAS MAP · PRELIMINARY (BETA)

Combine four factors knowable by 09:45 — overnight direction, gap, prior-day body, ORB15. When all four agree (~1 in 3 days), the day closed that way ~68% of the time and the continuation trade made +1.29R (NQ). Below that, direction is a coin-flip — so the read is STAND DOWN. Honest caveat: this is preliminary — a small out-of-sample sample (n≈60, ±~12pp), a conviction/stand-down filter, not yet a graduated edge. A directional bias, not a win rate. The Session Bias Map indicator →

LIVE FORWARD TEST · CROSS-ASSET CONFLUENCE (AND-GATE) · BETAACCUMULATING
FWD TRADES3since 2026-06-16
AVG R-0.06Rvs +0.62R backtest
WIN %33%vs 65% backtest
PROFIT FACTOR0.65vs 3.62 backtest

Our single out-of-sample-validated lead — NQ prior-day-H/L break that diverges from ES, confirmed by a NQ/ES/YM initial-balance triple — traded live, in the open, since 2026-06-16. Backtest: +0.62R/trade, 65% win, PF 3.62, max-DD 2.4R, ~14 signals/yr. Confirmation bar: ~20–30 forward trades holding ≥ +0.30R before we call it validated. If it fades here, we publish that too.

DATEDIRENTRYSTOPTARGETRESULT
2026-08-07LONG29,686.329,575.030,020.0-0.34R
2026-07-30LONG28,099.527,904.028,686.0+0.32R
2026-07-21LONG29,192.529,086.329,511.1-0.16R
GOLD REJECTION WICK · BETA · PUBLIC FORWARD TEST
since 2026-09-09
LIVE $/TRADEper MGC
BACKTEST+$19n=85
VS BACKTEST
LIVE WIN %backtest 56.5%
TRADES TAKEN0of 0 signals · 0 no-fill
PROMOTION AT~30trades above +$10 net
THE RULE (frozen 2026-09-08 · not tuned since)

GC (traded as MGC, $10/pt) · 5-minute · 08:20–13:30 ET (14:20–19:30 SAST). Reference: Overnight range 18:00 → 08:20 ET. 1. First bar to trade beyond the overnight range AND close back inside on the same bar, closing against the trade direction. 2. A later bar closes beyond that rejection bar's far extreme (within 12 bars). 3. LIMIT at the rejection bar's midpoint, working from the confirmation bar (no fill within 12 bars = no trade). Low swept → LONG · high swept → SHORT. 30-point target, 30-point stop, Flat at 13:30 ET.

Why this is a candidate, not an edge. We examined roughly 350 combinations of level, entry model, target and stop to find it — with that many cells a good-looking backtest is expected from noise alone, so the backtest is not the evidence; this ledger is. Two things we corrected while building it, both published: (1)we first described this as a “breaker block”, but our code never implemented one — every qualifying day was a same-bar sweep-and-reclaim, so it is named for what it is; (2) an earlier version entered at the reclaim close, which used the confirmation to select the day and then entered before it — lookahead. That version is void. Entering at the confirmation bar at market loses (−$26/trade); only the limit retest survives. Expect ~28 trades a year: roughly $532 a year on a single MGC contract. Full record →

No signals yet. The rule fires roughly 28 times a year, so the first entries may take a couple of weeks. Every one will appear here — including the losers and the days the limit never filled.

Educational research, not trading advice. $ figures are one MGC contract ($10/point) before commissions (~$2 round turn, i.e. ~10% of the backtest expectancy). Backtest max drawdown $803 on one MGC.
INSTRUMENTS
NQ · ES · GC
PUBLISH TIME
7:30 AM ET
FREQUENCY
Daily
DELIVERY
Email + Archive
EDGE INTELLIGENCE · DAILY BRIEFING

Get the EDGE before
the market opens.

Daily futures intelligence for NQ, ES, and GC traders. What institutions analyze before the open — delivered to your inbox every morning.

No spam. No hype. Tested market context with sample sizes, daily. Unsubscribe anytime.

DAILY COVERAGE

  • Daily gap fill probability
  • Regime classification
  • Volatility context
  • Opening range expectation
  • Session replay breakdowns
  • Macro event setups
  • Historical analog analysis
  • EDGE score daily read
DAILY BRIEFING ARCHIVENEWSLETTER ARCHIVE PENDING

The daily pre-open briefing publishes via the newsletter; the searchable archive lands here once the Beehiiv integration is wired. Until then, the live read is on the NQ dashboard, and the validated research above is the real, out-of-sample work. We'd rather show you nothing than invent sample articles.

Every stat above is computed from real historical price data with an out-of-sample split — and we publish the setups that fail, not just the ones that work. Market context, not trading advice.