1H CRT

Candle Range Theory on the 1H timeframe: hour sweeps prior 1H extreme, closes back in body, next move targets opposite liquidity. Recalibrated against true-CRT hit rate. The GOLDEN tier (PDF Soup-Room filter) layers a key-level proximity gate on top — hr=9 SHORT lifts to 78%, hr=2 SHORT to 65%.

The setup

Candle Range Theory (CRT) on the 1H: a 1H candle sweeps the prior 1H high or low, then closes back inside the prior candle’s body. The read: the sweep was a liquidity raid; the next move targets the opposite extreme.

Line ships on our chart HUD as 1H CRT LONG / 1H CRT SHORT (or 1H CRT PRIME on the top hours).

Entry rules

• Trigger: 1H candle closes; wick beyond prior 1H H or L; close back inside prior body
• Direction: opposite the wick (fade the raid)
• Target: opposite prior-1H extreme
• Stop (implicit): sweep-wick extreme
• TTL: 180 minutes

Viable hours (n ≥ 30, hit rate ≥ 52%):

All other hours: both directions below viability. Does not fire.

Depth adjustment (sweep-wick depth past prior extreme):

• LONG: <3pt −5% · 10–20pt +3% (peak hit rate)
• SHORT: <3pt −5% · ≥20pt +6% (deep sweeps 20–40pt hit 55%, 40pt+ hit 66%)

The recalibration story

This model used to fire 1H CRT PRIME 91% at hr=6–8 AM, which felt great. Then we built the CRT-reversal backtest and measured the true-CRT hit rate (price touches opposite extreme before sweep-wick stop, within 180 min). The gap was brutal:

The old probabilities came from “close-same-direction within 120m” drift — which is not the question PM’s target field asks. We fixed the metric and rebuilt the table against n=5,160 signals over 5 years.

PRIME combo

isPrime = (hr==18 && SHORT) || (hr==09) || (hr==07 && LONG)

Only these combos earn the PRIME badge. 18 SHORT is the strongest single combo in the dataset.

2026 regime watch

• hr=07 LONG dropped 64% → 29% in 2026 (n=7 — small, but worth watching)
• hr=09 SHORT stable at 55% in 2026 ✓
• hr=18 SHORT still 100% hit (n=2) ✓

If hr=07 LONG’s 2026 slump persists past n=30 we’ll drop it from PRIME.

Cross-contract stability

Why Tier 2 overall (Tier 1 for PRIME combos)

• Many hours below viability — gated hard.
• PRIME combos (07 LONG, 09 both, 18 SHORT) earn Tier 1 treatment individually.
• Rest are solid but not flagship. The model as a whole is Tier 2.

GOLDEN tier — key-level confluence (Soup-Room CRT)

A community framework called “Soup Room CRT” (PDF distillation of an ICT-style entry model) argues that a CRT sweep that occurs while testing a same-direction key level is qualitatively different from one in open space. We tested the claim against our 5,169-signal MNQ dataset.

Method

For every CRT signal, compute the distance from the swept level (the prior-1H extreme that got raided) to the nearest same-direction named level:

• SHORT: nearest of PDH, PWH
• LONG: nearest of PDL, PWL

Bucket signals by distance and compare hit rate.

Headline result

The PDF’s filter does not generalize across all 24 hours — pooled lift is ~3pp, noise. But within the PDF’s named hours, it lights up:

Per-hour, gated to <25pt

The bucket cliff is sharp at 25pt — beyond that, the lift collapses:

Below 25pt: ~75%. Above: ~62%. That’s the line.

What didn’t work

• Asia window (hr 18–21) — the PDF’s Asia model goes negative both directions (longs −10pp gated, shorts flat). The hr=18 SHORT PRIME combo we already trade is not an Asia-model sweep; it’s the daily-close hour on a different mechanism.
• hr=4 and hr=8 — both go more wrong when gated.
• PWL / PMH / PML — too few near-misses to register; PDH/PDL dominate.
• All-hours pooling — drowns the signal in 10–17 ET garbage hours we don’t trade anyway.

What ships as GOLDEN

isGolden = ((SHORT and hr in [2, 5, 9]) or (LONG and hr in [7, 9]))
        and swept_level within 25pt of [PDH, PWH] (S) or [PDL, PWL] (L)

When GOLDEN qualifies, the model’s hour-table probability gets +10pp lift, label promotes from 1H CRT / 1H CRT PRIME to 1H CRT GOLDEN, confidence flips to HIGH, and the signal is treated as Tier 1 in the consensus scorer (same regime/penalty calibration as PRIME).

Caveats

• Sample sizes for individual GOLDEN combos are modest (hr=5 SHORT n=27, hr=2 SHORT n=54). The +10 prob lift is the empirical mean of the per-hour deltas — it’s not Brier-tuned.
• Only PDH/PWH/PDL/PWL counted toward proximity. Adding 4H OB midpoints, ASIA H/L, ORB H/L, and 4H FVG edges (the PDF’s actual “key level” definition) would likely tighten the filter further. Open follow-up.
• Because gating cuts the sample roughly 5× in PRIME hours, GOLDEN fires are rare — expect ≈10–15 per year on MNQ. The PRIME tier still carries most of the trade volume.

History

• 2026-04-19 — Original ship with old probability table.
• 2026-04-23 — Full recalibration against true-CRT hit rate. Overconfidence gaps of 27–54pt closed. Missing hr=18 SHORT edge added. New PRIME combo gate.
• 2026-04-29 — GOLDEN tier added (PDF “Soup-Room CRT” key-level filter). Backtest backtest_crt_keylevel_overlay.py n=5,169 — gating lifts hr=9 SHORT to 78% and hr=2 SHORT to 65%. Wired into PredictionModel.cs Model 12, track_predictions.py, smc_analysis.py setup override, and TIER1 sets.