Candle Range Theory (CRT)

A single candle sweeps the prior candle's extreme, then closes back inside. Read: the sweep was a liquidity raid. Next move targets the opposite extreme.

The pattern

Candle Range Theory is an ICT primitive. On any timeframe, you’re looking for three things in sequence:

1. A candle forms — call it C1 — with a high and a low.
2. The next candle (C2) wicks beyond C1’s extreme — either above C1 high or below C1 low.
3. C2 closes back inside C1’s body — the wick was rejected; the breach was a sweep, not a breakout.

When that happens, the read is: the wick raided stops beyond C1’s extreme, inventory was taken, and the move that follows aims for the opposite C1 extreme as its magnet target.

Why the pattern works

Stops tend to cluster just past recent highs and lows. A wick that reaches through those stops and then fails to close past them is a classic signature of a liquidity-seeking move by larger participants — sometimes called a liquidity raid or sweep of lows/highs. Once the raid has filled inventory, the path of least resistance is back through the range toward the opposite extreme, where the opposing pool of stops lives.

This is the same logic underlying AMD (Power of Three) at the session level. CRT is its candle-level cousin.

Where we use it

On this site CRT shows up in the 1H CRT model, which scans every 1-hour candle for the pattern and fires a line on our chart HUD targeting the opposite 1H extreme. It’s a 2 Tier model overall with specific Tier 1 “PRIME” hours (07 LONG, 09 both, 18 SHORT) where the hit rate is 62–82%.

Why not every CRT is tradeable

We calibrated CRT on 5,160 signals over 5 years. Most hours don’t work. Below ~50% hit the signal is a coin flip or anti-predictive. Our model only fires in the hours where the measured hit rate is ≥52% with n ≥ 30.

This is the general pattern you’ll see across the site: an ICT concept is real, but naively trading every instance of it leaks edge. The work is in figuring out when the pattern has statistical backing and when it’s a campfire story.

The depth filter

A wick of 1pt past C1’s extreme is a different animal from a wick of 40pt past. Our calibration found:

• LONG sweeps: <3pt wick = −5% probability. 10–20pt wick = +3% (peak hit rate).
• SHORT sweeps: <3pt wick = −5%. ≥20pt = +6%. 40pt+ = 66% hit.

Deeper sweeps carry higher conviction because they’ve done more liquidity work. Shallow sweeps are often just noise at the extreme.

Related

• AMD / Power of Three — session-level version of the same raid-and-reverse dynamic
• Killzones — the time-of-day windows where CRT and AMD tend to resolve
• 1H CRT model — our implementation with calibrated hour-by-hour probabilities