The Trading Pattern With a 97% Historical Win Rate That Nobody Talks About

Jay Sharma · Founder, Botsfolio

Published August 18, 2026

Any headline with a "97% win rate" is usually clickbait. This one is not, but you should verify it yourself. All the numbers below come from our own backtest engine, which reruns walk-forward on real Coinbase candles going back to 2018. The methodology page lays out exactly how we compute it.

The pattern is the order block after a liquidity sweep, and on the 1H timeframe it has posted 85-97% historical win rate across BTC, ETH, and SOL depending on management style. Very few traders talk about it, because it does not have a name that fits neatly into a course syllabus.

Here is what it is, why it works, and where it fails.

The two patterns you already know

Every SMC trader has heard of order blocks. And every one of them has heard of liquidity sweeps.

An order block is the last opposite-color candle before a strong impulse. It marks the price zone where institutional flow entered the market. On its own, order blocks are a decent setup. Our backtest shows a raw order block on BTC 4H sits around 60-65% win rate depending on how you manage it.

A liquidity sweep is when price briefly extends beyond a visible high or low (grabbing stops parked there), then reverses. It is the market telling you the move was a stop hunt, not a real breakout. On its own, sweeps also sit around 60-70% win rate, again depending on management.

Neither one, standalone, is dramatic. Neither one earns a 97% headline. But when they occur together, in the right order, on the right timeframe, something different happens.

The confluence

The pattern: a liquidity sweep occurs, and then, immediately after the sweep candle closes, an order block forms in the reversal direction.

In other words, institutional flow ran the stops, and then the very next candle (or two) shows large-body accumulation in the opposite direction. That accumulation candle IS the order block. The sweep and the order block are two footprints of the same event: a large trader used the stop-hunt to fill their opposite-direction position.

When you see this combination on a 1H chart, you are seeing the entry footprint of a size-mover. Not a trader guessing. A participant that just did the work.

Why the structure logic makes sense

Standalone order blocks fail because you cannot always tell WHY the accumulation happened. It could be a genuine trend initiation. It could be a temporary bounce that fades. It could be noise.

Standalone sweeps fail because you cannot always tell what the smart money did with the liquidity it grabbed. Sometimes it uses the stops as an exit (bad for you as a reversal trader). Sometimes it uses them as an entry (good).

The confluence resolves both ambiguities. The sweep tells you liquidity was collected. The order block tells you what side that liquidity was collected FOR. Together they answer the only question that matters: "who is on the other side and which way are they positioned?"

That is why the win rate jumps from 60-65% to 85-97% depending on coin and timeframe. You are not trading a pattern. You are trading a confirmed footprint. If you want to see whether this footprint is on the chart right now, ask the Analyst directly.

The real backtest data

Here is what the confluence has actually produced across the coins we track. Full matrix, computed net of 0.12% round-trip fees.

Two important notes on how to read this:

1. Win rate is not the whole story. A 90% win rate at 1R target means expectancy is +0.8R after losses. A 60% win rate at 2R target is +0.6R. Both are edge. The confluence gives you the higher win rate with the same 1R target, which is what makes it unusual.

2. Sample size matters. Rows with under 20 historical instances should be treated as directional, not conclusive. The 1H and 4H rows on the majors all have 100+ instances.

When the 3-15% failure hits

No pattern is 100%. Here is what the failure looks like when it happens.

Failure mode 1: the sweep is not a sweep, it is a breakout. Sometimes what looks like a stop-hunt is actually the start of a real trend continuation. The order block forms, price bounces briefly, then the underlying trend takes over and takes out the order block from the other side. This is the most common failure and it happens roughly 5% of the time even on the strongest instances.

Failure mode 2: news bomb. A macro headline hits mid-setup (Fed statement, exchange outage, geopolitical event). Structure breaks because a non-structural force is now driving price. No pattern survives news, and the confluence is no exception.

Failure mode 3: multi-timeframe conflict. The confluence forms on 1H against a strong bearish 1D structure. The 1H setup takes but is quickly overwhelmed by the higher timeframe. This is why serious traders check higher timeframe bias before taking any pattern, confluence included. See our multi-timeframe analysis primer for how to do the check.

What "forming" looks like on the chart

The pattern is only tradeable if you can see it forming before the entry candle closes. Here is the sequence:

1. Setup building: price approaches a visible high or low with reasonable liquidity above or below (retail stops parked there).
2. The sweep candle: price extends beyond that level, then closes back inside the prior range. This is bar 1 of the pattern.
3. The order block candle: the very next candle prints an opposite-color body of at least average size. This IS the order block. This is bar 2 of the pattern.
4. Entry trigger: some traders enter on the close of bar 2. Others wait for a small pullback into the order block zone before entering, sacrificing a bit of edge for confirmation.
5. Stop: below the sweep low (for longs) or above the sweep high (for shorts). This is the level that, if broken, invalidates the entire structural read.
6. Target: 1R at minimum. Many traders take partial profits at 1R and let the rest run to structure.

You do not need to draw this by hand. Our detector flags the confluence in the live setups feed with the sweep, the order block, the entry, and the invalidation all marked. But you should understand the anatomy either way, so you can decide when to trust the machine's read and when to skip it.

FAQ

Educational analysis, not financial advice. Past performance does not predict future results.