Order blocks: we tested 2,142 on Bitcoin. Random rectangles held better.
We detected 2,142 order blocks across 826 days of Bitcoin data and measured retest rates, hold rates, and whether the zones beat random rectangles.
Order blocks are the load-bearing concept of smart money trading: the last opposite candle before a big move, marked as the footprint of institutional positioning, expected to hold when price comes back. The pitch is everywhere. The base rates are nowhere.
We measured them the same way we measured fair value gaps: fixed rules, full sample, and a control group, on 826 days of Bitcoin across three timeframes.
What an order block is supposed to be
The story goes like this. Institutions cannot buy size at one price, so they accumulate inside a candle that looks like ordinary selling. Then they drive price away with displacement. The original candle, the bullish order block, marks where their resting interest sits. When price trades back into it (the "mitigation"), that interest defends the zone and price bounces. A bearish order block is the mirror. A breaker block is the follow-up claim: when a zone breaks, it flips polarity, old demand becomes new resistance.
Notice the shape of the story. It explains a rectangle on a chart with an invisible actor's intentions. Whether an actual passive buyer sat in that zone is checkable, but not from candles. It shows up in executed flow: absorption on the footprint, a CVD that refuses to fall while price tests the level. The rectangle is a guess about flow. The tape is the flow.
We will come back to that after the numbers, because we have run this play before: the fake walls story had the same structure of invisible intent, and the data treated it the same way.
The test
Rules fixed before running, no discretion anywhere:
- Data: Coinbase BTC-USD hourly candles, 2024-05-01 to 2026-08-05 UTC, 19,826 candles (826 days), resampled to 4h and 1D. Same dataset as the fair value gap study.
- Detection (bullish): a down-close candle whose high is broken by a candle close within the next 3 bars, with the leg from the block's low to that close spanning at least 2x ATR(14). Zone = the candle's full range. The block is indexed at the close of the confirming candle, so nothing is drawn with hindsight. Mirror for bearish.
- Counts: 1,626 zones on 1h (817 bullish, 809 bearish, median height 0.60% of price), 456 on 4h (median 1.19%), 60 on 1D (median 3.28%). Median displacement 2.5x ATR. 2,142 zones total.
- Hold test: from the first retest of the zone edge, price races one full zone height in the claimed direction against a candle close through the far side. Win the race and the zone "held."
- Control: for every real zone, a phantom rectangle with the same height, planted at the same distance from price, at a uniformly random timestamp, put through the identical race.
Mitigation is nearly guaranteed, which is the problem
| Retest of the zone within | 1h blocks | 4h blocks | 1D blocks |
|---|---|---|---|
| 3 days | 80.8% | 66.5% | 33.3% |
| 7 days | 87.6% | 78.4% | 43.3% |
| 30 days | 93.6% | 88.8% | 68.3% |
| Ever | 97.7% | 95.8% | 85.0% |
Median time to retest on the hourly chart: 6 hours. When a mentor says "price always comes back to mitigate the order block," this table is why the claim feels prophetic.
It is the same near-certainty we found with gap fills. Bitcoin revisits nearby prices constantly, so any zone you draw near price will be "respected" with a retest on a short clock. A prediction that comes true 98% of the time regardless of the theory behind it is not evidence for the theory.
The hold test, and the control that deletes it
The tradable claim is what happens at the retest. Here it is, with the phantom rectangles alongside:
| Timeframe | Real order blocks | Random same-size rectangles | Edge | n (real / control) |
|---|---|---|---|---|
| 1h | 61.2% | 62.3% | −1.1pp | 1,589 / 1,584 |
| 4h | 57.0% | 61.2% | −4.3pp | 437 / 436 |
| 1D | 64.7% | 71.2% | −6.4pp | 51 / 52 |
The edge of institutional footprints over rectangles nobody drew is negative on every timeframe. Not small. Negative. Zones chosen by the theory did slightly worse than zones chosen by a random number generator, and the daily row has only 51 decided cases, so read that one as descriptive.
Where the 60% actually comes from
Both columns sitting near 60% instead of 50% deserves its own section, because it is where the marketing lives.
The race is asymmetric. The win condition is a wick touching the target. The loss condition is a full candle close through the stop. Wicks are cheap and closes are expensive, so any zone, real or phantom, wins more than half of these races.
That is an explanation, so we measured it. Same zones, same retests, same target distance, with one change: the target now also requires a close.
| Timeframe | Order blocks | Random rectangles |
|---|---|---|
| 1h | 51.2% | 53.2% |
| 4h | 49.1% | 48.0% |
| 1D | 47.1% | 61.5% |
Ten points of the hourly hit rate evaporate, and what is left is a coin flip. Nothing about the zones changed. Only the scoring did.
This is the mechanism behind every order block win rate you have been shown without a control column. Run a backtest with a wick-paid target and a close-charged stop, skip the control group, and you can print 60% for essentially any rectangle-drawing method. Our control column is what that 60% looks like with the costume off.
Direction: the zones point the wrong way
Forward returns from the moment of first retest, hourly chart:
| After retest of | Median 24h fwd | Median 72h fwd | % positive (24h) | n |
|---|---|---|---|---|
| Bullish OB (claim: up) | −0.03% | +0.02% | 49.4% | 807 |
| Bearish OB (claim: down) | +0.12% | +0.35% | 52.8% | 781 |
| Baseline, any hour | +0.05% | +0.12% | 51.1% | 2,829 |
Read the bearish row again. Price entering a supposed institutional sell zone went on to outperform the market's baseline drift by roughly three times over the next three days. Bullish zones, where the bounce is supposed to be, underperformed the baseline at 24 hours and went nowhere at 72. The 4h chart repeats the pattern: bullish retests median −0.17% over 24h against a +0.15% baseline.
This is worse than "no edge." A signal that reliably pointed the wrong way would at least be useful inverted. This one tracks noise with a story attached.
The rescues do not rescue
Every order block tutorial ends with qualifiers: only trade blocks with strong displacement, only with imbalance confluence, only breakers. We cut the sample along each one.
Displacement strength. Blocks launched by a 2-3x ATR leg held 61.1% on the hourly; blocks with 3x+ legs, 61.5%. Nothing. On 4h the strong-leg blocks held 50.5% against 58.8% for the modest ones, so harder displacement made the zone worse. The story says more conviction. The data says a market that just moved 3 ATRs is volatile, and volatile tape closes through stops.
FVG confluence. 80.2% of hourly order blocks already contain a fair value gap in their displacement leg, which tells you how much these two concepts overlap. Blocks with the gap held 62.4%, without it 56.5%. The six-point gain looks like something until you put it next to the control: random rectangles score 62.3%. Confluence buys you exactly the number a phantom gets for free.
Breaker blocks. When a bullish zone failed, price travelled a full zone height away, and then returned, the breaker story says old demand acts as resistance. It did, 58.8% of the time on 1h (n = 762). The bearish mirror: 59.5%. On 4h, 57.4% and 58.9%. Those are not coin flips, and they are not an edge either. They are the all-rectangles number again. The flip inherits the scoring asymmetry exactly like the original zone did, and contributes nothing of its own.
What was actually in the candle
Here is the part worth keeping. The order block story is a claim about passive positioning: someone absorbed the selling inside that candle before the markup. That is a real phenomenon, and it is directly observable, just not in the candle's shape.
Absorption prints on the footprint as heavy volume at prices that refuse to break. It shows up as price flat while CVD digs in, the signature we walked through in order flow trading. On the current tape you can watch it happen in the footprint view whenever a level gets defended in real time.
The difference matters because candles cannot distinguish a defended level from a vacated one. Two identical down candles before two identical rallies: one absorbed real size and may get defended on a retest, the other was a liquidity hole that filled itself, as in the stop hunt study. Same rectangle, opposite flow. Averaged over 2,142 rectangles, the informative ones drown, which is exactly what our numbers look like: a mean of nothing.
If you currently trade order blocks with a footprint or delta filter and it works, our result does not contradict you. It says the rectangle contributed nothing. Your filter was the strategy all along.
What this doesn't tell you
One venue, spot prices, one asset, 826 days, one mechanical definition per timeframe. ICT practitioners layer context we did not model: higher-timeframe bias, session windows, liquidity sweeps preceding the block, market structure shifts.
Fine. Those filters can only matter if the base object carries signal for them to refine, and the base rates above are the floor any filtered version has to demonstrably beat, with a control, before the win rate means anything. We published the floor.
The daily timeframe carries 60 zones and 51 decided races; treat those rows as descriptive, not conclusive. The hold race also resolves intrabar ties in the target's favour, which lifts every column in the table, real and phantom alike, and is precisely why the comparison is the number to read rather than the level.
The live-tape side of every claim here is queryable by agents over MCP, so the flow-versus-rectangle comparison can be re-run against today's market rather than taken on faith.
Frequently asked questions
What is an order block in trading?
The last candle that closed against a strong move before the move launched: a down candle before a rally (bullish order block) or an up candle before a decline (bearish). Smart money theory treats its range as a zone of institutional interest that price should respect on a return visit.
Do order blocks actually work?
In 826 days of Bitcoin data, retested order blocks won a one-zone-height race 61.2% of the time on the hourly chart, and height-matched random rectangles won the same race 62.3% of the time. The random rectangles were ahead on every timeframe we tested. Forward returns after retests pointed marginally the wrong way on both sides. Any working order block strategy is getting its results from filters around the zone, not from the zone.
Why do order block backtests show a 60% win rate?
Because of how the race is scored, not because of the zone. The target is paid when a wick touches it, while the stop is only charged when a full candle closes through it. Wicks are cheap and closes are expensive, so any rectangle wins more than half of these races. When we required a close on both sides, hourly order blocks fell from 61.2% to 51.2% and the random rectangles fell from 62.3% to 53.2%. Roughly ten points of a typical order block win rate is bought by the scoring asymmetry.
What is a breaker block?
A broken order block that has supposedly flipped polarity, old support acting as new resistance. After a confirmed break, flipped bullish zones held as claimed 58.8% of the time on the hourly chart and flipped bearish zones 59.5%, which is the same rate every other rectangle scores in this race, including the random controls. The flip does not add anything; it inherits the scoring asymmetry like everything else.
Are order blocks the same as supply and demand zones?
Functionally yes. Supply and demand zone trading draws the same rectangles with an older vocabulary, and every mechanical test here applies to it unchanged: near-certain retests, hold rates matched by random rectangles, and direction that tracks market drift rather than the zone's label.
Dataset. Source: Coinbase Exchange BTC-USD, hourly OHLCV, 2024-05-01T00:00Z to 2026-08-05T11:00Z (19,826 candles, 826 days), resampled to 4h and 1D. Detection: opposite-close candle, confirming close through its extreme within 3 bars, displacement ≥ 2x ATR(14) from the block's far edge to that close; zone = full candle range; confirmation-close indexing (no hindsight drawing). Counts: 1,626 (1h, 817 bull / 809 bear), 456 (4h, 214/242), 60 (1D, 28/32). Hold test: one-zone-height target on touch against a candle close through the far side, evaluated from the bar after first retest, intrabar ties to the target. Symmetric variant requires a close on both sides. Control: height- and distance-matched random rectangles, uniform random timestamps, fixed seed. Baseline returns: 24h and 72h windows sampled every 7 bars. Breakers require a confirmed break (one zone height beyond the far side) before the return. Computed 2026-08-05.
MarketTrace shows aggregated funding, order-book imbalance, CVD, open interest and cross-exchange liquidations for BTC, ETH, SOL, BNB, XRP, DOGE and HYPE across Binance, Bybit, OKX and Hyperliquid (liquidations cover Binance, Bybit and OKX). Descriptive microstructure, not financial advice.