Why do similar crypto charts have different outcomes?

A similar price shape does not mean everything else in the market was the same. In our test of 12 crypto markets from 2020 to September 2026, four to six of the ten closest past charts rose over the next 24 hours in 62.72% of 3,600 tests.

What counts as a close split among past cases?

In this study, a close split meant four to six rising outcomes among ten past matches. We checked each case after the same amount of time. An unchanged price did not count as rising.

Does disagreement mean the next price move will be small?

No. Counting rising cases tells you about direction, not the size of the moves. Check how far each price moved as well.

Does the Hyperliquid search use more than chart shapes?

Yes. It also checks buying versus selling, changes in the total size of open contracts, and forced trade closures. This study did not test whether those extra details reduce disagreement.

Similar crypto charts, different outcomes: how often?

Similar crypto charts often have different outcomes. In 3,600 tests across 12 coin-and-exchange markets, four to six of the ten closest past charts rose over the next 24 hours in 62.72% of tests. Each test used 48 hourly closing prices, at test times from January 2020 through September 29, 2026.

A split like that shows that similar shapes had different endings. It does not tell you which ending comes next.

For Bitcoin alone, see What happened to Bitcoin after similar charts?.

What does it mean when past crypto cases disagree?

A chart analog is a past stretch of prices that looks like the chart you are studying. A bar records the opening, highest, lowest and closing price during a set time. We used hourly bars and compared the changes between their closing prices.

At each test time, we found ten earlier matches for the same coin on the same exchange. A Binance Bitcoin search used only Binance Bitcoin history. We then checked whether the price rose after each past match.

We called the group closely split when four to six of the ten cases rose over the chosen period. An unchanged price did not count as rising. A group with three or seven rises still had disagreement, but we did not count it as a split.

Before comparing shapes, we put each chart's price changes on a common scale. This lets charts match even when their price levels and swing sizes differ. Our plain guide to chart analogs explains this step.

How often did similar crypto charts have different outcomes?

The table shows the share of test times with a close split. Each row has 300 tests, using 48 hourly closing prices per chart and ten earlier matches per test. All rows end at September 29, 2026, at 00:00 UTC; the exact first test times appear below. UTC is a shared world time standard.

Bitcoin is BTC, Ethereum is ETH, Solana is SOL, and Avalanche is AVAX. Binance rows use perpetual contracts, which follow a coin's price and have no end date. Coinbase rows use spot markets, where the coin itself is traded. USDT and USD name the currencies used to quote prices.

How often did similar crypto charts have different outcomes?
Coin and exchange	First test time, UTC	Split after 12 hours	Split after 24 hours	Split after 48 hours
BTCUSDT, Binance perpetual	January 1, 2020, 00:00	65.00%	65.00%	67.00%
ETHUSDT, Binance perpetual	January 18, 2020, 22:00	62.67%	64.33%	61.67%
SOLUSDT, Binance perpetual	October 10, 2020, 10:00	65.67%	59.67%	64.33%
AVAXUSDT, Binance perpetual	October 18, 2020, 04:00	73.00%	67.00%	69.00%
BTC-USD, Coinbase spot	January 1, 2020, 01:00	64.33%	64.33%	60.67%
ETH-USD, Coinbase spot	January 1, 2020, 04:00	61.00%	63.33%	62.33%

Across all twelve markets we tested (ten coins on Binance, plus BTC and ETH on Coinbase) and all three periods, close splits never fell below 59.67% of tests and reached 73.00% at most.

Is a split a sign that something went wrong?

No. It is the normal result. Think of flipping ten fair coins. Four to six of them land heads about two times in three. Our split shares sit close to that.

So ten past cases often split the way ten coin flips would. That does not prove the cases are random. But together with the direction test below, it suggests the chart shape alone told us little about which way price went next. Seeing that plainly is useful: it stops one good-looking match from standing in for the whole picture.

Does waiting longer make the cases agree?

The table gives no simple rule that waiting longer makes disagreement fade. For Binance Bitcoin, the share of close splits was higher at the longest period shown. For Coinbase Bitcoin, it was lower. Our Bitcoin look-alike results show both exchanges side by side.

Each column counts splits separately. It does not show whether the same searches stayed split throughout. A case can rise early and fall later, so agreement depends on when you check the price.

When you read a result, choose the time period that answers your question and keep it fixed. Picking whichever later time looks clearest hides how much the answer depends on that choice.

Bitcoin close splits by later period and exchange

Does a split mean price will stay flat?

No. The split counts directions. It says nothing about how far each case moved.

A group can contain large rises and large falls. An average near zero can hide both. Check the percentage changes as well as whether they were positive or negative.

The path matters too. A case that ends higher may have fallen along the way. Comparand's chart search, shape_analogs, shows each past case with its own price path and later moves, so you can see those differences.

These are past cases, not a forecast.

A habit you can use in your own research: when four to six of ten cases rose, write the result down as “split” instead of picking a side. Then note the largest rise and the largest fall among the cases. That pair tells you how wide the past outcomes were, which a vote count hides.

Could we combine the cases into a direction?

We tested that separately. We sorted the ten later percentage moves and found their median: the average of the middle pair. If that value was positive, the rule chose up. If it was negative, the rule chose down.

For the same 300 Binance BTCUSDT tests, this rule got the next 24-hour direction right 50.67% of the time. Always choosing down got 50.67% right at those same times. The median rule did no better than that simple comparison.

An imaginary trade following the median's direction had an average return of −0.30% over that same 24-hour period, after fees: the charges for opening and closing a trade. Each trade was measured separately. This was not the return of an account that kept reinvesting its money.

These results cover the whole Binance Bitcoin sample. They do not show how the rule did only when the cases were closely split. Our broader test of chart look-alikes explores the prediction question.

How we measured this

We used public Binance and Coinbase hourly prices, with 300 roughly evenly spaced test times per market over the dates in the table. We compared charts with 48 hourly closing prices, putting price changes on a common scale. We kept ten closest past stretches that shared no bars with each other. Each stretch ended before the chart being studied began, and its later results were already known at the test time. We counted close splits after 12, 24 and 48 hours. The trade test charged 0.05% to enter and 0.05% to exit; it left out some costs and required complete prices.

What are the limits of these results?

We chose test times without using the size or direction of later moves. But we required complete later prices for both the past cases and the chart being tested. A live search cannot know in advance whether those prices will be missing.

The trade test left out payments between traders with open contracts, borrowing costs, and gaps between quoted and actual trade prices.

The study includes markets that were still trading. Some tests reused the same past outcomes, and later price paths could overlap. That means the tests were not all separate pieces of evidence. We cannot use these counts alone to claim a reliable trading advantage.

The split measure counts cases equally, whatever their move size. It does not explain why the cases differed, or prove that a group with more agreement gives a better answer.

Try this in Comparand

With Comparand connected to your AI app, ask: “Find ten past Binance Bitcoin charts like the last 48 completed hourly bars of BTCUSDT. Show each case's path and later moves.” This uses shape_analogs.

You get:

Matches from Bitcoin's own Binance history, each with its own price path.
Each case's price changes after 12, 24 and 48 hours for this chart length, in basis points (100 basis points equals 1%).
Details about the source data and its limits.

Compare rising and falling cases at the same later time. The search gives no combined average or success rate; we calculated this article's summaries separately.

For hl_analogs, ask: “At my chosen time, find similar past Bitcoin moments on Hyperliquid. Include buying versus selling, changes in open interest (the total size of trades still open) and liquidations (forced closures when a trader's money runs out). Show what happened over the next four hours.”