How to find past crypto charts that look like today's

Finding past charts that look like today's takes five choices: a coin, one exchange, a candle size, a window length, and how many matches to read. The search accepts 5-minute to 1-day candles and windows of 20 to 200 candles, with 1-hour candles and 48 candles as the defaults. Your choices also set how much past there is to search: on Binance, 1-hour data starts in September 2019, but 5-minute data only in January 2026.

The matches are past cases, not a forecast. Below are the steps, then what we measured.

For how much the window length changes the matches, see How many candles for pattern matching?.

What is a chart look-alike search?

A bar, also called a candle, shows the open, high, low and close price over a set time. A chart-shape search takes the latest bars of one market, turns them into a shape, and finds the earlier stretches of that market whose shape is nearest.

Comparand compares a coin only with its own past on the same exchange. A Bitcoin search on Binance returns Bitcoin cases from Binance. It never returns another coin or another exchange. Our plain guide to what a chart analog is covers the idea, and why the same exchange matters explains the one-market rule.

Each match is one episode: it does not overlap the chart you searched, and you get one match per episode, not ten nearly identical hours in a row.

Step 1: Which coin and exchange should you pick?

Pick the coin and the exchange you actually trade or study. The search covers BTC, ETH, SOL and other big coins on Binance, Bybit, OKX and Coinbase, plus Hyperliquid perps. A perp is a futures contract that never expires.

Coverage is not equal everywhere:

• Binance: 10 perps. Coinbase: 7 spot markets, with no BNB, LINK or TRX. OKX: 9 perps, no LINK. Bybit: 10 perps.
• Hyperliquid: about 210 perps.

Before you search, ask for the venue list. That check is free. It tells you which exchanges, datasets and bar sizes are available.

Step 2: Which candle size?

The search accepts 5-minute, 15-minute, 1-hour, 4-hour and 1-day candles. One hour is the default.

Smaller candles have less history. On Binance, 1-hour, 4-hour and 1-day data start in September 2019. The 15-minute data starts in October 2024, and 5-minute data in January 2026. A search can only match your chart with windows from the past it has. With 5-minute candles, that past is much shorter.

Step 3: How many candles in the window?

The window is how many of the latest candles are compared. It can be 20 to 200 candles, and the default is 48. At one hour per candle, 48 candles cover two days.

This choice matters more than most people expect. We ran the same search times with 48 and 96 hourly candles. Both searches ended at the same moment, so the longer one just added older prices. We counted how many of the ten closest past cases were exactly the same in both lists.

Step 3: How many candles in the window?

Market (hourly, 300 searches each, 2020 to September 2026)	Average cases kept, of 10	Searches where none were kept
Binance BTCUSDT	1.96	35.00%
Binance ETHUSDT	1.90	29.67%
Coinbase BTC-USD	1.88	39.00%
Coinbase ETH-USD	1.52	42.33%

The typical search kept 1 case. Nine in ten searches kept 5 or fewer (4 or fewer for Coinbase ETH-USD). Even when we counted two matches as the same if they ended within 12 hours of each other, only 1.55 to 2.04 of ten lined up. Going from 24 to 96 candles kept even fewer: 0.40 to 0.57 of ten. The longer window mostly showed different past charts, not the same ones with a few extras.

Neither window was "right". A short window focuses on the recent move. A long one also shows the climb or fall that came before it. Pick the length that covers the move you care about, and choose it before you look at what followed. Our guide on how window length changes the matches shows more cases.

Step 4: How do you read each past case?

For each match you get its dates, its own price path, and what the price did after it. With a 48-candle window, the forward moves are measured at 12, 24 and 48 candles after the match, a quarter, half and full window length. On hourly candles, that is 12, 24 and 48 hours later.

Moves come in basis points (bp). One basis point is one hundredth of one percent, so 100 bp is 1%.

Read the matches one by one and notice the dates. If five matches come from the same crash week, they are really one story. Also check whether the matches come from different years or all from one.

Step 5: How do you read the spread?

The search shows each case and gives no average or hit rate. That is on purpose. You count the ups and downs yourself.

We measured what that count looks like. For 12 markets, we ran 300 searches each, 3,600 in all, using 48 hourly candles and 10 matches. We looked at the middle later move of the ten matches and asked if it pointed the right way.

Step 5: How do you read the spread?

Time after the search	Matches pointed the right way	Price was up (the guess-up score)	Searches where 4 to 6 of 10 matches were up
12 hours	50.44%	51.67%	65.36%
24 hours	50.22%	50.92%	62.72%
48 hours	50.25%	51.39%	63.64%

In every row, always guessing up scored a little higher than following the look-alikes. And in about two of three searches, the ten matches split almost evenly: 4 to 6 went up. Following the middle match move and paying a 0.05% fee on each side, the average 24-hour trade lost 0.07%.

So read the spread as a range of past outcomes, not as a vote. Our guide on when past crypto cases disagree explains how to read a split, and do chart look-alikes predict crypto breaks the 3,600 tests down by market.

What does this not show?

It does not show that look-alike charts are useless for research. It shows that, in our test, the middle of ten matches did not guess direction better than always guessing up. It says nothing about rare cases.

The 3,600 searches also overlap in history, so they are not 3,600 independent tests. The test used only surviving markets. It leaves out funding payments, spreads and slippage (the gap between expected and real fill price).

How we measured this

For the window-length test, we used hourly bars for Bitcoin and Ethereum on Binance and Coinbase, with 300 search times per market from January 2020 to September 2026. At each time we ran a 48-candle and a 96-candle search with ten matches, with matches not overlapping. We counted exact shared ending hours between the lists.

For the 3,600-search test, we ran 300 times in each of 12 markets, using 48 hourly candles, and compared each match list's middle forward move with what actually followed. Each match had to finish before the search began, and its forward moves had to be complete. A live search may meet gaps in the data that this test skipped.

Try this in Comparand

Ask your connected AI agent: "Which exchanges and bar sizes can you search? Then find the ten past charts closest to the last 48 hourly bars of BTC on Binance."

• venue_coverage lists the other exchanges and prediction markets served, with their datasets, kinds and bar sizes. It is free on every key.
• shape_analogs takes a venue and symbol. You can set timeframe, an end time, window (20 to 200, default 48) and k, the number of matches (1 to 20, default 10). It returns each past case with its own path and its forward move at a quarter, half and full window, in bp. It costs $0.01 per search from your credits.
• It never matches another coin or another exchange, and it gives no average or hit rate.

Try an older end time too: "Run the same search as of a past date, then check what followed." You can try 3 coins free, read the tool docs, or see how matching works. These are past cases, not a forecast.

The short version

• Pick a coin and one exchange; a search only matches that coin's own past on that exchange.
• Candle size sets how much history you have; the window sets the shape you compare.
• Going from 48 to 96 candles kept only 1.52 to 1.96 of ten matches, so choose the window before you look at outcomes.
• Read each case and its date, then count the spread yourself.
• In 3,600 tests, the ten closest matches called 24-hour direction right 50.22% of the time, below always guessing up at 50.92%.