Rally Radar
How Early Do Altcoin Rallies Show Up on a Scanner? We Measured 1,506
How early can a scanner spot an altcoin rally? We went back through five months of hourly data on 315 Binance pairs, found every run of +20% or more, and checked whether our scanner had flagged the coin beforehand. About a third of them had been flagged, typically around a day in advance. That sounds like foresight. Most of it isn't, and the reason is the most useful thing in this post.
What we counted
"A rally" needs a definition before it can be counted, so here is ours. A rally starts at any hourly close from which the price went on to trade at least 20% higher within the next 48 hours. One long run on one coin counts once, not once per hour it kept climbing. Across 315 USDC pairs between April and late August 2026, that gave 1,506 rallies.
A rally counts as caught if the scanner's net score reached 75 or more on that coin at any point in the 48 hours up to and including the hour the rally began. The lead time is measured from the first such signal, because the first time the scanner spoke is when the warning started.
We picked 75 because it is roughly where the scanner's ranking gets selective: fewer than 1 in 100 scored hours reach it. Other thresholds are further down.
The flattering answer
Of the 1,506 rallies, 483 had a signal of 75 or more beforehand: 32.1%. The median warning was 26 hours, and half of all caught rallies had between 11 and 39 hours of notice. Bigger moves were caught at about the same rate:
| Rally size (within 48h) | Caught by a 75+ signal | Median lead |
|---|---|---|
| +10% or more | 30.6% | 25h |
| +20% or more | 32.1% | 26h (half within 11–39h) |
| +30% or more | 30.6% | 25h |
| +50% or more | 33.5% | 25h |
If you stopped reading here, you would conclude that the scanner sees about a third of big altcoin moves coming a day early. That is how most tools would present this number, and it is not wrong. It is just missing the part that tells you what it is worth.
The question the flattering answer skips
How often does a 48-hour window contain a 75+ signal when nothing happens afterwards?
This is the number that makes recall mean something. A smoke alarm that goes off once a day catches every fire in the house, and tells you nothing. To know whether a signal before a rally is a warning or a coincidence, you have to know how often an arbitrary window, picked at random with no rally in mind, would have had one anyway.
For a 75+ signal, the answer is 26.3%. One randomly chosen 48-hour stretch in four already contains one. So the scanner catching 32.1% of rallies is not a jump from zero to a third. It is a jump from about a quarter to about a third: +5.8 percentage points of real preference for coins that were about to run. That is a genuine difference, and a modest one.
The threshold doesn't rescue it. Loosening the bar buys recall with noise, and tightening it buys precision with misses. The margin over chance stays small either way:
| Signal threshold | +20% rallies with a signal first | Arbitrary 48h windows with one |
|---|---|---|
| 60+ | 66.3% | 63.9% |
| 75+ | 32.1% | 26.3% |
| 85+ | 11.5% | 8.1% |
At 60 the scanner "catches" two-thirds of all +20% rallies, which would make an excellent headline, and it does barely better than a window picked at random. Any recall figure published without the column on the right is a statement about how often the tool fires, not about what it sees.
Why "26 hours early" needs care too
The lead time has two properties worth knowing before you lean on it.
It can't exceed 48 hours, because that is how far back we looked. And the scanner's score tends to stay high for a while once it gets there. A coin that sat above 75 for a day and a half before rallying will show a long lead time, but some of that reflects how persistent the score is, not how early anything was foreseen.
Rallies cluster on the same days. When the whole altcoin market lifts, hundreds of coins rally at once, and those are exactly the days the score runs high across the board. Part of the 5.8-point margin is the scanner noticing when the market is moving rather than which coin will move. The random-window comparison above is taken over the whole period, busy days and quiet ones together, so it doesn't remove that effect.
What the number is still good for
A tool that catches a third of big moves, a day early, at a rate modestly above chance is a very particular kind of useful. It is a watchlist, not an alarm.
- A signal buys you time to look, not a reason to buy. The value of a day's notice is a day to read the chart, check the volume and decide, instead of meeting the coin for the first time on a top-gainers list at +25%.
- No signal is not an all-clear. About two-thirds of +20% rallies had no 75+ signal in the two days before. Whatever started them was not visible to this score. A scanner built from price and volume can only see what price and volume show.
- Most signals are not followed by a rally. Roughly nine in ten 75+ signals were not followed by a +20% move within 48 hours. Treat a flagged coin as a candidate to examine, not a prediction.
The other direction of the same question, starting from the signal and asking what happened next, is on our track record page. There, coins flagged at a high score reached +10% within a day about twice as often as random picks in raw terms, and kept a smaller advantage once the day and the coin's volatility were accounted for. The two findings tell the same story from opposite ends: the ranking leans toward coins that are about to move, and the lean is real but not large.
What this measurement cannot tell you
It says nothing about profit. A rally here means the price touched +20% within 48 hours. It may have fallen 30% first, and nothing here says when anyone would have sold. When we did test the score as a trading strategy, with stops, exits and fees deducted, on data it had not been tuned on, it did not show a profitable edge. That result is published in full on the track record page.
It is descriptive, not a validated finding. Every figure here is measured across the whole period, with no held-out stretch kept aside to check it against. It can suggest a pattern; it cannot confirm one.
It covers one stretch of one market. Five months of one market regime, with a week-long gap in our candle data in late July. A pattern measured over that span may not survive a different regime.
If you use any scanner, ours included, the one question worth carrying away is this: compared to what? A rally caught, a signal that "worked", a win rate. None of them means anything until someone tells you how often the same thing happens by chance.
See what the scanner is flagging now
Rally Radar scans 200+ Binance pairs every 15 minutes and ranks them by net score, with the signals behind each one. Use it to decide what to look at, and do the deciding yourself.
Open the scanner →For how these tests are built and why a win rate proves nothing, read do crypto scanners actually work. For what the net score is made of, see what makes a good crypto setup. And for reading the chart once a coin is flagged, spotting accumulation before a breakout is the place to start.