How to Find Altcoins About to Move: What Our Scanner Does Well

Most of what we have published about our own scanner is about what it can't do. It has no demonstrated trading edge, and it misses most big rallies. This post is about the one thing it measurably does well: finding altcoins about to move and putting them near the top of the list. We tested that with the same rules we used for the failures, and we'll show the controls it had to survive.

What "does well" means here

The claim is narrow on purpose. It is not "the scanner makes money" (we tested that, and it doesn't). It is not "the scanner sees rallies coming" either, since about two-thirds of big rallies got no strong signal beforehand. The claim is this: among the coins the scanner ranks highest, unusually many go on to make a sharp move within hours. In other words, the ranking concentrates movers.

To test that, we went back through five months of hourly data on 315 Binance USDC pairs. Each time a coin's net score crossed into a band, we asked whether its price touched a target, such as +10%, within a fixed window. We compared that with the same question asked at random moments across the same coins and weeks. Repeated hours of one run count once, so a coin scoring high for twenty hours in a row is one observation, not twenty.

The ranking puts movers on top, in order

Here is the share of episodes that reached +10% within 24 hours, by the score at the moment of entry:

Two things matter in that table. The top band reached the target a bit over twice as often as a random pick. And the rows go up in order: each step up the score ladder hit the target more often than the one below it. Across 20 combinations of target size and time window, the four bands lined up in that order in 16 of 20. A ranking that only got lucky at the top wouldn't usually hold its order all the way down.

The effect is sharpest over short windows. For a +5% move within 4 hours, the same bands read:

That is more than three times the random rate. The top band is also rare: only about 0.2% of scored coin-hours reach 85. On a typical scan that is a handful of coins out of a couple of hundred, which is the size of list a person can actually look at.

Control one: is it just noticing busy days?

The most obvious way to fake a result like this is by accident. When the whole altcoin market lifts, scores run high everywhere, and random coins also hit +10% more often that day. A scanner that only noticed "today is a busy day" would look like a good scanner.

So each episode was scored against random entries from the same UTC day, and the confidence intervals were built by resampling whole days, because three hundred altcoins rallying in one afternoon are close to one event, not three hundred. After that correction:

Some of the raw gap was the day, as expected. Most of it wasn't, and every interval stays clear of zero.

Control two: is it just picking volatile coins?

This is the harder objection. The score is built from volatility, momentum, volume and distance to a breakout. A coin that already swings 8% a day will touch +10% more often than a stablecoin-like large cap, whoever picks it. If that explains everything, our scoring model is an expensive volatility filter and should be replaced by one line of code.

To test it, every 85+ coin was paired with coins the scanner did not flag, priced in the same hour, whose volatility, 24-hour momentum and liquidity ranked within a narrow band of its own. Close to the same kind of coin at the same moment, minus the score. Each step adds one more thing to hold equal. Here it is for +10% within 24 hours, measured on the stretch before the 2026-07-17 cutoff:

Holding volatility, momentum and liquidity equal took away roughly half the advantage. The other half stayed, and its intervals stay above zero. (The last row ticks up rather than down because adding liquidity leaves some coins with no comparable twin. It averages a slightly different set of episodes, so it shouldn't be read as liquidity giving part of the effect back.)

That remainder is the closest thing to a positive result this project has. It means the score is not only a volatility filter. Something in how it combines volume, structure and trend picks out coins that move more than equally jumpy coins at the same hour.

The honest asterisk on that result

The ladder above was measured on data we had already looked at. To check that it wasn't an artifact of designing the comparison after seeing the answer, we wrote one test down in advance: the fully matched comparison, +10% within 24 hours, on the weeks after 2026-07-17. We ran it once. It came back at +2.5 pp, with an interval of −2.0 to +6.9, on 411 episodes spread over 34 days.

Positive, but the interval spans zero, so it is inconclusive. Thirty-four days of entries isn't enough data to tell a few percentage points from noise. That is not a finding that the effect vanished. The test runs once more when the held-out stretch is about three times as long, around November 2026, and the track record page will change with whatever comes back, including if it comes back worse.

Where it stops working

Being clear about where a tool is good also means being clear about where it isn't.

• Longer windows. At 48 hours, the same-day correction turns several of the raw advantages negative. In the 60–74 band, the chance of +5% within 48 hours went from a small raw lead to −10.9 pp behind a same-day random pick. Over two days, the mid-scored coins looked good only because they were flagged on days the whole market was rising. The scanner is about the next few hours, not the next few days.
• Profit. "Touched +10%" says nothing about whether it fell 30% first, or when anyone would have sold. When we tested the score as a strategy with stops, exits and fees on data it hadn't been tuned on, it did not show a profitable edge.
• Most flagged coins still don't move. Even in the top band, about four in five did not reach +10% within the day. A doubled hit rate starting from 8% is still a minority.

How to use a ranking like this

Put together, the measurements describe a specific tool: a way to decide which few charts to open first when a couple of hundred pairs are moving at once. Used that way:

• Read the top of the list, not the whole list. The advantage sits in the highest bands. Below 60, the score does no better than random, because almost every coin-hour is there.
• Look soon. The lead is strongest over 4 to 24 hours. A coin flagged yesterday morning is old news.
• Treat a flag as a reason to check the chart, not a trade. The scanner's job ends at the shortlist. Entry, stop and size are still yours, and they decide the outcome far more than the shortlist does.
• Look at why it scored. Each coin on the scanner shows the signals behind its score. In our signal ablation, volume was the only component whose removal clearly hurt, while RSI mostly repeated the trend signals. So it is worth checking whether volume is part of a high score or whether the score is one uptrend counted three times.

What this measurement cannot tell you

It is mostly in-sample. Apart from the single pre-registered holdout, every number above was measured on the same five months used to study the score. It can show a pattern; only the holdout can confirm one, and so far it hasn't been able to.

It covers one market. April to August 2026, one regime, with a week-long gap in our candle data in late July. A ranking that concentrates movers in one kind of market may not do it in another.

It is not a forecast for any single coin. A shift of a few percentage points across a thousand episodes says nothing about the next one.

That is what a scanner can honestly claim to do well: not predict, not trade, but sort. Ours sorts movers toward the top, and does it noticeably better than chance over the next few hours. The full methods, including everything that came back against us, are on the track record.

For the tests a scanner's claim has to pass, and the result that went against us, read do crypto scanners actually work. For the opposite question, how many big rallies got a signal at all, see how early altcoin rallies show up on a scanner. And for what goes into the score, what makes a good crypto setup breaks it down.

← Back to all posts