What Makes a Good Crypto Setup? Breaking Down a Scanner's Scoring

A good crypto setup is one where several independent signals agree at the same time — momentum, volume, trend structure, and volatility all pointing the same way — on a pair liquid enough for that agreement to mean something. A scanner's score is nothing more exotic than that agreement counted up, penalised for risk, and put on one scale so two hundred coins can be compared at a glance.

Most scoring explanations stop at "we combine multiple indicators," which tells you nothing. This one gives you the actual arithmetic: which components earn points, how many, what subtracts them, and where the cutoffs sit. Then it does the part almost nobody publishes — it shows what happened when we measured whether the score is worth anything, including the results that came back against us.

A Good Setup Is Agreement, Not a Single Signal

Any one indicator, run across a whole exchange, returns garbage. "RSI below 30" hands you a list of coins that are falling for excellent reasons. "Volume spike" hands you wash trading and delisting panic. The signal is not in any individual reading — it is in the coincidence.

So the first design decision in any scoring model is which families of evidence have to agree:

• Momentum — RSI, MACD, Stochastic RSI. Is pressure building, or is the move already spent?
• Volume — is there participation behind the price, or is it drifting on nothing?
• Trend structure — moving-average alignment and higher lows. Is this a constructive base, or a bounce inside a downtrend?
• Volatility — is the range compressed, leaving room for a move to expand? That is the mechanical reading of a Bollinger Band squeeze.

These families are chosen to be as independent as possible, because the whole value of the exercise is that they can disagree. Three indicators that all measure momentum agreeing with each other is one signal wearing three hats. Momentum agreeing with volume is genuine confirmation.

What a Scoring Table Actually Looks Like

Here is the real breakdown Rally Radar uses. Each component contributes points toward a rally score capped at 100:

• Volume expansion — up to 25 points. Current volume against its 72-hour median. The single heaviest component, because it is the hardest to fake with a quiet drift.
• Price / range breakout — up to 20 points. Close above the prior recent range high.
• EMA alignment or reclaim — up to 15 points. EMA9 > EMA25 > EMA50 > EMA200, or a fresh reclaim of EMA25 from below.
• MACD confirmation — up to 15 points. Histogram positive and expanding, or narrowing toward a cross.
• Volatility compression — 12 to 14 points. A band squeeze in the act of releasing, with volume.
• StochRSI reversal — up to 12 points. Turning up out of oversold, again with volume behind it.
• RSI strength — up to 10 points. RSI sitting in a bullish zone of roughly 55 to 70 — not deep in overbought.
• Higher lows / structure — up to 10 points. Successive swing lows stepping up over the lookback.

Two things are worth noticing about that list. The weights are deliberately blunt round numbers, because precision there would be false — nobody knows that volume expansion is worth exactly 25 rather than 23. And volume appears in three separate places: as its own component, and as a required confirmation on both the compression and the StochRSI components. That is not redundancy by accident. A squeeze releasing on no volume is a non-event, and the scoring is built to say so.

The Penalties Do More Work Than the Points

The rally score alone would be a bad ranking, because the easiest way to score highly on momentum and volume is to already have run. So a separate risk score is subtracted:

• No meaningful volume — minus 30. Illiquid or inactive. This one penalty removes most of the noise on any exchange-wide scan: on a dead pair, a single order moves price 8% and every "signal" downstream of that is an artefact.
• Sharp dump — minus 25. A 1-hour drop past −8%, or a 4-hour crash past −15%. Falling knives light up momentum indicators on the bounce.

The final number is essentially net score = rally score − risk score. A coin at rally 80 with risk 30 ranks below a clean coin at rally 60, which is the entire point: the penalty is not a tiebreaker, it is a veto with teeth.

There is a stage reserved specifically for the failure mode this is fighting. When rally is at or above 75 and risk is at or above 25 at the same time, the coin is not labelled a breakout — it is labelled CHASE_WARNING. Every signal is screaming, and that is exactly the problem. Hold on to that label; the measurements later in this article come back to it.

From Score to Stage

A raw number is hard to act on, so the net score is bucketed into named stages:

• CONFIRMED_TREND — 78 and above. Strong trend with full confirmation.
• BREAKOUT — 65 to 77. Breaking out with volume and momentum behind it.
• EARLY_SIGNAL — 50 to 64. Early conditions appearing.
• WATCHLIST — 38 to 49. A base or reclaim forming.
• NONE — below 38. No actionable setup.

Alongside the stage, each pair gets a setup type — a plain-language name for the shape it is making, such as a squeeze release, a flat-base wake-up, a post-spike consolidation, or an oversold reversal. Two coins can arrive at a net score of 70 by completely different routes, and the setup type is what tells you which one you are looking at. Our guide to finding altcoins in accumulation automatically follows one of those setup types end to end.

The Question Everyone Skips: Does the Score Mean Anything?

Everything above is a design. Designs feel convincing, which is precisely why they should not be trusted. A scoring model built from sensible-sounding components can still rank coins no better than a dartboard, and you cannot tell the difference by reading the components.

The test that separates the two is a matched baseline. Not "did high-scoring coins go up" — in a rising market everything goes up. The question is whether high-scoring coins went up more than comparable coins at the same moment. Getting that comparison right takes two corrections that are easy to skip:

• Control for the day. A scanner that fires on days the whole market is running will look brilliant against a baseline averaged over calm days too. Compare each signal against what an arbitrary coin did on the same day.
• Control for the coin. The score's own inputs include volatility and momentum, so of course it selects volatile coins — and volatile coins hit any upside target more often. The honest comparison is against unflagged coins with similar recent volatility, momentum, and liquidity.

What We Found When We Ran It

We ran exactly that on our own data: 315 Binance pairs, roughly 887,000 hourly candles, from April through late August 2026. Overlapping alerts during one rally were collapsed into single episodes, so nothing is counted twenty times. The top bucket — net score 85 and above — is rare, about 0.2% of all scored bars.

• The headline cell. Of 1,327 episodes scoring 85+, 18.5% reached +10% within 24 hours, against 8.4% of matched random entries — a raw gap of about 10 percentage points. Correcting for which day it was, the gap narrows to +6.9 points (95% interval +4.2 to +9.6).
• Ordering holds. Across four score buckets and twenty combinations of target size and time horizon, higher buckets showed higher hit rates in 16 of 20. The score is not noise — rank generally tracks outcome.
• Lead time is real. Of 1,506 episodes where a coin rallied 20%, 32.1% had a signal of 75 or better somewhere in the preceding 48 hours, at a median lead of 26 hours.

That last figure is also a good lesson in reading your own results generously. Catching a third of big rallies sounds strong until you notice that a quarter of all 48-hour windows contained the same signal. Recall is cheap; a threshold loose enough will "catch" everything.

And What Those Numbers Do Not Say

This is where most scoring explainers would stop, having quoted the good number. Four caveats matter more than the headline:

• The pre-registered test was inconclusive. We fixed one cell in advance — 85+, +10% within 24 hours — and ran it on a held-out later period against volatility-matched controls. The result was +2.5 points with a 95% interval of −2.0 to +6.9. That interval spans zero. At 319 matched episodes it is a statement about resolution, not a clean pass and not a clean failure.
• Everything else is exploratory. The figures in the previous section were measured on the same window they were chosen from, with no correction for having examined eighty cells at once. They generate hypotheses; they do not confirm them.
• Chasing shows up in the data. In a few cells the raw advantage was positive while the same-day advantage was significantly negative. On those, the apparent skill was entirely about when the scanner fired, not what it picked: it fired on days everything was moving, and its picks then did worse than an arbitrary coin that day. That is the CHASE_WARNING failure mode, visible in the measurements rather than the design.
• None of this is a profitability claim. "Reached +10%" is a maximum favourable excursion with no stop, no exit rule, and no fees. An episode that touched +10% may have first traded 30% lower. Our separate out-of-sample test of trading the score after costs found no demonstrated edge, and nothing above revises that.

So the honest summary of our own scanner is this: it is a good discovery tool and an unproven trading system. It surfaces coins that move more often than comparable coins do, hours before they move. It has never been shown to convert that into profit after costs, because whether a touch becomes a gain depends on the stop, the exit, and the discipline — none of which live in the scanner. We publish this rather than a better-sounding number because we would rather you calibrate correctly than sign up enthusiastically.

One more result worth the embarrassment: we once weighted setup types by how they had historically performed, derived from our own outcome tracking. Tested out-of-sample, several of those weights were inverted — they had been fitted to noise. Any scoring adjustment that has not survived a holdout is decoration.

How to Read Any Scanner's Score

Whatever tool you use, the same four questions apply — and they are worth asking before you trust a ranking, not after:

• Can you see the components? A score you cannot break down is a score you cannot overrule when you disagree with it. You should be able to see which signals fired.
• Does it penalise, or only reward? A model with no risk term will hand you the coin that has already tripled, every time.
• What was it compared against? "Our signals average +14%" means nothing without a baseline. Compared to what, on which days, against which coins?
• Is a high score a claim about direction, or about attention? It should be the second. The score says this pair is worth thirty seconds of your time — nothing more.

That last distinction is the one that protects you. A score is a triage device that turns an unreadable exchange into a shortlist, which is the same job described in our guides to screening pairs without checking charts manually and scanning every Binance pair for a squeeze. What it is not is a prediction, and no arrangement of RSI and volume thresholds turns it into one.

Final Thoughts

What makes a good crypto setup is independent signals agreeing on a liquid pair, with nothing in the picture that should disqualify it — and a scanner's score is just that judgement written down as arithmetic so it can be applied consistently to hundreds of coins instead of inconsistently to a dozen. The points and penalties are the easy half. The hard half, and the one worth demanding from any tool you use, is a measured comparison against a fair baseline, published with the parts that did not work. Ours finds movers ahead of time. Whether you make money from that is a question about your exits, and the scanner has never claimed to answer it.

If you want the signal-by-signal version of what feeds these scores, start with how a free scanner reads momentum and volume, or the chart-level view in spotting accumulation before a breakout. For the measurement rather than the mechanics — what a scanner has to prove before its score means anything — see do crypto scanners actually work.

← Back to all posts