How to cut noise and false hits from a screener
A false hit is not a screener error: the condition really was met, there was simply nothing behind the move beyond one large order or a market-wide wave. Noise is cut by four settings — a liquidity floor, the interval, per-coin limits, and combined conditions — and all four live in the screener form, not in the head of whoever reads the chart.
In short
- An unwanted hit is not a malfunction: the condition was met, there was simply nothing behind the move.
- The main filter is a liquidity floor: it is set in millions of dollars and removes thin pairs with a single field.
- Repeats of one move are cut by the per-coin cap and pause, not by a tighter threshold.
- The wash-volume filter works on two scenarios only of the turnover screener, and is impossible on spot.
Why the number of unwanted hits is never zero
A screener measures rather than predicts: it compares the value now with the value N minutes ago and reports when the difference exceeds your threshold. Noise appears where the measurement is correct and means nothing:
- A thin market. One large order moves price by several percent, and the screener faithfully records it.
- A market-wide move. When Bitcoin moves sharply, dozens of altcoins fire after it — that is one event, not dozens.
- A threshold that is too sensitive. A 0.5% rise in a minute is an ordinary wiggle on most pairs.
Noise cannot be driven to zero, and it is not worth trying: a setup that lets nothing unwanted through usually lets nothing through at all. The practical goal is to bring the flow down to what you actually read. If dozens of alerts arrive a day and you open only a few, the threshold is too loose.
Filter one: the liquidity floor
Two fields in the Filters block exist on every screener, and both are set in millions of dollars: minimum 24-hour turnover and minimum open interest.
The numbers the presets actually use
This is not a recommendation but what the presets carry after being tuned against the observed alert flow:| Preset | Turnover, $m | OI, $m |
|---|---|---|
| Minute turnover spike ×30 | 10 | 5 |
| Price, OI and turnover rise together | 10 | 5 |
| Price rise with new positions | 10 | 3 |
| Price rise on rising turnover | 10 | 0 |
| Pullback in a trend (indicators) | 50 | 5 |
Why a turnover floor beats a tighter threshold
A tighter threshold cuts events by strength, and along with the noise it discards moderate moves on liquid pairs. A turnover floor cuts by a different property — whether there is a market behind the move — which is what lets the threshold stay moderate. A measurement showed this plainly: on the minute-turnover spike with floors of $0.5m and $1m, the screener produced 495 hits across 33 usable minutes, all of them inside one market episode. With $5m and $10m the same window left 15 independent episodes.An open-interest floor cuts off spot
Open interest does not exist on spot, so any non-zero floor on it removes spot venues entirely. That is why the preset for a price rise on rising turnover keeps it at zero: a price-plus-turnover combination makes sense on spot too. Why a dollar floor does not guarantee a fill at your price is covered in the article on liquidity.
Filter two: the interval, and what the threshold is compared with
The interval runs from 1 minute to 4 hours
The interval sets how old the value being compared is. A short window pulls order-book twitches into the sample; a long one leaves moves that went somewhere. Threshold and interval are chosen together: 3% in 5 minutes and 3% in 2 hours are different events with different frequencies.“Unusual move for this coin”
The price and open-interest screeners have a separate mode: the hit comes not from “rose by N%” but from “N times stronger than is usual for this coin”, measured on the same pair and venue. It works together with the percentage threshold, not instead of it — a move has to pass both, or a coin stuck in a dead range would report “five times its usual” on hundredths of a percent. While observations are few, the mode filters nothing.Letting the threshold follow the market
The price, open-interest, and turnover screeners have an adaptive switch: the screener moves the value itself inside the range you set — up when there are too many alerts, down when there are almost none. It is recalculated daily from your own flow; your own numbers stay put, and the value in force is shown in the form.How the screener helps
Liquidity floors, the interval, the cap, and the pause sit in the form of every screener — the Filters and Limits blocks are the same everywhere. So a threshold-plus-floors pairing you dial in once carries over to a neighbouring screener unchanged, and a ready-made preset is the easy way to start: floors and limits are already in place.
A price move without turnover is the first candidate for noise. If hardly anything traded, there is little behind the move.
“Not verified” is not “clean”. A spike the wash filter could not assess is passed through, not discarded.
Filter three: the cap and the pause on one coin
The screener sweeps the market every minute, and one long move satisfies the condition for many minutes in a row. The per-coin rate is handled by the Limits block and its two different limiters:
- Alerts per coin, plus the period. How many alerts on one pair are allowed within the period you pick — from 15 minutes to 24 hours. The cap is off by default.
- The pause between alerts per coin. The minimum gap between two alerts on one pair, from 0 to 1,440 minutes. Zero means off, and it works independently of the cap: the cap limits the total count, the pause the repeat rate.
Repeats of one move in the dashboard feed are also collapsed into a single card with a counter. That is why the feed shows which pairs fire again and again — the fastest way to find the coin that made your flow unreadable.
The wash-volume filter: what it checks and what it cannot do
Wash volume is trading with yourself to make a coin or an exchange look busier. A separate filter scores a spike on three observable properties and returns a verdict — clean, suspicious, or wash.
Three properties and two modes
The filter looks at what share of turnover turned into open positions, whether price moved while it did, and how deep the venue is — the pair's open interest against its own daily turnover. Two modes: soft cuts only obvious wash trading, strict cuts borderline cases too. A separate switch keeps suspicious spikes in the feed with a badge.Two scenarios of the turnover screener only
The filter applies to the minute-turnover spike and to the dollar threshold on minute turnover. A percentage change of turnover over 30 or 60 minutes it never touches: such a change is not a signature of wash trading.On spot it is impossible
The filter's main property is the share of turnover that became open positions, and there is no open interest on spot. The metric there is not “bad”, it is undefined, so a spot leg is marked unverified and passes through. A freshly listed coin behaves the same way.We make no claim about its accuracy
Its thresholds are derived from the definitions of the metrics rather than calibrated against a labelled sample — no such sample exists. The filter removes some spikes with a characteristic profile, and how accurate it is has not been measured. More is in the article on the turnover screener.Where filtering breaks
Settings cut the flow, but they do not replace a decision. Three things are worth keeping in mind:
- A screener's “accuracy” cannot be computed. An alert carries no direction — it reports what happened, not where to go. So neither “entry accuracy” nor a share of correct hits is defined, and promising them would be untrue.
- A tight setup is silence, not quality. Tighten a threshold a little at a time and watch how the flow changes. A useful habit is to note which setups bring alerts you actually open.
- Combined conditions filter better than one threshold. The custom screener requires price, open interest, and turnover to line up at once, which lets every threshold stay moderate. Combinations are worked through in the article on the custom screener, and its form fields in its settings.
Filtering noise FAQ
What counts as a false screener hit?
A hit where the condition was met correctly but nothing stood behind the move: one large order on a thin pair, a market-wide wave, or a threshold that is too sensitive. It is a property of measurement, not a malfunction.
How do I reduce the number of unwanted alerts?
Start with the liquidity floors in the Filters block: they are set in millions of dollars and remove thin pairs. Then lengthen the interval, turn on the per-coin cap and pause, and only after that tighten the threshold itself.
What minimum turnover should I set?
The presets use $10m and up, and the preset for the technical signals screener uses $50m. A measurement on the minute-turnover spike: with floors of $0.5m and $1m nearly all hits fell inside one market episode, while $5m and $10m left 15 independent episodes over the same window.
What is the difference between the alert cap and the pause?
The cap counts alerts per coin within the period you pick; the pause sets the minimum gap between two alerts on that coin. They work independently: the cap limits the total, the pause the repeat rate. Zero in the pause means off.
Does the wash-volume filter work on spot?
No. Its main property is what share of turnover became open positions, and open interest does not exist on spot. A spot leg is marked unverified and passes through: not verified does not mean clean.
Can a screener be configured so that false hits never happen?
No. A setup that lets nothing unwanted through usually sends nothing at all. The realistic goal is to cut the flow down to the volume you have time to read.
The core rule of working with a screener: an alert is not a trade. The screener takes over the routine and shows where something unusual happened; the trade, the position size, and the risk stay with you. How to take a single alert apart is in the article on reading an alert breakdown, and how to keep risk in check in the piece on risk management. Which market parameters screeners watch at all is covered in the overview of cryptocurrency parameters.
Read next
How to read an alert breakdown
The four lines under an alert: the 1h RSI and the daily SMA200, the share of coins that moved the same way, the Upbit premium, and a nearby announcement.
Open interest screener scenarios
Four ways to read open interest: which thresholds and intervals to set in the screener, what the alert says, and where each scenario breaks.
Price screener: scenarios and thresholds
Price screener scenarios: which percentage and interval to set for an impulse, a breakout, or a drop, what the alert carries, and where each one breaks.
