The universe is chosen before you touch any filters
Every screener begins from a list. On a stock screener that list is usually exchange-listed securities, sometimes only primary listings, sometimes excluding funds, trusts and instruments below a price floor. On a crypto screener it is whichever tokens the provider has chosen to track, which is a commercial and operational decision rather than a defined market.
This matters more than any individual criterion. Two screeners applying identical filters return different sets because they started from different universes, and neither is wrong. Before trusting a count, find the universe definition; it is normally in the documentation rather than on the screen itself.
What the filters actually remove
A filter is an exclusion, and the useful question is always what it excluded rather than what it kept. Set a minimum daily volume and you have removed everything thinly traded, which is a liquidity decision disguised as a quality one. Set a maximum price and you have removed large, expensive, often stable assets. Require a market capitalization above a threshold and you have removed every newer asset regardless of merit.
The subtler exclusion is the missing value. When a field is empty, with no reported supply figure or no available ratio, most screeners drop the row rather than flag it. That means a filter on any criterion silently removes everything for which the data is unavailable, and on the less-covered end of a crypto screener that can be a large share of the universe.
The exclusion nobody sets deliberatelyA screen with five criteria has usually applied a sixth without being asked: "has data for all five". On well-covered large assets that is harmless. On the long tail it is the strongest filter in the set, and it correlates with exactly the things a reader might have wanted to discover.
Which screening criteria survive the move to tokens
Some criteria transfer directly, because they describe trading rather than a business: price, volume, market capitalization, volatility, and the change over a period. These are the columns a crypto screener and a stock screener genuinely share, and they are all market data.
Almost nothing from the fundamental side transfers. There is no income statement for a protocol, so earnings, margins, book value and every ratio built on them have no counterpart. Substitutes exist: circulating against total supply, issuance schedule, fee revenue, active addresses. They are not equivalents, and comparing a token on issuance with a company on earnings is a category error rather than a clever cross-asset screen.
Market capitalization is the trap worth naming, because the words are identical and the arithmetic is not. For a listed company it is share price times shares outstanding, with a free-float variant that excludes locked holdings. For a token it is price times circulating supply, where the definition of circulating varies by provider and unlocked-but-unmoved supply may or may not count. The screening criteria look comparable in the column header and are not comparable underneath it.
Two screeners, one question, different answers
The most instructive experiment with any screener takes ten minutes: write one question, run it on two providers, and compare the counts rather than the names. They will differ, often by a large multiple, and the difference is entirely explained by the universe each started from and how each defined the fields you filtered on. Neither is broken.
What that exercise settles is how much confidence a single screen deserves. A result that survives on both providers is telling you something about the companies; a result that appears on one is telling you something about the data. Almost nobody runs the comparison, which is why screen output gets treated as a measurement rather than as one vendor's answer to a question you asked imprecisely.
The number that is never on the screen
Whatever a screen returns, one figure decides whether the result is usable and it is almost never a column: how much of the security actually trades. A criterion can be satisfied perfectly by an asset whose entire daily volume is smaller than the position you were considering, and the screen has no way to tell you that the price it matched on is one nobody has transacted at recently.
Adding average daily volume and the typical spread as explicit criteria fixes this, and it changes the output more than any refinement of the fundamental filters. It is also the criterion most often left out, because it feels like plumbing rather than analysis, which is exactly the reason it belongs in a screen rather than in a footnote after one.
Where a stock screener does more work
A stock screener has decades of standardized reporting behind it, which is why it can offer hundreds of fields. That standardization is also its main hazard: a field labeled P/E has silently resolved which earnings, over which period, adjusted or reported, and what to do with a negative number. Two providers resolve those differently under the same label.
The equivalent problem on a crypto screener is not definitional but temporal. Supply figures change on schedules, tokens are relisted and renamed, and history is frequently rewritten as providers correct their own records. A screen run today and repeated next month may differ because the past changed, not the present.
Building a screen from scratch, in order
Start by writing down the question in a sentence, before touching any control. "Which listed US companies grew revenue while reducing debt" is a screenable question. "Which stocks are good right now" is not, and no arrangement of filters will turn it into one; a screen can only test what you can state.
Then fix the universe deliberately rather than accepting the default: which exchanges, primary listings only or all, funds and trusts included or excluded, any price or capitalization floor. Add criteria one at a time and note the surviving count after each, because that sequence is the only way to see which filter is doing the work. A criterion that removes far more than expected is usually a data-coverage effect rather than a finding about businesses.
Finally, decide the ordering separately from the filtering. Filters answer yes or no; the sort order decides what you look at first, and it quietly becomes the ranking even though nothing about it was validated. Sorting a screen by the criterion you care about least is a cheap way to notice how much the ordering was doing.
A saved screen drifts even when you do not touch it
Screens are usually saved and re-run, and both sides of the comparison move underneath them. The universe changes as companies list, delist, get acquired or move exchange. Field definitions are revised by the provider. Historical figures are restated as filings are corrected, and on the token side supply figures change on schedules and assets are renamed or relisted entirely.
The consequence is that a screen run in January and re-run in April can differ because the past changed, not the present, and nothing on the screen announces that. Re-reading the universe definition and the row count each time is the only cheap defense, and a screen whose output count has moved sharply with no criteria edited is telling you something about the data rather than the market.
Using a crypto screener as a starting point, not an answer
The practical discipline is short. Read the universe definition first. For each criterion, say out loud what it removes. Check what happens to rows with missing data. Then treat the output as a list of things to look at by hand, not as an answer. The screen has told you which rows matched, and nothing at all about whether matching was the right test.
That is the same discipline the rest of this section applies to composite scores and to the statements themselves. A screener is faster than reading filings, and it is fast precisely because it discarded the context. Understanding how a crypto screener works is mostly understanding what got discarded.