Stock market screener: 1,018 to 5,207

On 2026-09-13 at 08:11 UTC this stock market screener offered 125 filterable fields. The thinnest of them is held for 1,018 listings and the widest for 5,207. The field in the middle is held for 3,800.

That matters because every filter you add to a screen does two things. It removes the companies that fail the test, which is what you asked for, and it removes the companies for which the figure does not exist, which is not. The second group never appears in the result and never appears in a count of what was rejected. It is simply gone.

So the useful question about any screener is not how many indicators it offers. It is how many companies each of those indicators is actually held for. GET /api/screener/fields reports that figure for all 125 fields, and every count below is its reading at 2026-09-13 08:11 UTC.

The count has a denominator, and the denominator is already a selection

The endpoint measured coverage against 5,207 listings. The pool of active US equity listings was 5,518 that day. The difference is 311 listings that held no computed indicator at all, and the endpoint says so in its own response: it names the population as active equity listings that also hold an indicator row, and reports the 311 it left out.

That is the honest way to report it, and it still flatters every number on the page by about six percent. Share price and market capitalisation are held for 5,207 of 5,207, which is where the phrase "held for every listing" comes from. Against the 5,518 listings you can actually type into the search box, the same two fields are held for 94.4 percent, and no field in the catalogue reaches every listing.

Neither row is wrong. The first answers "of the companies this screener has computed anything for, how many hold this field", and the second answers "of the companies I can search, how many". A screen returns results from the second population, so the second row is the one that describes what you will get.

No field is empty any more, and that is not the same as a field being fixed

Until 2026-09-07 two fields were offered as filters and held for nobody: insider ownership and institutional ownership. The code that dropped them records the count, which was zero of the 103,812 indicator rows stored at the time, and the reason: a filter held for no company returns an empty table, and an empty table reads as "no company qualifies" rather than as "this field does not exist here". A request naming either field now answers 422 and says which field it refused.

So the two are not in the catalogue today, and they were not filled in. They were withdrawn. That is the right answer for a filter nothing can be screened on, and it is worth knowing that it is the answer, because it means the absence of a zero-coverage row in a field list is not by itself evidence that every field on the list arrives.

Today no field is at zero. The floor is the PEG at 1,018 listings.

The thinnest fields are the ones people screen on

Twenty-two of the 125 fields reach 90 percent of the counted population. Thirty-four are held for fewer than half of it.

Sort the fields by the group they belong to and the pattern is not random.

The valuation group is the thinnest, and it is thinnest for a structural reason rather than a supply one. A profitability ratio needs a revenue line and a profit line. A valuation ratio needs those and a denominator that behaves: an earnings figure that is positive, a book value that is positive, a growth rate that is not zero. Wherever the denominator misbehaves the ratio is withheld, and the valuation group is where denominators misbehave.

Splitting the same 125 fields by what dates them rather than by what they measure gives a flatter picture. The 87 fields dated by a filed statement have a median coverage of 3,776. The 34 dated by the last quote have a median of 3,968. Four are dated by neither, and they are not what the shorthand suggests: two are per-employee ratios that need a headcount from the company profile, and two are this product's own composite ranks. There is no longer a group of bought-in figures sitting visibly below the rest, because the three that were retired on 11 September 2026 are gone from the catalogue: a modelled discounted cash flow value, the margin of safety measured against it, and Forward P/E.

Missing and withheld are different problems

A field being held for 39.9 percent of listings does not mean the data is bad for the other 60.1 percent. Most of the time it means the figure has been refused on purpose.

Price to earnings is the clearest case. It is withheld whenever trailing earnings are zero or negative, because a negative denominator produces a ratio that reads as a small positive number and means nothing. A screener that shows a P/E of 4 for a company that lost money is not more complete than one that shows nothing. It is wrong in a way that survives sorting.

Ford Motor Company is the worked example. Its FY2025 income statement, filed on 2026-02-11 under accession 0000037996-26-000015, reports a net loss of 8,162,000,000 dollars and earnings per share of -2.06. So the screener holds no P/E for Ford at all, and it holds an earnings yield of

-14.68 percent (stock_indicators.raw_values.earnings_yield for F, recomputed 2026-09-13 03:03 UTC)

in the same row. One field refuses the company and the other describes it, out of the same filing, and only one of them would have dropped Ford silently out of a screen.

The distinction matters because it changes what you should do about it. A withheld figure is telling you something about the company. An absent figure is telling you something about the data supply. The two look identical in a result count.

What this does to an actual screen

Take four conditions of the kind a defensive screen uses: a price to earnings ratio, a price to book ratio, a current ratio and a debt to equity ratio.

The arithmetic follows without any further query. A listing has to hold all four figures to be tested against all four thresholds, so the screen can be applied to at most 2,078 listings, which is the smallest of the four. That is 37.7 percent of the pool. At least 3,440 listings leave the screen before a single threshold is evaluated, and they leave it because no P/E exists for them.

The result you read at the end is therefore a selection from fewer than four listings in ten, and those four are not a random four. They are the ones that reported a profit last year and carry positive book value. Adding the other three conditions cannot make the sample broader, only narrower.

None of that makes the screen wrong. It makes the screen's population different from the population you had in mind, and the difference is large enough to matter.

Reading the coverage before the threshold

Three habits follow from the census, and all three are cheap.

Check the coverage of your scarcest field first. In the four-condition screen above, every property of the result was decided by the P/E, because it is held for the fewest listings. The other three conditions could not add anything the first one had already removed.

Prefer a field with a wider base when two answer the same question. Earnings yield is held for 5,003 listings against the P/E's 2,078, a difference of 2,925. The two are not reciprocals and should not be treated as one figure. What matters for a screen is that the yield survives a negative numerator and the P/E does not, which is the whole of the Ford case above, so it keeps 2,925 more listings in the population and orders them by the same economic quantity.

Read a small result as a question rather than an answer. A screen that returns a dozen rows out of 5,518 can mean the conditions are demanding. It can also mean one of them is held for a fifth of the market, which is what the thinnest field in this catalogue is held for today.

What this leaves out

Every figure above is a single reading of GET /api/screener/fields on 2026-09-13 at 08:11 UTC, except the Ford figures, which come from the production database on the same day: the FY2025 income statement as filed, and the indicator row recomputed at 03:03 UTC that morning. The pipeline rewrites these counts nightly and the pool itself changes with listings and delistings, so treat the numbers as dated observations rather than as the current state. The live coverage figures are on /coverage.

The two populations in this piece are different and are labelled on every count. The endpoint reported coverage against the 5,207 active equity listings that held an indicator row on 2026-09-13. The pool was 5,518 that day. The 311 listings in between held no computed indicator, so they appear in no coverage figure the endpoint publishes, including the two that read 100 percent.

Group medians are medians of the per-field coverage counts inside each group, not a count of companies holding every field in the group. A group whose median is 74.6 percent contains fields far above and far below that.

The census says how many listings hold a figure. It does not say whether the figure is right. A ratio computed from a stale filing counts as held.

Coverage is measured per listing, not per company. A company with two listed share classes counts twice.

Ford's earnings yield is computed against the market capitalisation stored at the moment of the recompute, so dividing the filed net loss by the market capitalisation shown elsewhere on the site today will not reproduce the second decimal place. The sign and the order of magnitude are the point, and both come straight off the filing.

The four-condition arithmetic uses the smallest of the four coverage counts as the ceiling on the tested population. That is exact as a ceiling, and it is not the tested population itself, which can only be smaller. Measuring the exact overlap of four fields would need a count the public endpoint caps.

Nothing above is a view on any company or any screen. It is a description of where the data behind a filter stops.