Distill Markets · Methodology · Version 2026.10.1 · Revised 24 October 2026 · Last reviewed 24 October 2026

Methodology

How the data is built, from the filing to the served field, and how it changes. Every per-company value names the filing and the XBRL concept it came from, so it can be checked against EDGAR. Computed metrics and aggregates are our work, so their rules are stated here in full, with where they stop.

Scope and sources

Coverage is US domestic filers of 10-K and 10-Q. At 2015-12-31 the as-of universe holds 3,925 issuers (listed, over the revenue floor; the covered-company count on the docs ramp is a wider basis), read 24 October 2026. The floor and the way screen and export counts differ are under Revenue floor.

Companies filing 20-F or 40-F are outside every computed surface: the screener, sector aggregates, Piotroski, Altman and Sloan, the health ratios, capital cycles, revisions and every as-of view. They resolve on the company profile and the raw fundamentals lookup only. A 404 from a computed endpoint for a ticker that resolves on the profile means outside coverage. Absence from a screen is not evidence about a company.

Universe and identity

A row is keyed by the issuer’s CIK, the SEC registrant number, which is the only identifier that survives a symbol being reused or changed. Each row also carries three labels that answer different questions: the symbol the issuer trades under today, the symbol it traded under on the row’s date, and whether its common was exchange-listed on that date.

The identity fields

In the panel export the same values are the columns symbol_at_date, symbol_at_date_source and listing_status_at_date. Their known gaps are under Limitations.

One holder per symbol per date

Symbols get reused: BBBY belonged to Bed Bath & Beyond Inc and now resolves to the Overstock successor that adopted it. A symbol belongs to one issuer per date in the as-of universe, so two rows never share a label (checked nightly).

Listed equity

The screener covers US-listed common equity of US domestic filers. Many entities file XBRL without ever being a listed stock: captive finance subsidiaries, bond vehicles, preferred-stock trusts, ETFs. Each CIK is classified from its EDGAR submissions record, and preferred-share symbols are excluded.

Revenue floor

The universe is issuers with revenue of at least $10M. The panel export applies the floor as a retention rule. An issuer enters on the first quarter-end at which an annual report carrying a twelve-month income statement that clears the floor is on file, and stays until its listing ends. A company whose revenue later collapses is still a row, with its revenue as filed. A newly listed company, a spin-off or a merger successor joins with its first 10-K, not its first 10-Q. For a bank that tags no revenue total, the floor is read off revenue_financial_basis (see Financials, real estate and IFRS).

The as-of screen applies the floor to each row separately and reads revenue alone, so it can omit an issuer the export keeps, among them every pre-2018 bank that tags no revenue total. It reports the size of its own universe at the requested date as universeCount, so its count for a date can be lower than the export’s.

Ticker to CIK

A ticker in a request resolves to a CIK through the live SEC mapping first, then through a ticker history built from EDGAR submissions, which covers delisted and renamed issuers. Where a symbol has been reused, the most recently active mapping wins. A company with several share classes (GOOG, GOOGL) files once under one CIK, so it is one row in a screen and one input to a sector median.

Facts and concepts

From tag to concept

Filers tag the same line under different us-gaap tags. Revenue alone is tagged several ways. A registry maps the tags in use onto the served concepts and is also the ingestion whitelist: a tag outside it is never loaded. Where several mapped tags are present for one period, one is chosen.

Every served fundamental names the tag it came from (revenueConceptUsed and its siblings), so the choice can be checked against the filing.

Where the facts come from

The nightly sync reads the SEC companyfacts feed. A new 10-K or 10-Q can take days to weeks to appear there, so the sync keeps asking for 21 days before returning the filer to the ordinary rotation. Each night, 10-K and 10-Q filings of ticker-mapped filers made 5 to 120 days ago with no facts on file (up to 100 a night, listed issuers first) are read from their XBRL instance document, the document companyfacts is built from, and loaded under the same key with their accession and filing date. Only non-dimensional numeric facts of the filer itself are kept. The fact versions behind point-in-time reads come from a different feed, the SEC’s quarterly financial statement datasets (see How the data changes).

Plausibility and staleness

Period ends on a 52- or 53-week year

Filers on a 52- or 53-week fiscal calendar end their year on a weekday near month-end, such as 29 or 30 December. The stored periodEnd is normalised to month-end, so it can sit about a day from the date printed on the filing. The value is unchanged, and periods stay comparable across filers.

How the data changes

A point-in-time read returns the filing record we hold, under the methodology version in force when it is computed. It changes for three reasons only: a filing that reaches us late (fact versions are loaded from the SEC’s quarterly financial statement datasets, so a read for a recent date fills in when the SEC publishes, each filing keeping its real filing date); a correction to our own parsing or computation; or a rule change, which moves the methodology version. Corrections and rule changes are listed, dated, at /changelog. Every as-of screen response and every panel export row names the build that produced it in vintage and built_at (builtAt in the JSON screen response). Earlier builds are kept but not served, so to re-run a backtest on exactly the data it used, save the export and its vintage with your results.

Latest-value reads (fundamentals, health, events, the live screen) report the most recent filed position and move whenever a company files or amends.

Computed on request, or read from a build

What the vintage identifies

vintage is pit. followed by twelve hex characters of a SHA-256 hash taken over the text of the as-of computation: the concept mapping, the plausibility guards and the formulas. It is derived, never set by hand, and changes whenever any of those change.

It names the computation only. The fact record grows backwards as coverage gaps are repaired, and the listing model and symbol history are rebuilt. Each can move a past date’s answer with the vintage unchanged. built_at says when the computation last ran for the date. Together they tell two files apart. They do not freeze a number.

Kinds of change, and what we publish

Fact versions are appended

Since 2009 each filing’s own statement of each fact, annual and quarterly, is stored keyed to its accession and filing date. A revised value is stored beside the earlier one, never over it. A row with no filing date is skipped. Text-derived pre-2009 facts are kept per parser version. One repair moved the filing date of 504 accessions where two SEC datasets disagreed onto the companyfacts date, which matched the EDGAR header in every case checked. No value changed.

What is visible as of a date

Each value carries its accession and filing date, so one as-of panel can mix accessions: each fact is the newest statement of itself on the date. The pre-2009 corpus follows different restatement semantics; see Pre-2009 text-derived data.

Detecting a revision

A revision is found by comparing the earliest version held of an annual fact with the latest. The relative change is |change| / max(|original|, 1,000,000 in the fact’s own unit), with a default threshold of 1%. The earliest version counts as a first print only when it was filed within 105 days of period end. Otherwise the row says “earliest held”.

A changed value is not by itself a restatement. Each row carries a changeType naming the mechanism where the evidence identifies one. The public change ledger applies a stricter rule: a change between 2% and 100%, made by a filing within 24 months of the period, that is not an entity change (no revenue originally, revenue later: the de-SPAC signature), counting only the first re-presentation of each change.

Nulls, fallbacks and withheld values

Null means unknown or not applicable, never zero. Four rules produce a null, or fill one, on purpose. Each is stated here so a null can be read for what it is.

Not computed for Financials and Real Estate

Seven screener metrics are defined on a statement shape that banks, insurers and REITs do not have. They are null for those sectors.

A margin filter (net_margin, gross_margin, operating_margin) excludes Financials and Real Estate from the result set, because their margins are not comparable to an operating company’s. The response note says so whenever it happens. Every other metric is computed in every sector. Banks and real estate filers have measures of their own, under Financials, real estate and IFRS.

Altman Z″ and derived liabilities

On the screener and the as-of panel the derivation subtracts parent-only equity, so a filer with significant noncontrolling interests has them counted as liabilities. The bias runs one way: a derived score reads slightly more distressed than a filed one would, never less. The derived totalLiabilities on the fundamentals summary subtracts total equity including noncontrolling interests.

Fallback aliases

Eight concept groups (Equity, Cash, AccountsPayable, RnD, ShareBuybacks, Dividends, Inventory, PPE) accept twelve secondary XBRL concepts that a minority of filers use in place of the primary ones, such as PartnersCapital for equity. About 540 filers gain a group they otherwise lack. Revenue has its own fallbacks: the industry revenue tags filers used before the 2018 revenue standard (UtilityRevenue, OilAndGasRevenue, FoodAndBeverageRevenue and others). A tagged zero counts as empty.

Values withheld as tagging errors

A filed fact that sits an exact factor of a thousand or a million from the issuer’s own settled level for that metric, where the series returns to that level afterwards, is read as a tagging error in the filing and served as null. Nothing is rescaled and no substitute is supplied. A filer that tagged 185 trillion shares outstanding for one period has no share count for that period here, and the rendered filing will show the difference.

The test needs at least 9 periods to establish a level. It exempts per-share figures, where a thousandfold move is usually a split.

Financials, real estate and IFRS

The routes that serve these fields are listed in the API docs. A field not computed for a filer is absent from its response. A non-bank carries no revenueFinancialBasis.

Banks and insurers

On the financial-health panel, banks that tag the lines also get efficiencyRatio, allowanceToLoans, depositsGrowthYoY and netInterestIncomeGrowthYoY on each period, each null unless every input is filed for that period over the same window, and each with a *Basis field naming its inputs. The allowance basis changed at CECL (2020 for large filers, 2023 for smaller ones), so allowanceToLoans is not comparable across that date; the basis field says which side of it a row sits on.

Real estate

Real Estate filers carry ffoNareit, funds from operations on the NAREIT definition: net income plus depreciation and amortization, less gains on property sales, plus real estate impairments, from the filed lines of one fiscal year on the consolidated statement as tagged. It approximates the NAREIT figure. The filer’s own reported FFO can adjust for items the tags do not isolate. ffoNareit is null unless every component, and a real estate investment property line, is tagged for the same year. ffoBasis names the year and the concepts used.

A realEstate block carries ffoPayout, propertyGrowthYoy, rentalRevenueGrowthYoy and debtToGrossProperty, each with its own *Basis field and each null unless every input was tagged for the same period. The block is omitted for a Real Estate filer that tags none of the inputs. Occupancy is not an XBRL concept, so it is not served.

Foreign private issuers (IFRS)

An IFRS filer is outside every computed surface; see Scope and sources.

Computed metrics

Annual basis

The cycle metrics are computed on a fiscal-year basis: the latest fiscal year with income-statement financials. Quarterly rows are excluded from the cycle views, because a single quarter’s capex over a partial quarter’s revenue is lumpy and its growth rate explodes.

Where a per-company call returns the latest quarter, it also carries a trailing-year figure (trailing*). trailingBasis reads “TTM” when four contiguous quarters were available and “FY” when the fiscal year was the only clean window. SEC filings carry no standalone fourth quarter, so where a trailing year closes on a fiscal year end, the fourth quarter is the annual less the first three. Each trailing figure resolves its own window, and trailingBasis names them separately (“mixed: revenue TTM, netIncome FY”) when they differ.

Formulas and bounds

A ratio with a small denominator produces nonsense: a $3M-revenue shell with a one-time gain shows a 300% net margin. Two guards apply: a $10M revenue floor on the screener, and the per-metric bounds above, outside which a value is returned as null. The bounds are wide enough to keep genuine extremes and cut only the impossible.

Sector aggregates

Sector figures are medians across the in-scope companies in each sector, computed per fiscal year so the cycle views form a series. One outlier cannot move a median.

Ownership

Institutional holdings (13F)

Every manager holding $100M or more in 13F securities files a 13F-HR within 45 days of quarter end. Three rules govern aggregation.

Two coverage limits apply. The ticker-to-manager view lists a manager only when the ticker is among their larger positions (books are stored to 95% of value, capped at 200 positions), so it is a partial holder register. Quarterly history spans 2017-03-31 to the current quarter. Values are normalised to whole dollars (pre-2023 filings reported thousands).

Blockholders (13D/G)

Schedules 13D and 13G and their amendments, parsed from the structured XML filings since the SEC’s December 2024 cutover and from filing text back to January 2015. Issuer, CUSIP and percent of class are taken from the document as filed; the percent is never recomputed.

Insider transactions (Form 4)

One row per transaction line, parsed from the Form 4 XML. Issuer identity comes from the document itself, which re-anchors the roughly 30% of filings EDGAR indexes under the insider’s own CIK. Summary figures (open-market buys, sells, net shares) count non-derivative lines only; derivative lines are listed and counted separately.

Limitations

Every limit we know of, with its status: planned for a later release; addressed by the field that answers it; inherent in the source; accepted, with no work under way; or a stated convention. A correction that moves a served value is listed, dated, at /changelog.

• New CIKs and first annual reports

  planned for a later release

  An issuer enters the as-of universe at the first date on which an annual report under its own CIK is knowable. Alphabet and PayPal are absent at 2015-12-31. Kraft Heinz is absent through 2016-12-31, because its first 10-K reports successor and predecessor columns that the SEC’s companyfacts feed does not carry undimensioned. Expect the first year of a new registrant to be missing.

• Symbol history before 2019

  accepted

  The symbol history starts with the 2019 cover-page tag. A few issuers whose only earlier symbol on file is a preferred (DuPont at 2015 and 2016 among them) are dropped by the preferred-share filter at those dates. They are present by CIK on the single-name as-of route. The dated symbol of a row that is present is served as symbolAtDate.

• Symbol at the date: what the Form 4 record can show

  inherent

  symbolAtDate is read from each issuer’s own Form 4 filings. An issuer with no insider filing in the trailing year (commodity and currency trusts, funds, limited partnerships, SPAC shells, many small filers) carries the current label, with symbolAtDateSource reading current-label. A change inside a quarter reads from the next quarter’s first filing, and an insider can enter the old symbol on a later form. Sightings begin with 2009Q1, so the first grid dates look back less than a year. The selection rules are under Universe and identity.

• One listed-equity flag per issuer

  addressed by listing status at the date

  is_listed_equity is one value per issuer, so an issuer that was private for a span and later re-listed (Toys R Us, 2005 to 2018) reads as listed at dates inside that span. Never-listed debt-only filers that reach the universe are labelled by CIK. The dated reading is listingStatusAtDate.

• Listing status at the date: gaps in the filings

  accepted

  listingStatusAtDate is derived from each issuer’s own filings and inherits their gaps. A move from an over-the-counter quote to Nasdaq before August 2006 was registered under Section 12(g), so no filing dates it, and the listing reads as starting at the first insider or ownership filing. A listing with neither a registration nor a Form 25 on file is known only from today’s exchange venue, so once it has ended it reads never. Commodity, currency and bitcoin trusts file no insider forms and read never although their shares trade on an exchange.

• Listing status at the date: ranges that stay open or close early

  accepted

  A listing with no Form 25 under the issuer’s own CIK stays open. A few hundred open ranges have no exchange venue today (uplistings registered and never completed, listings carried on by a successor under a new CIK), so those read exchange past the point they stopped. A debt-only filer whose Form 8-A12B registers notes rather than common reads exchange. Four issuers (CRH, Luxfer, Global Self Storage, Flutter) read registered from a Form 25 onward while their shares still trade on an exchange.

• Banks and other financials before 2018

  convention

  Where no GAAP revenue concept was tagged, revenue is empty and revenue_financial_basis carries net interest income plus noninterest income for the same period. Filter on either column. From 2018 both fill.

• Extension-tagged revenue

  planned for a later release

  A small number of issuers tag revenue under a company-specific concept the registry does not map (Affiliated Managers, E*TRADE and Ensco at 2015 among them) and appear with an empty revenue although a filing exists.

• Foreign private issuers

  accepted

  Companies filing 20-F or 40-F are absent from every computed and screened surface, because those surfaces are built from 10-K and 10-Q filings only. See Scope and sources.

• Aggregates are our calculation

  inherent

  Any per-company input can be checked against EDGAR; a sector median is our calculation under the rules on this page, and no single filing contains it.

• The XBRL record starts in 2009

  inherent

  The as-filed, versioned record begins with the XBRL mandate, so point-in-time queries are bounded there. Earlier years come from the text-derived dataset, never merged with the XBRL record, and every such fact says which dataset it came from.

• Sector on an as-of screen is the raw SIC mapping

  accepted

  Some large companies file under a catch-all SIC code. The live screener carries a curated correction; the as-of screener does not, so on a historical date Walmart and Costco return Consumer Discretionary and Visa and Mastercard return Industrials. Filter on the metrics and join on cik.

• Blockholder and 13F identities do not join

  inherent

  The 13D/G surface keys a filer by its name as filed, because Schedule 13G carries no filer CIK; the 13F surface keys a manager by a family slug or CIK. There is no reliable way to ask whether a blockholder is also a 13F filer.

• Small caps are noisier

  inherent

  Below the revenue floor, sparse or unusual filings make individual ratios less reliable.

Revisions

The version changes when a formula, bound or universe rule on this page changes. Wording and structure changes are listed without a bump. Every API response names its version in the X-Methodology-Version header. A row lists a change under the version that first states it: the rules listed for 2026.10.1 are served from the API tag that carries that version, and a response that names 2026.09.3 was served under the earlier rule set. The page is read against the served data at each release train, and the reviewed date records the last read.

Version 2026.10.1 · Revised 24 October 2026 · Last reviewed 24 October 2026

Methodology version 2026.10.1, last revised 24 October 2026. This page and the accuracy benchmark are our representation as to the accuracy of the data, under section 8 of the Terms of Service.