SEC Filings Pulse

Methodology

Everything on this page is derived from SEC EDGAR, a work of the US federal government and therefore public domain. We add no editorial judgement to a row: the company said it, the SEC published it, we counted it and linked back to it.

Filing signals — SEC's own item codes

The 8-K tables above are built from the items field SEC publishes alongside every filing ("5.02,9.01"), and the Form-4 counts from SEC's own transaction codes S (sale) and P (purchase). No model reads anything to produce these rows. A count is a count of filings and reported transactions, nothing more — it is not a claim about why an insider sold, and a routine 10b5-1 sale looks identical to any other.

Form 4 — structured parse of sworn fields

Leadership facts on company pages (CEO, board member, officer, ≥10% owner) are read directly out of the ownershipDocument XML a Form 4 filer swears to: officerTitle, isDirector, isTenPercentOwner. These carry the filed tier because a person under penalty of perjury typed the field we are quoting.

Exhibit 21 — structured parse of sworn subsidiary lists

Subsidiary and parent relationships come from the Exhibit 21 list attached to a 10-K, parsed as a table or list, never inferred from prose. We cap the parse at 300 subsidiaries per filing — a handful of financial holding companies file lists longer than that, so their tail is not represented.

8-K narrative extraction — measured, and withheld

We also run a local relation-extraction model (no LLM) over the narrative text of 8-K item sections and their earnings-release exhibits. Those facts are not published on this site. On 2026-08-23 we audited a random seeded sample of 25 of them against the source filings by hand: 8 exactly right, 8 imprecise, 9 wrong — about 32% accurate as written and about 36% simply wrong. The failures are systematic (a ticker parenthetical read as a corporate parent, a reporting segment read as a subsidiary, a syndicate lender read as a partner), not random noise. They stay held back until a re-audit reaches 80% exactly-right. We publish the number rather than the facts because hiding either would make this page worse than useless.

Company links, and where they are wrong

Filings are attached to companies through SEC's CIK identifier. That matcher produced duplicate company rows (one accession pointing at several near-identical entities) and a small number of outright mis-links — the clearest one we have found attributes LSI Industries' filings to LSI Corporation, a different company. This page guards against both: the CIK inside a filing's own sec.gov URL must match the company's CIK, only one canonical company row per CIK is shown, and rows are deduplicated on accession number. Rows that fail those checks are dropped rather than shown wrong, so the counts here are a floor, not a total.

What this is not

Not investment advice, not a complete picture of any company, and not real-time — the ingest runs daily and this page is cached for ten minutes. If a row matters to you, click the accession and read the filing.