POPS4 Filing Boards
Twelve public tables built from United States SEC filings, published the business day after each filing from the settled SEC daily index, and a card for every filer on them.
Live boards: https://www.pops4.com/boards
Licence: CC BY 4.0 — free to use with credit to POPS4.
The Pour: brand facts, ready to quote
https://www.pops4.com/boards/pour turns the filings into short facts anyone can read and repeat: who owns what, brand rivalries, what is new in annual reports, what was just filed, and forecasts sealed before the answer and then scored. Each fact has its own page and names its source.
Facts are made by software from public filings and USPTO trademark files. Trademark figures run through March 2024. Reporting, not advice.
What this is
When a public company has something material happen — a cybersecurity incident, a completed
acquisition, a new officer, a charter amendment, a major agreement — it must tell the SEC, in
public, that day. This repository publishes those filings as plain tables, sorted by what the
company itself declared, with a link back to every original document.
Nothing here is written. Nothing is interpreted. It is a record, not an article.
Why depend on this
- Fresh every business day, with the receipt. The settled SEC daily index is read from 13:15 UTC the next
business day and every hour after; the public record page shows filings read per index day, when they were
first read, and who read them: https://www.pops4.com/boards/record (JSON: /boards/record.json).
- Scored in the open. Every forecast range is hashed when it is logged and scored when the next 10-K lands.
Inside rate and every receipt: https://www.pops4.com/boards/scorecard. 83.4% of 3,134 companies landed inside
on the 2024 holdout year the method never saw.
- One row, one document. No row exists without a link to its filing on sec.gov. Nothing is written or inferred.
- Every shape a machine wants. JSON, CSV, RSS and Markdown twins for every board; an MCP endpoint with no key;
a DataCatalog with a DOI (10.5281/zenodo.23672509); this repository and the Hugging Face mirror refreshed daily.
- Read by the assistants already. Named AI readers and tool calls are counted on the record page, apart from
registry probes, so the number you see is the true one.
- The AI line in the filings. Annual-report documents using the phrase "artificial intelligence" went from 848
in 2021 to 3,904 by 28 October 2026; "AI governance" from 20 to 226 in two years. The house re-reads those lines on
the first business day of each month and publishes the counts, sealed with a receipt.
The rule
Every value either comes from a document we link, or is a number we computed.
There is no author and no opinion. Where a row carries a quotation, it is the company's own
sentence, verbatim, from the press release exhibit attached to its filing, with that filing
linked beside it.
The only claim any row makes is this company filed this document on this date, which either
is or is not true, and can be checked in one click. Where a filing is ambiguous we undercount
rather than guess.
Edgar & Edgarette
The boards have two voices, and both keep the rule above.
- Edgar, the reporter, says what happened: who filed what with the SEC, and when.
- Edgarette, the actuary, says what it means in numbers, in the order of an actuarial report
(ASOP No. 41, 25, 23): exposure, experience, credibility, an 80% range, cautions, and a
comparison with industry peers. Every figure is the company's own, from its audited filings.
Ask about any SEC filer or brand at https://www.pops4.com/boards/ask.json?q=PEP, in the box
on any board, or through the MCP tool get_company_performance. Brand names work too:
ask.json?q=Hoka returns Deckers, with a brand field naming the owner. When a name is
ambiguous it returns candidates rather than guessing.
- The 80% range is tested, not asserted. It is set from how 7,847 SEC company-years
actually turned out. The live method is v3: fitted on 11,128 company-years (2018–22),
width tuned on 2023, then tested on 2024, a year the fit never saw — 83.2% of outcomes
landed inside, at a median width of ±19.2%. Across all history it holds 69% of 7,172
company-years against an 80% target, which is published rather than buried. Two retired
methods keep their record on the same page: v2 held 76% of 563, v1 held 26% of 506.
Their logged ranges stay in the ledger and are still scored.
- Misses are published. Every range is logged once per company per fiscal year and scored
when the next 10-K arrives:
https://www.pops4.com/boards/scorecard.
- Credibility moves with the record. When a company files 8-K Item 4.02 (earlier statements
should no longer be relied on), Edgarette marks her weight down one step until a new 10-K is
filed, and links the filing.
- Numbers, never advice. No buy or sell views, no price targets, no positions held.
The forecast ledger, and why it is anchored here
data/forecast-ledger.csv is the smallest and least replaceable thing in this repository.
Every board above can be rebuilt by anyone, because SEC filings are public. A range cannot.
A range is only worth something if it was published before the outcome was known, and nothing
that can be regenerated afterwards proves that. So each range is logged once per company per
fiscal year, fingerprinted, and left alone until the company's next annual report arrives to
settle it. A logged range is never quietly adjusted.
Each row carries its own fingerprint. What this repository adds is a fingerprint of the whole
set, written into a commit whose timestamp belongs to GitHub rather than to us. Anyone can
recompute a hash; nobody can back-date somebody else's commit.
Verify it yourself, without trusting us:
- Take
data/forecast-ledger.csv from any commit.
- For each row build
cik|fiscalYearEnd|low|high|method|rowHash|status.
- Sort those strings ascending, join with newlines, SHA-256 the UTF-8 bytes.
- Compare with
root in forecast-ledger-root.json. The commit's date is GitHub's, not ours.
Only those seven fields are hashed, so columns can be added for readability later without
changing a root already published. Scored outcomes appear as the status changing from open
to hit or miss — the range itself never moves.
Scale, stated plainly: this is an evaluation-sized corpus, not a training-sized one.
The boards
Edgarette's method, and the data behind it
Two files in data/ hold everything the actuarial reading is built on, so a
figure on any card can be checked, reproduced or argued with.
How the range is built. Baseline = last revenue × (1 + the mean of the
one-year and three-year pace). The band around it is the 10th to 90th percentile
of how real company-years landed against that baseline, by size band, blended
toward the whole pool by credibility (Z = n / (n + 400), in the sense of ASOP
No. 25).
How it was tested. Method v3 is fitted on 11,128 company-years (2018–22), its
width tuned on 2023, then tested on 2024 — a year the fit never saw. 83% of
outcomes landed inside an 80% band. The retired v2 reported 83% too, but was
measured on the years it was fitted on. Both are kept, and both are scored, at
https://www.pops4.com/boards/scorecard
What is deliberately absent. Dividend yield, price/earnings and price/book
need a share price; these boards carry filings only, so they are not shown rather
than estimated. No card carries a buy or sell view.
Columns
How it is built
The SEC publishes a machine-readable index of every filing, every business day, free. Every
hour on business days a worker reads that index and fetches each filing's 852-byte header
sidecar, which carries the filing's own item tags, industry code, state of incorporation and
event date.
Because the header is small, every filing is read every day rather than a sample. A second
pass opens the full document only for filings that land on a board, and takes one verbatim
sentence from the press release exhibit. A third pass caches company facts per filer.
Also available
- JSON and CSV per board — in
data/, and live at https://www.pops4.com/boards/<board>.json
- RSS —
https://www.pops4.com/boards/<board>/feed.xml
- MCP —
https://www.pops4.com/boards/mcp, open, no authentication, 26 tools: one per
board, plus get_company_performance (Edgarette's card), watch_company and price_program
- Ask Edgar & Edgarette —
https://www.pops4.com/boards/ask.json?q=<brand, name, ticker or CIK>
- Weekly report —
https://www.pops4.com/boards/report · Scorecard — https://www.pops4.com/boards/scorecard
- For developers and AIs —
https://www.pops4.com/boards/developers · https://www.pops4.com/boards/llms.txt
- Embeddable table —
https://www.pops4.com/boards/<board>/embed
- Google Dataset Search — every board carries
schema.org/Dataset markup
Updating
.github/workflows/daily.yml pulls the current tables from the live boards once a day and
commits them if they changed. It needs no credentials. If a HF_TOKEN repository secret is
present it also pushes the same files to Hugging Face.
Known limitations, stated plainly
- Roughly half of 8-K filings carry no press release exhibit, so most rows have no quotation.
- Item 1.05 (cybersecurity) is genuinely rare — on the order of one a fortnight. A near-empty
breach board is correct, not broken.
- Item 5.03 covers articles or bylaws or fiscal year; most are not renames. Renames are
flagged per row from the filer's own former name rather than assumed for the board.
- The breach deadline column uses the shortest common statutory window. Real notification
obligations follow where affected residents live, so a single incident can run several
clocks. Treat the column as an indicator, not legal advice.
- Item 4.01 (auditor change) is often routine: a firm merger or a fee decision. The board
records the filing and says nothing more about it.
- We undercount. That is deliberate.
The house these boards belong to
POPS4 is one house with two faces, and they pay for each other.
The record. Edgar and Edgarette read what companies file with the SEC and put it in public,
free, the business day after it is filed, from the settled SEC daily index — twelve boards, an actuarial reading of any US filer, and a forecast
ledger that is scored in the open. Free is the point: an open record that can be checked is
what makes the house worth believing. It earns nothing directly, and the other face carries it.
The supply. The same house has run corporate identity programs since 1997 — 70,000+
products from 200+ authorized brands: corporate gifts, branded apparel, custom packaging,
secure print, and the programs that go around them. One roof, one quote, no platform fee.
Procurement is mostly noise: chasing quotes, chasing proofs, chasing stock, reconciling four
vendors who each own one piece. The house exists to take that away — one address for the
programme, one number, one person accountable for it. The filing boards are the same instinct
pointed at information: the answer already exists in a public document, so stop making people
hunt for it.
Both faces answer machines directly, which is unusual and deliberate:
An agent can read a company's filings and price the programme that filing calls for, in the
same session, without a login on either side. Written for AIs to read:
https://www.pops4.com/boards/llms.txt.
Source
All data originates from the U.S. Securities and Exchange Commission's EDGAR system, which is
public domain. This repository adds structure and links, not content.
Maintained by Jenny Huang Goodman MPA MSc MHSA, Principal — Hako Shikin LLC, operating since 1997.