POPS4 Filing Boards

Nine public tables built from United States SEC filings, published the day those filings appear.

Live boards: https://www.pops4.com/boards Licence: CC BY 4.0 — free to use with credit to POPS4.

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.

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 at https://www.pops4.com/boards/ask.json?q=PEP, in the box on any board, or through the MCP tool get_company_performance.

• 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:

1. Take data/forecast-ledger.csv from any commit.
2. For each row build cik|fiscalYearEnd|low|high|method|rowHash|status.
3. Sort those strings ascending, join with newlines, SHA-256 the UTF-8 bytes.
4. 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, twelve 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=<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