What Is an SEC Accession Number? EDGAR's Primary Key

What an accession number actually identifies

EDGAR is the SEC's electronic filing system. Every submission that lands in it receives exactly one accession number at acceptance. The thing being identified is the submission, which is best thought of as an envelope. One envelope can carry several documents. A 13F submission typically contains a cover-page XML document plus a separate information table listing every position. A Form 4 submission carries structured XML and usually a rendered human-readable version. All of those documents share the same accession number because they arrived together.

Be precise about what an accession number is not:

• It is not a company identifier. Companies and filers are identified by CIK (Central Index Key). One CIK accumulates many accession numbers over its life; one accession number belongs to exactly one submission, forever.
• It is not a document identifier. Documents inside a submission are addressed by filename within the folder that the accession number names.
• It is not stable across amendments. An amended filing (a 13F-HR/A or a Form 4/A, say) is a new submission, so it gets a brand-new accession number. The original keeps its own and stays in the archive unchanged.

That last point breaks naive pipelines. Model "this fund's Q1 13F" as a single record and an amendment, arriving under a different accession number, either duplicates your data or silently contradicts it. Supersession is a chain of submissions, each pinned by its own number; the downstream consequences are covered in how accurate is 13F data.

The format, decoded

Every accession number has the same 20-character shape: 18 digits plus two dashes, as in 0001234567-26-000123.

Two details trip people up.

First, the leading ten digits are the CIK of the submitter, which is frequently a filing agent rather than the company you actually care about. Law firms and financial printers transmit filings on behalf of many clients, so the prefix of an accession number often identifies the agent, and the sequence counter counts that agent's submissions for the year, not the subject company's. You cannot reliably recover "who is this filing about" from the accession number alone. For that you need the submission's metadata, which lists the subject CIKs and form type.

Second, the dashes are presentational. EDGAR's own systems accept and emit both the dashed form (0001234567-26-000123) and the squashed 18-digit form (000123456726000123), and the directory layout on sec.gov uses the squashed form. Treat the two as the same key and normalize before storing. The SEC documents the structure and crawl etiquette in its Accessing EDGAR Data guide, worth reading once if you plan to fetch filings yourself.

From accession number to permanent URL

Given an accession number and a CIK associated with the filing, you can construct a stable URL straight into the EDGAR archive:

https://www.sec.gov/Archives/edgar/data/<CIK>/<accession-number-without-dashes>/

That folder lists every document in the submission. Appending the dashed accession number plus -index.htm gives the human-readable index page EDGAR renders for the filing. From there you can open the primary document, the information table, the raw XML, whatever the envelope held.

This is what makes accession numbers so valuable: the address is permanent and public. A filing accepted years ago resolves today at the same path, byte for byte, with no authentication and no license. If you want to search across filings by text rather than by identifier, the SEC's EDGAR full-text search returns matches with their accession numbers, which you can then resolve to the archive.

For ownership data, the accession number bridges an analytical claim and its legal source: "this manager held this position at quarter end" is an extraction from a specific information table inside a specific submission, and walking the number back to the archive lets you re-derive the claim yourself. The mechanics of reading what you find there are covered in how to read a 13F filing, and the SEC's own Form 13F FAQ answers the regulatory questions around who files and what gets reported.

Why per-datapoint provenance keys on it

Language models paraphrase, and paraphrased numbers drift. When an AI agent tells you a fund added shares of NVDA last quarter, the only honest answer to "says who?" is a pointer to the filing itself. The accession number is the finest-grained stable citation EDGAR offers: it survives amendments (each version keeps its own number), it never collides (one submission, one number), and it resolves to the primary source with no intermediary.

That is why Arkolith keys provenance on it at the datapoint level. Every holding row in the Q1 2026 dataset (1,824 institutional filers, 1.87 million long positions, $53.7 trillion in reported value) carries the accession number of the submission it was parsed from, and the same treatment applies to the 51,000+ insider transactions tracked from Forms 3, 4, and 5. An agent consuming that data over MCP or REST can cite the accession number instead of asserting an unsourced figure, and anyone can verify the claim against sec.gov directly.

This is an anti-hallucination mechanism, not a compliance flourish. An agent that says "value X, per accession number Y" is making a falsifiable statement. An agent that says "value X" is making a plausible-sounding one. The difference compounds across a workflow with dozens of figures in flight; the full argument is in stop AI hallucinating market data. For a concrete browse, the Warren Buffett fund page surfaces positions you can chase back to EDGAR by accession number.

Deadlines set the rhythm of new accession numbers

Accession numbers pulse with the disclosure calendar, and knowing the rhythm tells you when fresh data exists to be pinned.

The 13F deadlines produce the heaviest bursts. Many managers file in the final days before each cutoff, so the archive fills with a dense wave of new numbers four times a year. Form 4 submissions flow continuously instead, which is why insider data feels live in a way quarterly holdings never will. The full 2026 calendar is in 13F filing deadlines 2026, and the differences between the ownership forms are mapped in 13F vs 13D vs 13G vs Form 4.

Amendments cluster after deadlines too. A wave of /A submissions follows each 13F window as managers correct errors, and every correction is a new accession number that must supersede the original in your data rather than sit beside it as a duplicate.

Querying it through an API instead of crawling

You can crawl EDGAR yourself. The archive is public, the layout is documented, and the accession number makes everything addressable. The cost is everything around it: rate limits, identifier resolution (filings list positions by CUSIP, not ticker), amendment supersession, and parsing each filer's slightly different rendering of the same form. That labor is invisible until it is wrong.

The alternative is an API that has done the plumbing and kept the provenance attached. With Arkolith, one key covers the surface. List tracked institutional filers with their CIKs:

curl -H "Authorization: Bearer YOUR_KEY" "https://arkolith.com/api/v1/funds"

Resolve a name to an entity:

curl -H "Authorization: Bearer YOUR_KEY" "https://arkolith.com/api/v1/search?q=berkshire"

Then pull holdings, with amendment supersession already applied and each row traceable to its source submission:

curl -H "Authorization: Bearer YOUR_KEY" "https://arkolith.com/api/v1/funds/1067983/holdings"

The same data is exposed over MCP, so a tool-using model can call it natively and cite accession numbers in its answers rather than asserting figures from memory. Setup takes a few minutes via the quickstart. The honest framing: if you need one filing once, EDGAR's website is faster. If you need clean, amendment-aware, identifier-resolved data across thousands of submissions, with every datapoint carrying its accession number, that packaging is exactly what you are paying for.

Frequently asked questions about SEC accession numbers

Is an accession number the same as a CIK?

No. A CIK identifies a filer or subject entity and persists across all of its filings, while an accession number identifies one submission. They are related only in that the first ten digits of an accession number are the CIK of whichever account transmitted the filing, often a filing agent.

Does an amended filing get a new accession number?

Yes, always. An amendment such as a 13F-HR/A or a Form 4/A is a separate submission with its own accession number, and the original remains in the archive untouched. Consumers have to chain amendments to their originals and decide which version currently governs.

Can I build the EDGAR URL from just the accession number?

Almost, but not quite. The archive path needs a CIK as well as the accession number, and the number alone only tells you the submitter's CIK, which may belong to an agent. In practice you look the number up via EDGAR search or an API, take the subject CIK from the metadata, and build the archive URL.

Why attach an accession number to every datapoint instead of every filing?

Because verification happens at the level of individual claims, not documents. A holding or insider trade tagged with its accession number can be checked against sec.gov in one step, which keeps agents from asserting unsourced figures and lets a human audit any single number.

This article explains public filings and data concepts. It is not investment advice.