The Org.AI Foundation · Standards
We copy other people's records and never correct them.
The few we wrote ourselves sit on a second shelf, and say how unfinished they are.
We do change the file format. We do not change what the record says.
- Copied from others
- 18 folders · 43 collections with anything in them · 466 original files keptCounted on disk on 28 July 2026. Git-tracked top-level folders, excluding the dot-directories, node_modules and site/; collections inside them that hold at least one row; the original files kept beside them. All three are countable by hand.
- Written down
- 478,507 entries · 861,416 links between themCounted from the files in this repository the same day. The raw figure was 628,019. The same entry turns up in more than one view of itself, and 478,507 is what survived removing the repeats.
- How much of each
- Not measured yet.Some of these hold a fraction of their standard, and every one of them says so in words on its own row. We have not measured any of them against its publisher's own total, so no share is printed here.
Work
If it is a list of jobs, of the steps inside them, or of how you train for them, it is here.
- O*NET
- every U.S. occupation, its tasks, its skills, its toolsFeeds occupations · loaded db_30_0_text, the O*NET database version 30.0. Public domain, a U.S. Government work.
- APQC PCF
- every process a business runs, named the same way across industriesFeeds process · Process Classification Framework 7.4 Cross-Industry, plus the industry frameworks at 7.2.1 and 7.2.2. APQC membership or purchase; the framework file itself grants a perpetual royalty-free licence with attribution to APQC and IBM.
- BLS
- how many people hold each job, and what each paysFeeds occupations and industries · Occupational Employment Statistics, May 2024; the projections matrix on 2022 NAICS and 2018 SOC. Public domain. One large series file is documented here and has not been decoded.
- Advance CTE
- which course of study leads to which jobFeeds skills and courses · Framework Crosswalk, September 2025. Free with attribution.
- ISCED
- UNESCO's levels of education and its fields of studyFeeds education · ISCED 2011 for the levels, ISCED-F 2013 for the fields. UNESCO.
- CEDS
- the U.S. common vocabulary for school dataFeeds education · no version stated. A sample of its elements, not the set it publishes.
- CASE
- the format for publishing a set of learning standardsFeeds education · no version stated. The frameworks and the types, and none of the competencies underneath them.
A careers adviser wants to know where the task list on a job page came from.
She reads one row. O*NET, the U.S. Labor Department's database, version 30.0, the folder we loaded. Not a summary of it, not our reading of it, not a number we worked out. The same words the department published, and the name of the file they came out of. If she wants to check us she now knows exactly what to download.
None of these bodies has a page of ours yet. Every name on this shelf is text, and every one of them belongs to somebody else.
Our own notes still say O*NET 29.1. The folder on disk says 30.0. The folder is what we loaded, so 30.0 is what this page prints.
Made and sold
If it is a way of naming what a business makes, buys or ships, it is here.
- NAICS
- how North America names its industriesFeeds industries · NAICS 2022. Public domain.
- NAPCS
- the products those industries sellFeeds products · NAPCS 2022. Public domain.
- UNSPSC
- one code for anything a business buysFeeds products · no version stated; the download requires registration. The largest thing we hold, and the only one with no page of its own yet.
- GS1
- what a product is, where it went, and the words for bothFeeds products and logistics · product classification as of May 2025, with the November 2024 network release beside it; the event standard at 2.0.1; the web vocabulary at 1.17. GS1 General Specifications licence, free for members.
- eCl@ss
- the European classification for industrial and technical productsFeeds products · no version stated. The forty-nine segments at the top, and none of the forty thousand classes its own documentation says sit underneath them.
- ETIM
- the classification electrical wholesalers useFeeds products · no version stated. Eighty example classes out of the several thousand it publishes.
- Schema.org
- the vocabulary the web uses to describe things to machinesFeeds schema · no version stated. CC BY-SA 3.0.
A buyer is told a supplier's catalogue is GS1 ready, and does not know what that would mean.
One row tells her GS1 is three different things with three different dates: a product classification as of May 2025, an event standard at 2.0.1, and a web vocabulary at 1.17. She can now ask the supplier which one, which is the real question, and the row got her there without a single badge.
Our own superset of Schema.org is on the other shelf, under What we wrote ourselves.
Health
If a clinic, a pharmacy or an insurer has to write it on a form, it is here.
- ICD-10-CM
- every diagnosis a U.S. clinician can write downFeeds health · fiscal year 2026, which runs 1 October 2025 to 30 September 2026. Public domain.
- HCPCS Level II
- the codes for supplies and services a doctor bills forFeeds health · January 2026. Public domain, from CMS.
- NDC
- the FDA's directory of finished drug productsFeeds health and products · downloaded 5 December 2025. Ten thousand products out of the hundred and thirty-five thousand the directory holds. We took a sample, and this is it.
- NPI provider taxonomy
- the code that says what kind of clinician someone isFeeds health and occupations · version 25.1, released 1 July 2025. Public use, maintained by NUCC.
- HL7 FHIR
- the shape of a health record moving between two systemsFeeds health · R5, version 5.0.0. CC0. The resources and the datatypes.
- SNOMED CT
- clinical termsFeeds health · the top-level hierarchy only. The full set needs a licence from SNOMED International and we do not hold one.
- RxNorm
- normalised drug namesFeeds health · the monthly release, October 2025. Free to use. The term types, and none of the drugs under them.
- LOINC
- the identifiers for lab results and measurementsFeeds health · 2.81, released 12 August 2025. Its axes and its categories, and none of its codes. Use is governed by the LOINC licence agreement.
- CPT
- the codes for medical proceduresFeeds health · category information only. The codes are the AMA's and need a licence we do not hold.
A hospital analyst is checking whether the diagnosis codes behind an answer page are current.
The row says fiscal year 2026, and says what that means: 1 October 2025 to 30 September 2026. One row down she learns we hold ten thousand drug products out of a hundred and thirty-five thousand, because that is the sample we took. She now knows one of these is safe to rely on and the other is not, and she found that out in eleven seconds.
Two of these need a licence we do not hold: SNOMED CT and CPT. We hold the parts their publishers make public and nothing else, and the rows say which parts.
Money and paperwork
If it identifies who is paying, or it is the message that carries the order, it is here.
- ISO 20022
- the message one bank sends anotherFeeds finance · no version stated. The message catalogue. ISO TC68 maintains it; SWIFT holds the registry.
- LEI
- the twenty-character code that says which company this isFeeds business · no version stated. The registration authorities. GLEIF maintains it.
- ISIN
- the code for a securityFeeds finance · no version stated. The national numbering agencies. ANNA maintains it.
- MCC
- the four digits on a card payment that say what kind of shop it wasFeeds finance and business · ISO 18245:2023. ISO TC68 and the card networks maintain it.
- BIC
- the code that identifies a bankFeeds finance · ISO 9362:2014. The structure and its country codes, and not the register of banks itself. SWIFT maintains it.
- X12
- North America's electronic purchase orders and invoicesFeeds logistics and products · no version stated. Its transaction sets, its segments and its elements.
- UN/EDIFACT
- the same idea, for the rest of the worldFeeds logistics · directory D.20B. Public domain under the UN terms of use.
- EANCOM
- GS1's version of EDIFACT for consumer goodsFeeds logistics and products · no version stated. GS1 maintains it.
- Peppol
- Europe's e-invoicing network, now used well beyond EuropeFeeds logistics · no version stated. OpenPeppol maintains it.
A payments engineer wants the merchant category code list and wants to know whose it is.
The row names ISO 18245:2023, and names ISO TC68 and the card networks as the ones who maintain it. We are neither, and the row does not pretend otherwise. What we are is the place that copied it and wrote down the date. That is a smaller claim than most pages about standards make, and it is the one we can keep.
UN/EDIFACT publishes 317 message types. We hold 195 of them.
The lists everyone quotes
If the other standards on this page cite it rather than compete with it, it is here.
- ISO 3166 · 4217 · 639
- country codes, currency codes, language codesFeeds places, units and language · no edition stated. Some ISO standards are free and others must be bought.
- UN/LOCODE
- a code for every port, airport and inland place goods pass throughFeeds places and logistics · our notes cite the loc241 release. Public domain under the UN terms of use.
- UN M49
- the UN's regions of the worldFeeds places · no version stated. The regions only, and none of the countries underneath them.
- IANA time zone database
- the zones, and the rules for when the clocks changeFeeds units and places · taken from the vvo/tzdb mirror, downloaded 7 December 2025. Public domain.
- W3C
- the specifications the web is written toFeeds tech and schema · twenty-two of them, from JSON-LD and RDF to WCAG and ARIA. Per-specification versions are not stated. W3C Software and Document Licence, with MDN's licence cited for some tables.
A developer wonders why country codes are filed beside the web's specifications.
Because nearly every other row on this page cites one of them. When a shipping record names a country it is naming ISO 3166; when it names a port, UN/LOCODE; when it stamps a time, the IANA database; when it is published as linked data, a W3C specification decides what that means. This section is the bottom of the pile, and the rows above it lean on it.
Four of the twenty-two specification folders filed under W3C are not W3C's. DOM and HTML are WHATWG's, Dublin Core is DCMI's, FOAF is the FOAF project's. The folder name is a shelf label, not an attribution.
Government registers
If one government publishes it and everyone trading with that government has to use it, it is here.
- Census Bureau
- state and county codes, and the metro-area boundariesFeeds places · FIPS codes with the July 2023 metro-area delineation, downloaded 7 December 2025. Public domain.
- SEC
- the older industry list, still used on filingsFeeds industries · the four-digit SIC list, downloaded 7 December 2025. Public domain.
- USITC
- the tariff schedule for goods entering the United StatesFeeds products · downloaded 7 December 2025. Public domain. The chapters, and nothing beneath them.
- GSA
- the codes the federal government buys underFeeds products and services · Product Service Code Manual, April 2025, with the supply classes beside it. Public domain. The classes, and almost nothing beneath them.
- SBA
- what counts as a small businessFeeds business · 13 CFR 121.201. Public domain. The business and contract types, and not the size thresholds by industry.
- USPTO
- the classes a trademark or a patent is filed underFeeds law and products · Nice Classification 12th edition 2025, with Locarno, CPC and USPC beside it, downloaded 7 December 2025. Public domain. The Nice and Locarno classes are whole; the patent classifications are a handful of rows.
A contractor is trying to work out which product-service code a bid falls under.
The row tells her what we hold and, more usefully, what we do not: the April 2025 manual's classes, and almost nothing beneath them. She learns in one line that this is not the place to finish that job. A page that had let her find that out on her own would have cost her the afternoon.
Four of these six are the beginning of a register rather than the register. Each row says which.
What we wrote ourselves
If nobody outside this Foundation governs it, it is here. That is the whole test.
- AXP
- what an API has to publish so an agent can use it on first contact, with nobody reading the documentationDraft 0.3.0, dated 21 July 2026. Published by apis.ax and verified by api.qa. Home: axp.org.ai
- MDXLD
- MDX with $id, $type and $context in the frontmatterNo version number. A superset of the open MDX format, whose frontmatter is itself a superset of YAML-LD. $context is always https://schema.org.ai, with no version in it. Home: mdx.org.ai/docs and mdxld.org
- PRODUCT.md
- one Markdown file saying who a product is for, how it speaks and what it costsDraft. Its own page calls itself the one-page home of the specification and says plainly that it is a specification we write, not a standard we host. The owner's ruling records it at 0.9; the file carries no version string, so this page says draft. Home: product.md
- TODO.md
- a task list a machine can also read, in a plain Markdown fileNo version number. In use in this Foundation's own repos today. Home: todo.md.org.ai
- schema.org.ai
- our superset of Schema.org, extended for work machines doNo version number. Adds our own types on top of the ones Schema.org publishes. CC BY-SA 4.0; the vocabulary it extends is CC BY-SA 3.0. Home: schema.org.ai
- AGENTS.md
- the frontmatter an AGENTS.md file carries so one file can compile to AGENTS.md, CLAUDE.md and .cursorrulesNo version number. Home: agents.md.org.ai
A developer is deciding whether to build against AXP.
The row says draft. It says 0.3.0. It says 21 July 2026. And the section says nobody outside this Foundation has yet built anything that depends on it. Then the row links to the specification, which is live and readable today. Nothing has been oversold, and the decision is hers with the facts already in front of her.
One of the six carries a version and a status in its own file: AXP, draft 0.3.0. The other five carry no version number at all. None of them is at 1.0, and none will be until at least two people outside this Foundation have built something that depends on it. That has not happened for any of them.
How a page says where it came from
The same line sits at the bottom of every page in this estate, so you never have to ask where a fact came from — and on this shelf most of that line is blank, which is the point.
Five plain phrases, the same five on every page in this estate. On a page like this one only the first two ever fill in, because a record we copied was not built, minted or operated by anybody here.
Where a step did not happen the line shows a blank instead of a plausible name. Three of the five are blank above, and none of the three will ever fill in. A blank is information.
This is a specimen. The page it describes is not published. The row it is drawn from is in this repository today.
What this is
This is the Org.AI Foundation's standards register. One shelf holds other people's records, copied and never corrected, with the date and the version we loaded.
The other holds the handful of standards we wrote ourselves, each one saying how unfinished it is.
Where we hold a piece of something rather than all of it, the page says which.