--- pretty_name: 3char - Same Code, Different Worlds license: other license_name: mixed-per-system license_link: https://huggingface.co/datasets/brennercruvinel/3char#licensing language: - en tags: - code-systems - controlled-vocabulary - taxonomy - identifiers - three-letter-codes - knowledge-base size_categories: - 100K/`; joins to the `systems` configuration | | `code` | string | The entry as the source writes it, without case folding | | `label` | string | Name of the entry, from the source or from the label source named in the system's `notes` | | `description` | string | Longer description, when a source provides one | | `parent_code` | string | The parent the source assigns, in 17 systems; see below | | `status` | string | Status value published by the source, or null | | `aliases` | list | Other strings that refer to the same entry | | `extra` | string | JSON object with the fields particular to the source | `parent_code` usually points to another entry of the same system: an ICD-10-CM category, an ATC group, a Library of Congress class, a schema.org supertype. In three systems it names a group that has no row of its own: the EC sub-subclass in `bio/ec` (`1.1.1.-`), the top-level type in `cmp/mim` (`application`) and the ISO 3166-1 alpha-2 country in `trp/unl` (`BR`). `status` is null in 758,213 rows. Elsewhere it holds the value the source publishes: `approved` for every HGNC symbol (44,997 rows); `deprecated` in `bio/ec`, `cmp/mim`, `fin/mic`, `lng/b47`, `lng/i63` and `trp/ica2` (7,357 rows); `current` for the remaining airlines in `trp/ica2` (1,069); `unassigned` in `cmp/sts` (436); `obsolete` for withdrawn PDB entries (43); and the Unicode emoji qualification values in `wrt/uni`: `fully-qualified` (3,944), `minimally-qualified` (1,029), `unqualified` (243) and `component` (9). Each source has its own vocabulary, so filter on `status` within a single system. `extra` is a JSON string rather than a struct because the sources have little in common beyond the code and the label: the 59 systems that use it have 142 distinct keys between them, and only 27 of those keys occur in more than one system. ### Identifying a record `code` is unique within 60 of the 64 systems. In the other four, a record is identified by the fields below, as computed by `keys.py`: | System | Rows | Distinct codes | Key | Remaining duplicates | | --- | ---: | ---: | --- | --- | | `trp/unl` | 116,213 | 26,162 | `extra.locode` (country plus code) | 139 locodes occur on more than one row as bilingual variants of the same name, adding 146 rows | | `lng/b47` | 9,295 | 9,037 | `code` and `extra.type` | None | | `wrt/tz` | 777 | 769 | `code` and `extra.kind` | None; eight strings, such as `CET` and `EST`, are both a zone name and an abbreviation | | `wrt/abj` | 84 | 83 | `code` and `label` | None; the repeated code, U+05BC, appears once as `dagesh` and once as `mapiq` | Across systems, a record is identified by `(system_id, code)`, or by `(system_id, extra.locode)` in `trp/unl`. A code without its system does not identify a record. ### Loading ```python from datasets import load_dataset ds = load_dataset("brennercruvinel/3char", "all") # every system in one table, 817,340 rows iata = load_dataset("brennercruvinel/3char", "trp-iata") # a single system meta = load_dataset("brennercruvinel/3char", "systems") # 64 rows: authority, URL, license, regex, version, retrieval date ``` There are 66 configurations: `all`, `systems`, and one per system, named `-`, so the `trp-iata` configuration holds the rows whose `system_id` is `trp/iata`. Each configuration has a single `train` split; the data is reference material and has not been divided into training and evaluation sets. For reproducible work, pin a revision with `load_dataset(..., revision="")`. ## Uses - **Validating and normalizing codes.** A system's rows tell whether a string is an entry in the recorded edition of the list; its regular expression tells whether the string has the form the source defines. - **Grounding language models and search.** The lists serve as lookup tables or retrieval corpora that map codes to names and back. [Retrieval](#retrieval) shows how much the way a record is written changes what comes back. - **Studying codes across systems.** The same string often occurs in several lists, sometimes for the same entity and sometimes not, and [Results](#results) measures both. The closed study [poc-nsi](https://github.com/brennercruvinel/poc-nsi) used seven systems this way to test whether language models confuse such codes. - **Linking tables.** Country codes in `extra`, such as `extra.country` in the airport and UN/LOCODE systems and `extra.iso3` in `lng/ioc`, connect several systems to ISO 3166-1; [Similarity between systems](#similarity-between-systems) lists every such field. Two cautions. The data is a snapshot of the editions listed in [Construction](#construction): for work that depends on the current edition, such as billing with ICD-10-CM, compare `version` and `retrieved_at` with the source's latest release. And 3char is not a crosswalk: apart from the links tested in [Lists that describe the same entities](#lists-that-describe-the-same-entities), it does not state that entries of different systems denote the same thing. ## Results Every figure in this section is computed from the published files. The census, the keys, the equivalences, the similarity measures and the Wikidata comparison come from scripts in [3char-pipeline](https://github.com/brennercruvinel/3char-pipeline) (`census.py`, `keys.py`, `equivalences.py`, `similarity.py`, `crosscheck_wikidata.py`), whose outputs are kept in its `deposit/` folder. The retrieval results come from [3char-bench](https://github.com/brennercruvinel/3char-bench). ### Codes shared between systems Codes are compared after case folding, and a code is shared when the same folded string occurs in two or more systems. Of the 687,733 distinct folded codes in the 64 systems: | Codes | Count | | --- | ---: | | In exactly one system | 672,328 | | In two or more systems | 15,405 | | In three or more systems | 10,269 | | In eleven systems, the maximum | 2 (`and`, `car`) | | In two or more systems, counting the derived pair `lng/b47` and `lng/i63` once | 15,339 | Several lists are almost entirely contained in others. Every code in eight systems also occurs in at least one other: `bio/aa1`, `bio/aa3`, `bio/cdn`, `lng/i31`, `lng/i39`, `lng/i63`, `lng/mdy` and `wrt/iau`. In `trp/iata`, `lng/ioc`, `lng/i15`, `trp/ica2` and `fin/i42`, more than 98% of codes do. Without case folding there are 697,391 distinct strings, of which 21,771 occur in more than one system. Within `trp/unl`, 18,748 codes repeat because the key is the country plus the code; these repetitions are not counted as sharing between systems. A shared string is not a shared meaning. `CAR`, one of the two codes claimed by eleven systems, is among other things Caribou Municipal Airport in the IATA list, the Galibi Carib language among the language subtags, the constellation Carina, Avis Budget Group among SEC tickers, a top-level domain and the schema.org and vCard term `Car`. The census counts strings; the next two sections ask when they denote the same thing. ### Similarity between systems `similarity.py` compares every pair of systems that share at least 10 codes, 196 pairs in all. For each pair it counts the shared codes and how many of them carry the same name in both systems, after removing accents, case and punctuation. ![Share of each system's codes that another system also has](https://huggingface.co/datasets/brennercruvinel/3char/resolve/main/figures/similarity_matrix.svg) *Each cell gives the share of the row system's codes that the column system also has; pairs sharing fewer than 10 codes are left blank. Opened on its own, the SVG shows the counts of each cell on hover.* ![Pairs of systems by shared codes and by the share of those codes with the same name](https://huggingface.co/datasets/brennercruvinel/3char/resolve/main/figures/similarity_pairs.svg) *Each dot is a pair of systems. The horizontal axis counts the codes they share; the vertical axis gives the share of those codes that carry the same name in both.* Leaving out the derived pair `lng/b47` and `lng/i63`, 8 of the 195 pairs give the same name to more than half of their shared codes: schema.org with Activity Streams (31 of 31), vCard (30 of 30), SIOC (10 of 10), FOAF (14 of 15) and SHACL (13 of 14); ISO 3166-1 with the IOC list (116 of 126); and the BCP 47 registry with ISO 639-1 (152 of 183) and ISO 15924 (182 of 224). At the other end, 179 pairs name fewer than 1% of their shared codes the same way. The largest overlaps are coincidences of the three-letter space: the IATA list shares 8,999 codes with UN/LOCODE and 5,795 with the language subtags, and `AAA` is Anaa Airport in the first, Araçuaí in the second and the Ghotuo language in the third. Name matching undercounts the IATA and UN/LOCODE overlap, where one list names the airport and the other the place; the next section tests that link with its own rule. Some systems point to others explicitly. 32 fields have at least 90% of their values among the codes of another system, such as `extra.iso3` in `lng/ioc`, which resolves one to one in ISO 3166-1; the aliases of Glottolog languoids, 95.9% of which are three-letter language subtags; and `extra.country` in the airport and UN/LOCODE systems, which resolves to ISO 3166-1 alpha-2 region subtags. The full list, with coverage and cardinality, is in `deposit/similarity.json`. ### Lists that describe the same entities The same entity often appears in several lists, sometimes under the same code and sometimes under a different one. `equivalences.py` tests each such link with a stated rule. Names are compared after removing accents, case and punctuation, so two different spellings of one name do not match. ![Lists in 3char that describe the same entities](https://huggingface.co/datasets/brennercruvinel/3char/resolve/main/figures/equivalences.svg) *Solid lines: same code, same entity. Dashed lines: different codes, same entity. Dotted line: same code, entity not checked. Each label gives the entries of the first list that pass the test, out of all its entries.* **Same code, same entity.** The BCP 47 subtag registry contains most of three other lists: 152 of the 183 ISO 639-1 codes, 182 of the 226 ISO 15924 codes and 233 of the 249 ISO 3166-1 alpha-2 codes are subtags in it with the same name, and all 8,084 entries of `lng/i63` are taken from it. The IOC uses the ISO 3166-1 alpha-3 code for 116 of its 226 countries. The largest overlap is between airport codes and UN/LOCODE: 7,106 of the 9,056 IATA codes are also UN/LOCODE codes in the same country, at a location with the airport function. Names cannot confirm this match, because one list names the airport and the other the place. **Different codes, same entity.** The one- and three-letter amino acid codes map one to one (20 of 20). Through `extra.iso3`, 94 IOC codes point to a different ISO 3166-1 code with the same name, as `GER` does to `DEU`. In Glottolog, 6,716 glottocodes carry an ISO 639-3 code that is a three-letter subtag with the same name. Of the IATA airports, 7,725 match exactly one ICAO airport with the same name, municipality and country. And 25 characters have both an ITA2 combination and a Morse signal. The 244 ISO 3166-1 alpha-2 codes that are also top-level domains are drawn but not counted as the same entity, because the TLD list names the registry that manages each domain, not the country. Across the whole collection, counting the derived pair once, 501 of the 15,339 shared codes carry the same name in two systems. Name matching misses equivalences whose names differ, the IATA and UN/LOCODE overlap among them, so this count is a lower bound on shared meaning, not an estimate of it. ### Shared strings by code length For codes made only of ASCII letters, the table compares the share of claimed strings that more than one system claims with the share expected if each system drew its codes uniformly at random from the 26^L strings of its length. | Length | Possible strings | Claimed | In two or more systems | Observed share | Uniform null | Ratio | | --- | ---: | ---: | ---: | ---: | ---: | ---: | | 1 | 26 | 26 | 25 | 0.962 | 0.999 | 0.96 | | 2 | 676 | 521 | 394 | 0.756 | 0.751 | 1.01 | | 3 | 17,576 | 17,151 | 12,703 | 0.741 | 0.919 | 0.81 | | 4 | 456,976 | 19,560 | 853 | 0.044 | 0.015 | 2.9 | | 5 | 11,881,376 | 4,476 | 65 | 0.015 | 0.0001 | 121 | The three-letter space is nearly full: 97.6% of all possible strings are claimed by at least one system, and the observed share of sharing is below the uniform null. From four letters on, the observed share is 2.9 and 121 times the null. The null ignores letter frequencies and word-likeness, and no null matched to those properties has been computed, so the ratios measure departure from a uniform draw and nothing more. ### Agreement with Wikidata format constraints Thirteen systems correspond to a Wikidata external-identifier property with a format constraint (P1793). Every code in these systems was tested for a full match against the property's regular expression, as retrieved on 2026-10-13, in the form the property expects: ICD-10-CM codes with the dot (the CDC file omits it), complete UN/LOCODEs, and, for BCP 47, language subtags only, because P305 describes complete tags. | System | Property | Tested | Rejected | Cause, verified in the source file | | --- | --- | ---: | ---: | --- | | `lng/i39`, `lng/i63`, `lng/ioc`, `bio/pdb`, `fin/i42`, `lng/i15`, `wrt/tz`, `cls/loc` | P218, P220, P984, P638, P498, P506, P6687, P1149 | All | 0 | None | | `trp/unl` | P1937 | 116,213 | 0 | None for the complete locode; the three-character code alone cannot match the pattern | | `cls/icd` | P4229 | 98,403 | 628 | 600 FY2027 codes have a letter in the third position (`C4A`, `D3A`, `M1A`) and 28 belong to the new `QA0` category; the expression requires digits in the second and third positions | | `lng/glt` | P1394 | 27,177 | 3 | `b10b1234`, `3adt1234` and `ww2p1234` have a digit in the first block; the expression requires four letters | | `bio/gen` | P353 | 44,997 | 14 | Approved symbols containing `_` (`C4A_2`) or readthrough segments longer than seven characters (`HSPB2-C11orf52`) | | `lng/b47` | P305 | 8,275 | 1 | The private-use range `qaa..qtz`, a range entry rather than a subtag | In each divergence, the rejected codes are in the authority's current file and the Wikidata constraint does not admit them. This compares two snapshots; it is not an assessment of Wikidata's accuracy in general. ### Retrieval [3char-bench](https://github.com/brennercruvinel/3char-bench) asks how a code list should be written into a dense-retrieval index for the right row to come back. It uses ten systems (2,123 records, seed 20260623) and 51 questions in Portuguese, each with one answer record. - **One record per chunk.** Written as a sentence, records reach 92.2% recall@1 with multilingual-e5-small, at 22.6 tokens per answer. The same fields as `key=value` pairs reach 60.8%, and blocks of 20 records 47.1%. With all-MiniLM-L6-v2, an English model, the best record-level result is 52.9%. - **One system per chunk.** The right list ranks first for 82.4% to 90.2% of questions, but a reader must still find the row among 6,000 to 7,500 tokens. - **Bare shared codes.** A code shared by several sampled systems retrieves a record of the intended system in 33.0% of 182 cases, the rate of a random choice among those systems. With the system name in both the chunk and the query, it does so in all cases. With 51 questions, one question is worth 1.96 points of recall, and the results describe this sample and these two models only. The ten systems are unchanged in this release, so the results stand. ## Construction The data is produced by [3char-pipeline](https://github.com/brennercruvinel/3char-pipeline), which downloads each source, parses it into the common schema and checks the result. Codes, descriptions and aliases are those of each source; no annotation was added, and the curated layer is the `systems` table. A source is included when three conditions hold, each recorded in a column or a file of the pipeline: | Condition | Recorded in | | --- | --- | | The body that maintains the list is identified, and the file comes from that body or from a derivative of it | `systems.authority` and `systems.url` | | A regular expression written against the actual file matches every published entry | `systems.code_pattern`, enforced by `validate.py`; `pattern_status` is `derived` for all 64 systems | | The SHA-256 of every file read is recorded | `CHECKSUMS.tsv` | The license and its status are recorded for every system and published with it (see [Licensing](#licensing)). For the two community conventions, the maintaining project is the one named in `systems.authority`. **Labels.** When the main source lists bare codes, the label comes from a second official source, named with its URL and retrieval date in the system's `notes` and checksummed like the main file: wwPDB and RCSB entry titles for `bio/pdb`, the IANA Root Zone Database for `trp/tld`, the NAME line of each manual page for `cmp/git`, the W3C Activity Streams and ActivityPub recommendations for `grf/act`, OpenID Connect Discovery and the IANA OAuth metadata registry for `grf/oid`, RFC 6742 and RFC 7208 for nine `cmp/dns` types, shared-mime-info, GitHub Linguist and mime-db for `cmp/ext`, and the SEC CIK lookup file for `fin/cus`. In `cmp/reg` the label is the LLVM register class of each register, in `wrt/tz` the name tzdb's `theory.html` gives each abbreviation or the UTC offset it stands for, and in `grf/gql` and `grf/lpg` the grammar rules that use each keyword. Every build verifies each file against its recorded checksum and tests each entry against its system's regular expression, and either check stops the build on a mismatch. Time zone abbreviations are those the reference compiler `zic` writes for the default tzdb build. The main files of 54 systems were retrieved by 2026-06-16, and those of the other ten (`bio/aa1`, `bio/aa3`, `cls/icd`, `cmp/ern`, `cmp/reg`, `com/bau`, `com/mor`, `com/tlg`, `grf/sio`, `lng/mdy`) on 2026-09-10; the label sources were retrieved on 2026-10-14. Ten systems come from sources that publish a version string, stored in `systems.version`: ENZYME 28-Jan-2026, ICD-10-CM FY2027, IANA media types 2026-06-05, the ISO 4217 list of 2026-01-01, the IANA Language Subtag Registry of 2026-05-05 (used by both `lng/b47` and `lng/i63`), Open English WordNet 2024, the IANA TLD list 2026060600, tzdb 2026b and Unicode emoji 17.0. For the other 54, `retrieved_at` is the only date. Updates are made by re-running the pipeline, which detects any change in a source through its checksum. ## Limitations Labels come from each source and are not reconciled across systems. The same airport is `São Paulo/Guarulhos–Governor André Franco Montoro International Airport` in `trp/iata` and `trp/icao`, and `Guarulhos Apt / São Paulo`, under `BRGRU`, in `trp/unl`. Airports have no city-name alias. Where a label comes from a second source, it is that source's name for the entry, which is not always the name the main source would use: `fin/cus` carries the name EDGAR currently registers for each CIK, which may differ from the issuer's name when the CUSIP was mapped; `trp/tld` names the organization that manages each domain; and `cmp/ext` records in `extra.label_source` which of its four sources gave each name. Coverage follows what sources publish. Standards bodies that publish in English and computing registries are well represented; national schemes outside the United States and the UN system are not yet included. The census describes this collection, so adding a system changes it. Some systems need care when read: - `trp/unl`: `code` is unique only within a country, and the complete locode is in `extra.locode`. 139 locodes occur on more than one row, as bilingual variants of the same place name. - `lng/b47`: the IANA registry lists subtags, not complete language tags. Seven subtag types are distinguished in `extra.type`, and a code is unique only within its type. - `lng/i63`: the three-letter language subtags published by IANA, not ISO 639-3. RFC 5646 omits the three-letter form when a two-letter subtag exists, so `eng`, `deu` and `por` are absent, and the list includes 115 ISO 639-5 collections and 224 deprecated subtags. - `cls/icd`: codes are stored without the dot, as in the CDC order file. In the FY2027 edition, 628 codes have a letter in the second or third position. 11,153 codes that use the placeholder `X` have a `parent_code` that is not itself a code in the file, such as `E0837X`. - `trp/iata`, `trp/icao`: derived from the public-domain OurAirports file, not from IATA or ICAO publications. `trp/ica2` is named after its OpenFlights source, which lists IATA and ICAO airline codes; it contains 5,840 three-character ICAO designators and one two-character row. - `com/tlg`: parsed from the DjVu XML of a scanned 1901 book, with the columns found on each page. The 15,672 code words come from the pages of the code proper; the front matter, the index and the number tables are left out. Code words keep their OCR errors, and 11,856 carry the OCR of their code number in `extra.code_number_ocr`. - `wrt/tz`: 599 zone names and 178 abbreviations. The abbreviations are those of the default build, without the historical zones in `backzone`. - `cmp/sts`: 500 rows for 64 registered status codes, because IANA's unassigned ranges are expanded to one row per code, labelled `Unassigned`, with the range IANA writes in `extra.range` when it gives one. - `trp/olc`: the test vectors of the reference encoder, not a registry of assigned codes; the label is the coordinate each code encodes. - `lng/ioc`: ISO 3166 alpha-3 codes are in `extra.iso3` and FIFA codes in `extra.fifa`; `aliases` holds only codes that differ from the IOC code. ## Licensing The dataset has no single license. Each row of the `systems` configuration records `license` and `license_status`. The `license` column takes 28 distinct values: an SPDX identifier where one applies, and otherwise a short tag defined in `LICENSES.md`. Of the 64 published systems, 56 have the status `ok`, 2 have the status `check` and 6 have the status `problem`. - **Pending clearance (`problem`).** `cls/atc`, `cls/bsc`, `cls/dew`, `fin/cus`, `fin/mic` and `trp/w3w` are published with `license_status = problem`: their authorities' published terms restrict redistribution, and permission has been requested from each. - **No stated terms (`check`).** `fin/cfi` and `fin/i42` come from SIX Group, which publishes both lists free of charge but states no terms for redistribution. They are included because the codes are plain facts in worldwide use; this is the curator's judgment, not a grant of rights. Anyone who redistributes them commercially should first obtain SIX Group's terms. - **Copyleft and share-alike (seven systems).** `cmp/frc` (GPL-2.0-or-later) and `cmp/git` (GPL-2.0-only) are lists extracted from GPL source trees; `cmp/vim` is under the Vim license; `grf/sch` and `grf/sio` are under CC-BY-SA-3.0; `lng/i31` is under CC-BY-SA-4.0, the license of its mirror; and `trp/ica2` is under ODbL-1.0. Their attribution requirements apply to any redistribution, and the share-alike terms also apply to redistributed copies of the tables. - **ISO code lists.** `lng/i31` and `lng/i39` have the status `ok` on the basis of their mirrors' licenses. ISO asserts copyright over its code lists, and a mirror's license does not change the upstream terms. - **Not included.** SIL's ISO 639-3 table is not part of the collection; `lng/i63` is a different list, the three-letter subtags of the IANA registry. Only Parquet files are published. Raw source files are downloaded by the pipeline and are not stored in either repository. ## Related resources | Resource | Contents | Relation to 3char | | --- | --- | --- | | [Bioregistry](https://bioregistry.io) (2,936 prefixes on 2026-10-14) | A registry of identifier systems, mostly biomedical, with prefixes, regular expressions and resolvers | Describes systems without redistributing their contents; 3char distributes the lists, across ten domains | | [N2T](https://n2t.net) compact identifier prefixes | Resolver prefixes shared with identifiers.org | As for the Bioregistry. Four 3char IDs (`elm`, `icd`, `mim`, `pdb`) are also N2T prefixes, and two of them denote other systems there: `elm` is the ELM database and `mim` is OMIM | | Wikidata external-identifier properties (9,846 with a format constraint, 2026-10-13) | One property per system, with a regular expression and identifiers attached to entities | Identifiers are statements about entities rather than the authority's list; the two are compared in [Agreement with Wikidata format constraints](#agreement-with-wikidata-format-constraints) | | pycountry 26.2.16 | ISO 3166-1 (249), ISO 3166-2 (5,046), ISO 4217 (178), ISO 639 (7,923) and ISO 15924 (226), as a Python package built from the Debian iso-codes files | Covers five ISO standards; 3char takes ISO 3166-1, 4217, 639-1 and 15924 from other sources and adds 60 other systems | | `datasets/country-codes`, OurAirports, OpenFlights, IANA registries | Source files for single domains | Sources of individual 3char systems, cited in `systems.url` | ## All 64 systems The reference inventory, grouped by domain. Kinds are those described in [The data](#the-data); licenses, versions, retrieval dates, label sources and curator notes are in the `systems` configuration. ### Biology (`bio`) | ID | System | Kind | Authority | Codes | License status | Pattern | | --- | --- | --- | --- | ---: | --- | --- | | `bio/aa1` | Amino acid 1-letter codes | Symbol | IUPAC-IUB JCBN | 20 | `ok` | `^[A-Z]$` | | `bio/aa3` | Amino acid 3-letter codes | Symbol | IUPAC-IUB JCBN | 20 | `ok` | `^[A-Z][a-z]{2}$` | | `bio/cdn` | NCBI genetic codes (codons) | Symbol | NCBI | 64 | `ok` | `^[ACGTU]{3}$` | | `bio/ec` | Enzyme commission numbers | Assigned code | IUBMB / ExPASy | 8,441 | `ok` | `^\d+\.\d+\.\d+\.(n?\d+\|-)$` | | `bio/gen` | HGNC gene symbols | Assigned code | HUGO Gene Nomenclature Committee | 44,997 | `ok` | `^[A-Za-z0-9@#\-\._]+$` | | `bio/pdb` | PDB entry IDs | Assigned code | wwPDB / RCSB | 254,978 | `ok` | `^[1-9][A-Z0-9]{3}$` | ### Classification (`cls`) | ID | System | Kind | Authority | Codes | License status | Pattern | | --- | --- | --- | --- | ---: | --- | --- | | `cls/atc` | WHO ATC drug classification | Assigned code | WHO Collaborating Centre for Drug Statistics Methodology | 6,440 | `problem` | `^[A-V](\d{2}([A-Z]([A-Z](\d{2})?)?)?)?$` | | `cls/bsc` | BISAC subject headings | Assigned code | Book Industry Study Group | 3,887 | `problem` | `^[A-Z]{3}\d{6}$` | | `cls/dew` | Dewey decimal summaries | Assigned code | OCLC | 525 | `problem` | `^\d{3}$` | | `cls/elm` | Chemical element symbols | Symbol | IUPAC | 118 | `ok` | `^[A-Z][a-z]{0,2}$` | | `cls/icd` | ICD-10-CM | Assigned code | US CDC / NCHS | 98,403 | `ok` | `^[A-Z][A-Z0-9]{2,6}$` | | `cls/loc` | Library of Congress classification outline | Assigned code | Library of Congress | 234 | `ok` | `^[A-Z]{1,3}$` | ### Computing (`cmp`) | ID | System | Kind | Authority | Codes | License status | Pattern | | --- | --- | --- | --- | ---: | --- | --- | | `cmp/asm` | x86 instruction mnemonics | Vocabulary term | Intel / AMD | 1,309 | `ok` | `^[A-Z][A-Za-z0-9]{1,15}$` | | `cmp/css` | CSS functions | Vocabulary term | W3C CSSWG / MDN | 105 | `ok` | `^[a-zA-Z][a-zA-Z0-9\-]*$` | | `cmp/dns` | DNS resource record types | Assigned code | IANA | 97 | `ok` | `^([A-Z][A-Z0-9\-]*\|\*)$` | | `cmp/ern` | POSIX errno names | Assigned code | IEEE / The Open Group | 77 | `ok` | `^E[A-Z0-9]+$` | | `cmp/ext` | File extensions | Community convention | Community | 286 | `ok` | `^[a-z0-9][a-z0-9.+\-]*$` | | `cmp/frc` | FourCC codec identifiers | Community convention | Community / VLC | 672 | `ok` | `^[A-Za-z0-9\-.:_]{2,4}$` | | `cmp/git` | Git subcommands | Vocabulary term | Git project | 158 | `ok` | `^[a-z][a-z0-9\-]{1,17}$` | | `cmp/hex` | CSS named colors | Vocabulary term | W3C CSSWG | 148 | `ok` | `^[a-z]{3,20}$` | | `cmp/htt` | HTTP methods | Assigned code | IANA | 41 | `ok` | `^([A-Z][A-Z\-]*\|\*)$` | | `cmp/mim` | Media types | Assigned code | IANA | 2,300 | `ok` | `^[a-z]+/[a-zA-Z0-9!#$&\-\^_\.\+]+$` | | `cmp/reg` | x86-64 register names | Vocabulary term | LLVM Project | 257 | `ok` | `^[a-z][a-z0-9.()]*$` | | `cmp/sts` | HTTP status codes | Assigned code | IANA | 500 | `ok` | `^\d{3}$` | | `cmp/vim` | Vim motion commands | Vocabulary term | Vim project | 289 | `ok` | `^\S{1,40}$` | ### Communication (`com`) | ID | System | Kind | Authority | Codes | License status | Pattern | | --- | --- | --- | --- | ---: | --- | --- | | `com/bau` | Baudot / ITA2 teleprinter codes | Symbol | ITU-T | 30 | `ok` | `^[A-Z0-9\-]+$` | | `com/mor` | Morse code | Symbol | ITU-R | 48 | `ok` | `^([A-Z0-9\u00c9]\|[^\w\s]{1,2})$` | | `com/tlg` | ABC universal commercial telegraphic code | Symbol | William Clauson-Thue (1901) | 15,672 | `ok` | `^[A-Za-z]{4,22}$` | ### Finance (`fin`) | ID | System | Kind | Authority | Codes | License status | Pattern | | --- | --- | --- | --- | ---: | --- | --- | | `fin/cfi` | ISO 10962 CFI codes | Assigned code | SIX Group / ISO | 1,216 | `check` | `^[A-Z]{6}$` | | `fin/cus` | CIK to CUSIP mapping | Assigned code | CUSIP Global Services / SEC | 43,485 | `problem` | `^[0-9A-Za-z]{6}$` | | `fin/i42` | ISO 4217 currency codes | Assigned code | SIX Group / ISO | 178 | `check` | `^[A-Z]{3}$` | | `fin/mic` | ISO 10383 market identifier codes | Assigned code | ISO 20022 RA / SWIFT | 2,853 | `problem` | `^[A-Z0-9]{4}$` | | `fin/tck` | SEC company tickers | Assigned code | US SEC | 10,405 | `ok` | `^[A-Z][A-Z\-]{0,6}$` | ### Graph and vocabulary (`grf`) | ID | System | Kind | Authority | Codes | License status | Pattern | | --- | --- | --- | --- | ---: | --- | --- | | `grf/act` | ActivityStreams 2.0 terms | Vocabulary term | W3C | 141 | `ok` | `^[A-Za-z][A-Za-z0-9]*$` | | `grf/foa` | FOAF vocabulary terms | Vocabulary term | FOAF project | 75 | `ok` | `^[A-Za-z][A-Za-z0-9_]*$` | | `grf/gql` | openCypher keywords | Vocabulary term | openCypher / Neo4j | 61 | `ok` | `^[A-Z][A-Z_]*$` | | `grf/lpg` | Property graph schema terms | Vocabulary term | arXiv 2211.10962 | 26 | `ok` | `^[A-Z]{2,9}$` | | `grf/oid` | OpenID Connect discovery fields | Vocabulary term | OpenID Foundation | 17 | `ok` | `^[a-z][a-z0-9_]*$` | | `grf/org` | W3C Organization ontology terms | Vocabulary term | W3C | 45 | `ok` | `^[a-zA-Z][A-Za-z0-9]*$` | | `grf/rdf` | RDF Schema terms | Vocabulary term | W3C | 15 | `ok` | `^[a-zA-Z][A-Za-z0-9]*$` | | `grf/sch` | schema.org types and properties | Vocabulary term | schema.org / W3C CG | 2,987 | `ok` | `^[A-Za-z0-9][A-Za-z0-9]*$` | | `grf/shc` | SHACL vocabulary terms | Vocabulary term | W3C | 221 | `ok` | `^[A-Za-z][A-Za-z0-9\-]*$` | | `grf/sio` | SIOC vocabulary terms | Vocabulary term | SIOC Project / DERI | 99 | `ok` | `^[A-Za-z][A-Za-z0-9_]*$` | | `grf/vcd` | vCard ontology terms | Vocabulary term | W3C | 146 | `ok` | `^[a-zA-Z][A-Za-z0-9\-]*$` | ### Language (`lng`) | ID | System | Kind | Authority | Codes | License status | Pattern | | --- | --- | --- | --- | ---: | --- | --- | | `lng/b47` | BCP 47 language subtags | Assigned code | IANA | 9,295 | `ok` | `^[A-Za-z0-9\-]+(\.\.[A-Za-z0-9\-]+)?$` | | `lng/glt` | Glottolog languoid codes | Assigned code | MPI-EVA Leipzig | 27,177 | `ok` | `^[a-z0-9]{4}\d{4}$` | | `lng/i15` | ISO 15924 script codes | Assigned code | Unicode Consortium (RA) | 226 | `ok` | `^[A-Z][a-z]{3}$` | | `lng/i31` | ISO 3166-1 country codes | Assigned code | ISO 3166/MA | 249 | `ok` | `^[A-Z]{2,3}$` | | `lng/i39` | ISO 639-1 language codes | Assigned code | ISO 639/RA | 183 | `ok` | `^[a-z]{2}$` | | `lng/i63` | BCP 47 three-letter language subtags | Assigned code | IANA | 8,084 | `ok` | `^[a-z]{3}$` | | `lng/ioc` | Country codes composite | Assigned code | Community / datasets.io | 226 | `ok` | `^[A-Z]{2,3}$` | | `lng/mdy` | Month names | Vocabulary term | Unicode CLDR | 12 | `ok` | `^[A-Z][a-z]{2}$` | ### Lexical (`nte`) | ID | System | Kind | Authority | Codes | License status | Pattern | | --- | --- | --- | --- | ---: | --- | --- | | `nte/wnt` | English WordNet 2024 synset IDs | Assigned code | Global WordNet Association | 120,630 | `ok` | `^oewn-\d{8}-[nvasr]$` | ### Transport and place (`trp`) | ID | System | Kind | Authority | Codes | License status | Pattern | | --- | --- | --- | --- | ---: | --- | --- | | `trp/iata` | IATA airport codes | Assigned code | IATA | 9,056 | `ok` | `^[A-Z]{3}$` | | `trp/ica2` | IATA and ICAO airline codes | Assigned code | IATA / ICAO | 5,841 | `ok` | `^[A-Z0-9]{2,3}$` | | `trp/icao` | ICAO airport codes | Assigned code | ICAO | 10,153 | `ok` | `^[A-Z]{4}$` | | `trp/olc` | Open Location Code (plus codes) | Test vector | Google | 292 | `ok` | `^[023456789CFGHJMPQRVWX]{4,}\+[23456789CFGHJMPQRVWX]*$` | | `trp/tld` | Top level domains | Assigned code | IANA | 1,437 | `ok` | `^[A-Z0-9\-]+$` | | `trp/unl` | UN/LOCODE location codes | Assigned code | UNECE | 116,213 | `ok` | `^[A-Z0-9]{3}$` | | `trp/w3w` | what3words API surface | Vocabulary term | what3words | 6 | `problem` | `^[a-zA-Z][a-zA-Z0-9_]*$` | ### Writing and time (`wrt`) | ID | System | Kind | Authority | Codes | License status | Pattern | | --- | --- | --- | --- | ---: | --- | --- | | `wrt/abj` | Hebrew script characters | Symbol | Unicode / community | 84 | `ok` | `^\S$` | | `wrt/iau` | IAU constellation abbreviations | Assigned code | International Astronomical Union | 88 | `ok` | `^[A-Z][A-Za-z]{2}$` | | `wrt/tz` | tz database zone names and abbreviations | Assigned code | IANA / tzdb | 777 | `ok` | `^[A-Za-z0-9+\-_/]{1,32}$` | | `wrt/uni` | Unicode emoji sequences | Symbol | Unicode Consortium | 5,225 | `ok` | `^[0-9A-F]{4,6}( [0-9A-F]{4,6})*$` | ## Versions and citation Releases are numbered `MAJOR.MINOR.PATCH`: a change to the schema increments the major version, a change to any Parquet file the minor version, and a change limited to the documentation the patch version. This card describes 0.2.0. ```bibtex @misc{cruvinel_3char_2026, title = {3char: 64 code systems and controlled vocabularies in one schema, with per-system provenance and license status}, author = {Cruvinel, Brenner}, year = {2026}, version = {0.2.0}, url = {https://huggingface.co/datasets/brennercruvinel/3char} } ```