Canonmap maps inconsistent labels and short descriptions to stable concepts in a taxonomy you choose. It also shows where the taxonomy is ambiguous, redundant, or missing useful structure.
Built for vocabularies with thousands of possible concepts, where exact matching and one-prompt classification stop being dependable.
1,012 conceptsCustom software development services for business clients
541511Custom Computer Programming ServicesThe description refers to software built for individual clients.
513210Software PublishersNot selected because the description does not indicate a standard software product sold to many customers.
Databases group categorical values by their stored representation. Unless a mapping has been defined, custom app development, bespoke software consulting, and contract software engineering are three different values.
That fragments frequency counts, changes rankings, prevents reliable joins, and leaves downstream rules dependent on spelling. A canonical concept gives equivalent observations a shared identifier while preserving the original text.
541511522110236115The raw table describes wording. The canonical table describes the concepts the field was intended to measure.
Large taxonomies contain close neighbors, uneven hierarchies, outdated names, and concepts that differ by one operational detail. Some inputs are incomplete or describe something the taxonomy does not contain.
Similarity can retrieve a useful neighborhood. Canonicalization must still determine whether any candidate represents the same concept.
Upload the concepts your data is expected to use. Canonmap validates the taxonomy, makes an initial mapping build available, and continues improving difficult decisions behind the same versioned API.
A label is sufficient to start. Existing IDs, definitions, aliases, parent concepts, and examples improve the source material when available. Canonmap retains source IDs or creates stable IDs when they are absent.
Source taxonomy -> validated concepts and stable identifiersCanonmap makes an initial mapping build available after validation and indexing. The first build uses the same response contract as later versions, so evaluation can begin while managed improvement continues.
Validated taxonomy -> initial versioned mapping buildEvery result is either a match or ambiguous. A match returns a stable concept from the selected taxonomy; an ambiguous result preserves the source value and plausible candidates without forcing a label.
Source value -> match or ambiguousCanonmap evaluates difficult cases and publishes qualified changes as new immutable builds. Separate recommendations can identify overlapping labels, reporting groups, and useful subcategories without silently changing the source taxonomy.
Evaluation evidence -> qualified, reversible buildA taxonomy defines the valid answers and the distinctions that matter. Change the taxonomy, and the meaning of the mapping changes with it.
1,012 six-digit industries
An establishment and its primary economic activityClassifies establishments by their primary economic activity. Use it for local industry labels and descriptions of what a business location does.
"custom software development services for business clients"illustrative targetCustom Computer Programming Services
2,344 active codes
The predominant product or service in a contract actionClassifies the predominant product or service in a federal contract action. Use it for procurement descriptions, statement-of-work titles, and local spend categories.
"IT telecom application development support labor"illustrative targetIT and Telecom - Application Development Services
4,516 research topics
The primary research topic of a scholarly workClassifies what a scholarly work is about. Use it for titles, abstracts, grant descriptions, and local research-area labels.
"artificial intelligence for healthcare diagnosis"illustrative targetArtificial Intelligence in Healthcare
Language models are useful when a distinction depends on context. They are less practical as the whole mapping system: repeatedly sending thousands of concepts is expensive and slow, smaller models struggle with large choice sets, and an unconstrained answer may not be a valid concept.
Canonmap prepares the taxonomy once, retrieves a small candidate set, constrains the decision to a specific version, and allows uncertain inputs to remain ambiguous.
Product and data teams can inspect concepts, candidates, ambiguous results, and evaluation evidence. Engineers receive the same mapping contract through code.
bashcurl "$CANONMAP_API_URL/v1/taxonomies/naics/normalize" \
-H "Content-Type: application/json" \
-d '{
"value": "custom software development services for business clients"
}'json{
"taxonomy_id": "naics",
"taxonomy_version": "2022",
"source_value": "custom software development services for business clients",
"decision": "match",
"concept": {
"id": "541511",
"label": "Custom Computer Programming Services"
},
"mapping_build": "<immutable-build-id>"
}Inspect a public example, test it against real values, and decide whether the mapping contract belongs in a production data path.