ColumnAnalyzerOverrideEntry

org.sagebionetworks.repo.model.search.table.ColumnAnalyzerOverrideEntry

Assigns one TextAnalyzer to one column. The referenced TextAnalyzer is the single source of truth for both index-time and search-time analysis on this column: its analyzer.default entry drives OpenSearch's analyzer field mapping, and its analyzer.default_search entry (if declared) drives the search_analyzer field mapping. To express asymmetric index/search analysis (e.g. an edge n-gram chain at index time paired with a non-ngram chain at search time), declare both entries inside the TextAnalyzer's settings; do not split the configuration into two TextAnalyzer records.

Field Type Description

columnName STRING The name of the column to override. Silently skipped at index-build time if the column is not present in the target SearchIndex's schema — a single override bundle can therefore be applied across several indexes that share some column names.

semantic

BOOLEAN

Field	Type	Description
columnName	STRING	The name of the column to override. Silently skipped at index-build time if the column is not present in the target SearchIndex's schema — a single override bundle can therefore be applied across several indexes that share some column names.
semantic	BOOLEAN	When true this column also feeds the index's `semantic_search` vector field, so a neural clause can match against it. Only STRING, STRING_LIST, MEDIUMTEXT, LARGETEXT, LINK, and JSON columns may be flagged; flagging any other type, or flagging any column on a source with more than 50,000 rows, fails the index build. A JSON value is embedded as its JSON text. Independent of `analyzer` — an entry may set either, both, or `semantic` alone. When several overrides name the same column, only the first entry is used, so a `semantic` flag on a later entry for that column is ignored. Long values are truncated for embedding. All of a row's flagged columns are joined into a single text of `Name: value` lines, and that text is capped at 8,000 tokens before it is embedded. Text past the cap contributes nothing to the vector. The columns themselves are unaffected: each value is indexed in full, returned in full on a hit, and remains fully matchable by keyword clauses.
analyzer	OBJECT	The analyzer to use for this column. Either a reference to a saved TextAnalyzer (preferred — supports reuse) written as `{"$ref": "{organizationName}-{name}"}`, or an inline analyzer literal pasted directly. $ref form: `"analyzer": { "$ref": "biomed-acronym_exact" }` Inline form — the bare OpenSearch `settings.analysis` block (same shape as a TextAnalyzer record's `settings` field, with no surrounding identity / audit fields and no outer `settings` wrapper). Allowed root keys are `char_filter`, `tokenizer`, `filter`, and `analyzer`. The inner `analyzer` map must declare `default` (required), and may optionally declare `default_search` for asymmetric search-time analysis. Refs to SynonymSets are not permitted inside an inline analyzer literal — save a TextAnalyzer and reference it by qualified name to use synonyms. `"analyzer": { "analyzer": { "default": { "type": "custom", "tokenizer": "keyword", "filter": ["lowercase"] } } }` The analyzer's `analyzer.default` entry is bound at index time; if it also declares `analyzer.default_search`, that entry is bound at search time.

When true this column also feeds the index's semantic_search vector field, so a neural clause can match against it. Only STRING, STRING_LIST, MEDIUMTEXT, LARGETEXT, LINK, and JSON columns may be flagged; flagging any other type, or flagging any column on a source with more than 50,000 rows, fails the index build. A JSON value is embedded as its JSON text. Independent of analyzer — an entry may set either, both, or semantic alone. When several overrides name the same column, only the first entry is used, so a semantic flag on a later entry for that column is ignored.

Long values are truncated for embedding. All of a row's flagged columns are joined into a single text of Name: value lines, and that text is capped at 8,000 tokens before it is embedded. Text past the cap contributes nothing to the vector. The columns themselves are unaffected: each value is indexed in full, returned in full on a hit, and remains fully matchable by keyword clauses.

analyzer

OBJECT

The analyzer to use for this column. Either a reference to a saved TextAnalyzer (preferred — supports reuse) written as {"$ref": "{organizationName}-{name}"}, or an inline analyzer literal pasted directly.

$ref form:

"analyzer": { "$ref": "biomed-acronym_exact" }

Inline form — the bare OpenSearch settings.analysis block (same shape as a TextAnalyzer record's settings field, with no surrounding identity / audit fields and no outer settings wrapper). Allowed root keys are char_filter, tokenizer, filter, and analyzer. The inner analyzer map must declare default (required), and may optionally declare default_search for asymmetric search-time analysis. Refs to SynonymSets are not permitted inside an inline analyzer literal — save a TextAnalyzer and reference it by qualified name to use synonyms.

"analyzer": {
  "analyzer": {
    "default": {
      "type": "custom",
      "tokenizer": "keyword",
      "filter": ["lowercase"]
    }
  }
}

The analyzer's analyzer.default entry is bound at index time; if it also declares analyzer.default_search, that entry is bound at search time.