Citations_search_citations_minimal

Minimal citation search for discovery (90-95% context reduction).

Use for high-volume pattern discovery before detailed analysis. Essential 8 fields: application, publication, art unit, citation ID, category, tech center, date, examiner indicator.

Solr/Lucene Query Examples:

• Field search: criteria='groupArtUnitNumber:2854'

• Date range: criteria='officeActionDate:[2017-10-01 TO *]'

• Boolean: criteria='citationCategoryCode:X AND techCenter:2100'

• Wildcard: criteria='citedDocumentIdentifier:US*'

• Combined: criteria='groupArtUnitNumber:2854 AND officeActionDate:[2023-01-01 TO 2023-12-31]'

Ultra-minimal mode: Pass custom fields list for 99% token reduction (2-3 fields only). Example: fields=['citedDocumentIdentifier', 'patentApplicationNumber'] for PFW integration.

Date handling: USPTO documents this API as office actions mailed 2017-10-01 to ~30 days ago. In practice ~44% of TC2100 records carry an earlier officeActionDate (verified against PFW document dates back to 2010-2012). Do NOT add a blanket officeActionDate:[2017-10-01 TO *] clause unless you specifically want the documented window — it discards records the index actually serves.

Lane routing — TRY BOTH: this is the ENRICHED lane (passage locations, claim mapping, quality scores, NPL flag, date filtering). For completeness-sensitive questions also run Citations_search_oa_citations_minimal (raw 892/1449 lists, statutory basis, broader applicant-IDS coverage) and union the results — neither lane is a superset of the other. See Citations_get_guidance(section='oa_citations').

CROSS-LANE JOIN KEY: every row carries referenceKey, the normalised reference identifier, and it is the ONLY correct key for unioning this lane with the OA lane. The two lanes write the same reference differently: on app 12849948 the OA parsedReferenceIdentifier reads '20060075466' while the enriched citedDocumentIdentifier reads 'US 2006/0075466 A1'. Joining those two raw fields finds zero overlap on every application; the true answer there is four references in both lanes. referenceKey is digits only (a leading US, spaces, slashes, hyphens and the kind code stripped, series markers such as RE kept), derived from publicationNumber first and citedDocumentIdentifier second, and carried on both lanes at every tier including a custom fields list.

ROWS WITH NO REFERENCE: referenceKey is null when the row carries no usable identifier, and the response envelope reports how many such rows the page holds as rows_without_reference_identifier (always present, 0 included). An absent citedDocumentIdentifier key, a null one and an empty string are ONE state, not three: a row can carry an empty publicationNumber with the citedDocumentIdentifier key missing from the JSON entirely. Measured: 2 of 5 on app 11752072, 4 of 8 on 12849948, 4 of 26 on 18407147. Those rows are real citations and must be reported as unresolved, never dropped.

IDENTIFIERS: patent_number takes either a GRANTED patent number (7-8 digits; commas, spaces and a US prefix are accepted) or an 11-digit pre-grant publication number. A granted patent number is crosswalked to its application serial with one USPTO ODP applications-search call and queried as patentApplicationNumber; an 11-digit value queries publicationNumber directly. The response reports which reading was used in patent_number_resolution {input, interpreted_as, resolved_application_number when crosswalked, source}. A number that resolves to no application is a 400 naming the accepted forms, not a zero-result. application_number remains the application serial; passing one that disagrees with the crosswalked patent number is also a 400.

Note: Returns citation metadata only. For the office action text itself, use the PFW MCP's PFW_get_oa_text / PFW_get_oa_rejections (direct, no document-bag + OCR round trip).

For complex workflows and cross-MCP integration, use Citations_get_guidance(section). Quick reference: 'fields' section for Solr syntax, 'workflows_pfw' for PFW integration.