Skip to main content
Partitions are read-optimized views over data source data. Each partition declares:
  • Which resources and relationships are exposed — Select collections per source or per unified MDM collection.
  • Traversal rules — Define which related properties can be projected into the query context. Learn how this powers filtering in Traversed property filters.
  • Search modality — Pick between STANDARD (lexical) and HEALTH_GRADE_SEARCH (hybrid semantic) to tailor relevance.

​
Partition types

Partitions can point to two data layers:
  • Source-backed partitions deliver the quickest time-to-first-query. They expose collections straight from a single source—ideal when you need to surface one system of record without cross-source reconciliation.
  • MDM-backed partitions draw from unified collections. They require entity resolution and reconciliation to run first so that results are deduplicated and normalized across sources.
Both partition types use the same query DSL and share the same lifecycle controls, which makes it easy to mix fast paths with unified paths in the same workspace.

​
Traversed properties

Traversal rules let you bring attributes from related resources or contained structures into the current query scope. That enables filters such as “clinics that offer a related service with code cardiology” without copying data. Traversals are explicit and bounded, which keeps query plans predictable and prevents accidental fan-out.

​
Search modality

  • STANDARD uses an inverted index tailored for lexical matching, faceting, and structured filters.
  • HEALTH_GRADE_SEARCH (HGS) adds vector retrieval tuned for healthcare vocabulary while preserving lexical operators. Use it when natural-language queries must coexist with compliance filters.
Choose the modality per partition so that experimentation does not affect production workloads.

​
MDM partitions

MDM partitions reference unified collections that have already gone through entity resolution and property reconciliation:
  • Reconciliation strategies — Define how conflicting attribute values are merged (latest, earliest, aggregated, or custom logic).
  • Cross-source joins — Unified relationships let you traverse across systems without denormalizing the data first.
  • Governance — MDM policies ensure that only curated, steward-approved data feeds downstream applications.
Use MDM-backed partitions when consistency across multiple systems matters more than latency.

​
Keep exploring

Source partitions

See how single-source partitions expose raw collections with minimal transformation.

MDM partitions

Understand how unified records are reconciled before hitting downstream apps.

Traversed properties

Learn how to project adjacent resource attributes directly into partition search.
⌘I