For data architects

You decide what the platform masters and how records resolve into golden records. Almost everything else is configured against your decisions, so the order below matters.

Start here

Where most of this role's work happens.

1. The platform model — what the pipeline does
2. Entities, crosswalks, and golden records
3. Attributes and data types

Model the domain

1. Model your data — entity types and attributes
2. Manage reference data — controlled vocabularies
3. Standardize data — canonical values

Do these in order. Matching compares canonical values, so standardization is part of modelling rather than a later refinement.

Configure resolution

1. Matching and merging
2. How probabilistic scoring works
3. Create a match profile
4. Tune thresholds

Decisions that are expensive to reverse

Decision	Why it is hard to change
Cardinality	Widening later means re-examining records that already lost values
Composite or scalar	Flattened structure is not recoverable
Lookup codes	Codes are stored on records; changing one orphans them
Reference binding	Retrofitting means cleaning every existing value

Spend your time here. Thresholds are tuned continuously; a data model is not.

Next

• Install seed packs — start from a close fit rather than blank
• Data model reference

---

Last verified against commit 944761c7 (2026-08-03)