Data modeling at a glance

What you define

Concept	Is
Entity type	A kind of thing you master
Attribute	A named piece of data on that type
Relationship type	A kind of link between two types

The four attribute decisions

Made in this order, because each constrains the next.

Decision	Options	Expensive to reverse
Data type	Scalar or composite	Yes — flattened structure is unrecoverable
Cardinality	Single or multiple	Yes — widening means re-examining records that lost values
Required	Yes or no	No
Reference binding	Bound to a lookup type, or free	Yes — retrofitting means cleaning every value

Three of four are expensive. Spend your time here.

Single versus multi-value

Single-value	Multi-value
One meaningful value — a legal name, a birth date	Several — emails, phones, addresses
Compared atomically when matching	Every combination compared, best pairing taken
One survivorship winner	Optionally every distinct value preserved

A multi-value attribute can bind a usage type (home, work, mobile). Survivorship then resolves each usage independently.

Composites

An address is named parts, not a string. Declaring it composite is what lets matching compare postal codes exactly while comparing street lines loosely.

Comparing two whole composites with a scalar comparator is refused — two unrelated addresses would otherwise compare equal.

Reference binding

The code is stored; the label is resolved when reading. Relabelling does not rewrite stored data, and search matches the resolved label as well as the code.

Order of work

1. Entity types and attributes
2. Reference data, and bind attributes to it
3. Standardization rules
4. Match profiles
5. Load records

Each depends on the one before. Matching unstandardized values produces false negatives.

Next

• Define an entity type
• Data model reference

---

Last verified against commit 1e1de1ad (2026-08-03)