Principle

ER modelling for Conform

How to shape Conform so Identity, Span, and Publish stay honest — without joining an Inmon / Kimball / Vault tribe war. Conform is an enterprise assertion model; Publish may look dimensional to consumers, but must not invent grain or KPIs. The entities you draw are the canonical ∩ — shared meaning across same-type vendors / systems / sources — not a paste of any one source’s tables.

Optional industry-model lenses

Industry standards are optional bias when locking the entity list — not a wholesale copy into CORE. Use them as a pattern lens after forensic inventory, not as the programme model by decree.

Automotive / dealership programmes often use STAR BOD / DMS as a pattern lens — not a global statute like ACORD or FHIR.

Stance vs familiar schools

Why we avoid FACT_*

A Kimball fact encodes a chosen analytic grain. Conform needs native keys, tritemporal correction history, Identity lag, and first-class relationships. Calling early tables FACT_* invites collapsing history and inventing KPIs before Publish. Prefer entity names: PARTY, ACCOUNT, MONETARY_POST, …

Headers and assertion rows

Identity subjects use a sparse header plus versioned assertions:

Singletons use consume_key = header_key until Identity clusters. Anti-pattern: one dense “current row” master with no header/assertion split.

{SUBJECT}_HEADER
  {subject}_header_key   -- stable enterprise join id

{SUBJECT}                -- assertion versions
  {subject}_row_key      -- PK of this version
  {subject}_header_key   -- FK → header
  valid_* / system_* / warehouse_*
  extensions             -- source_natural_id, lane payload

Entity vs domain

• Entity — Conform table type (how you store assertions).
• Domain — business subject area (what you steward). Often appears as link_to_domain.

When to use _LINK / roles — and types

Use a link or role entity when you assert a relationship, not a second copy of the subject. Qualify with a governed type/usage and optional scope:

• link_to_domain — governed domain code (kind of target), not the entity table name.
• link_to_key — target header key (which instance).
• Prefer one generic role/link table over exploding per-domain physical duplicates when the shape matches.
• Do not put attachment on the subject row — location_usage lives on the link, not location_kind on the place.

Suffix cheat-sheet

Relationship grain stays relationship grain

Span/Publish may collapse identity subjects to one consume key. Roles and links stay at Conform relationship grain — filter by type; do not fold them into the collapsed subject facade.

Checklist

1. Name Conform tables as entities — not FACT_/DIM_.
2. Header hub + versioned assertions for each identity subject.
3. Relationships in ROLE/LINK entities with type/usage + link_to_*.
4. Prefer generic link tables when the join shape is shared.
5. Money/events are assertions; Publish names KPIs later.
6. Never collapse relationship entities into identity facades.
7. Lock the entity list (forensic + AI + human) before adaptors bind.