Data model / /

Data model & schema

The GII-US data model in three representations: the Decision Impact Record JSON Schema, the Policy Lineage Graph, and the relational mapping — a table catalog realizing the logical stores.

14Tables
10Logical stores
132Columns
intactIntegrityThe example rows validate against the catalog's own foreign keys.

The Decision Impact Record is published in three shapes: as a JSON Schema (the wire contract), as the Policy Lineage Graph (typed relationships), and — here — as a relational mapping: a normalized table catalog a reviewer can materialize as SQL and load. The catalog realizes the logical stores of the registry architecture.

Rendered live from @gii/relational — 14 tables across 10 logical stores.

Table catalog#

Raw evidence

TableWhat it holds
raw_evidenceAppend-only immutable retrieved files and API payloads, content-addressed by checksum.

Normalized source

TableWhat it holds
normalized_sourceParsed, typed, quality-checked records linked back to their raw evidence.

DIR analytic core

TableWhat it holds
dir_eventThe scoreable Government Decision Event the scorecard describes.
dir_indicatorA preregistered outcome indicator within an event, with its E/R/P/S magnitude and orientation.
dir_coverage_entryA preregistered applicable-outcome weight for coverage (GII-V), defined before results.
dir_distribution_groupA benefit or burden incidence group for distributional concentration (GII-K,).

Decision Event Registry

TableWhat it holds
decision_eventStable government-action identity/authority object; versioned, corrections retain prior vintages.

Outcome Registry

TableWhat it holds
outcome_definitionAn outcome definition: codebook family plus denominator, sources, and revision history.

Analysis Registry

TableWhat it holds
analysisA causal-analysis record: design, estimand, model, code, diagnostics, and result.

Score Registry

TableWhat it holds
scoreA versioned scorecard with its decomposition.

Publication Registry

TableWhat it holds
publicationA release, citation, correction, or deprecation.

Policy Lineage Graph

TableWhat it holds
lineage_nodeA Government Decision Event as a node in the Policy Lineage Graph.
lineage_edgeA typed, dated, sourced, reviewable edge; subject → object.

Identifier crosswalk

TableWhat it holds
crosswalk_entryMaps a preserved external identifier to a GII entity with match method and confidence.
Round-trip. The fictional example maps to dir_event 1 · dir_indicator 4 · dir_coverage_entry 4 · dir_distribution_group 2 · lineage_node 5 · lineage_edge 6, and is referentially intact true — every foreign key resolves. Versioned registry records carry a composite (id, version) key so corrections retain prior vintages; cross-references between them resolve to the latest non-retracted version at the application layer.

The emitted schema#

The catalog emits deterministic PostgreSQL — CREATE TABLE with primary keys and checks, foreign keys as ALTER TABLE so table order never matters, and COMMENT ON carrying each table’s purpose into the database itself.

The statements are not published, and neither are the column definitions above them. A complete schema alongside the outcome codebook is not documentation of an index — it is an index ready to be populated. The table names and the stores they belong to describe the shape of the store, which is what a reader needs to understand how a published value is held and versioned. Operators emit the full DDL with gii relational.

Machine-readable at GET /api/v1/relational or gii relational. The storage architecture these tables realize is described on registries.