A real policy effect commonly results from several institutional actions — Congress authorizes a program, appropriates funds later, an agency implements it, states administer it, courts alter coverage, and a later Congress extends or repeals it. Crediting the full outcome to every document multiplies impact; crediting only the latest erases authority and implementation. The Policy Lineage Graph preserves both facts: who did what legally, and which actions contributed empirically.
Implemented in @gii/lineage · the graph below is rendered from the fictional example and its derived facts are computed live by the library.
Relationship types#
Edges are typed, dated, sourced, and reviewable. The graph distinguishes a legal relationship from an empirical causal one.
| Relation | Meaning | Default nature |
|---|---|---|
| authorizes | Supplies legal authority for another event | legal |
| funds | Provides or removes resources needed for implementation | legal |
| implements | Converts authority into operational rules, systems, or delivery | causal |
| interprets | Defines legal or administrative meaning without necessarily changing implementation | legal |
| modifies | Alters scope, obligations, timing, eligibility, or mechanism | legal |
| stays_blocks | Temporarily prevents operation or enforcement | legal |
| vacates_invalidates | Removes legal effect in a specified scope | legal |
| affirms | Sustains a challenged decision or legal interpretation | legal |
| supersedes | Replaces an earlier decision prospectively or retrospectively | legal |
| depends_on | Cannot produce the measured pathway without another event | causal |
| mediates | Transmits some effect between decision and outcome | causal |
| overlaps_with | Shares outcome, population, time, or mechanism and requires non-duplication analysis | causal |
A worked lineage#
The fictional Hypothetical National Reporting Rule: an enabling statute authorizes a final rule; the rule implements the statute; two states implement the rule; a district court partially stays it; and the two state programs overlap on shared reporting infrastructure.
Legal provenance versus causal contribution#
Legal provenance answers which institutions supplied authority, implementation, interpretation, relief, or amendment. Causal contribution answers which changes in exposure or mechanism produced the observed outcome, and by how much. GII publishes the entire institutional lineage but allocates quantitative contribution only where evidence supports it — if separate contributions cannot be identified, the impact remains at the lineage level rather than being arbitrarily divided.
Judicial decisions#
- Code the operative holding, relief, scope, precedential status, jurisdiction, mandate, and later treatment.
- Distinguish opinion date, mandate date, stay, remand, and actual implementation.
- Do not infer welfare direction from which party won or from an ideological database code.
- Represent circuit splits, limited injunctions, and temporal stays explicitly.
- Place Supreme Court decisions in the standalone Supreme Court view and once within Judiciary — the graph flags a SCOTUS citation that appears as more than one node.
Transfers and shared effects#
Fiscal transfers, fees, taxes, grants, and price changes benefit one group while burdening another. Payer and recipient sides are reconciled before aggregation: a government payment is not simultaneously counted as full household benefit, full business revenue, full public expenditure, and full macroeconomic impact unless the measures represent distinct mechanisms and non-overlapping welfare concepts. Overlapping events declare a shared fraction so the overlaps_with relationship can drive an explicit non-duplication adjustment.