Intervention Register Specification

The machine-readable contract is schemas/intervention.schema.json. The authoring template is templates/intervention.template.yaml. Entries live in data/public/interventions/, one YAML file per intervention.

Required fields

field	note
intervention_id	urn:worldfabric:intervention:<slug>
corridor_ref	the canonical corridor record it acts on
targets_bottleneck	reference into that corridor's bottleneck register
intervention_type	taxonomies/intervention_types.yaml
summary	one sentence, no adjectives
epistemic_status	fixed to scenario by the schema
proposed_change	functional link, metric, direction, magnitude band
expected_effect	corridor metric, magnitude band, reasoning
displacement	next binding constraint, and why
counter_effects	at least one
preconditions	at least one
evidence	source-registry references
evidence_sufficiency	taxonomies/evidence_sufficiency.yaml
falsification	what would show this is the wrong intervention
confidence	reuses system_states.yaml → confidence
priority_rationale	argued in words, never a score
status	taxonomies/intervention_status.yaml

review is optional and its fields accept null while undecided, consistent with the other templates in this package.

Magnitude, not measurement

proposed_change.magnitude_band and expected_effect.magnitude_band take a band from taxonomies/magnitude_bands.yaml, never a number. A single figure attached to an effect that has not been observed is invented precision, and it would be quoted as though it had been measured.

uncertain is a legitimate band and is the correct default. It is not a failure to state a band.

Status transitions

candidate → evidence_review → analytical_review → validated → implemented
        ↘ blocked_insufficient_evidence
        ↘ withdrawn
validated → contested | superseded

Two conditional rules are enforced by the schema rather than by review discipline:

• status: candidate requires evidence_sufficiency: insufficient;
• status: validated requires at least one evidence entry and an evidence_sufficiency of sufficient or sufficient_for_qualified_publication.

An author cannot promote an entry by editing one field.

Validation

scripts/validate_records.py validates every file in data/public/interventions/ against the schema, alongside the corridor records. It is also run against the template, so a schema change that breaks the template is caught immediately rather than at the next authoring session.

Rendering rule

Wherever interventions are displayed, the tier must be visible in the same view as the intervention itself — not in a footnote, not on a separate page. A candidate carries an explicit non-citable notice.

Counts of candidates are never presented as coverage or as progress. They measure how much has been hypothesised, which is not a measure of how much is known.