# Choose an architectural fact

Distinguish classification, Context, Relation and Contribution before choosing an emission.

One provider attribute can support several different architectural claims.
Choose the meaning first, then prove its target. A dependency tells Terraform
about ordering; it does not choose an architectural fact for Rootform.

## Classification is not a connection

A Concept answers what an instance represents. `as = rf.concept.subnet`
classifies it without establishing any network Context. A Rule may emit facts
without assigning a Concept, and an instance without a Rule remains represented.
Use an RF Vocabulary Concept only when its
[complete meaning](https://docs.rootform.dev/language/reference/rf-vocabulary/) fits. Define provider-specific
Concepts locally when it does not.

## Choose the claim and direction

| Need | Fact | Official example |
| --- | --- | --- |
| Record placement in a named dimension | Context, source to placement target | AWS subnet to VPC through `rf.context.network` |
| State a directed domain predicate | Relation, source to related target | AWS SNS subscription to topic through `aws.relation.subscribes-to` |
| Record an instance contributing to another | Contribution, contributor to beneficiary | AWS bucket versioning to its bucket |

Network placement does not prove reachability. A subscription Relation does
not prove messages were delivered. A versioning Contribution records the
configuration instance's support for its bucket; it does not absorb that
instance or prove all storage protections.

Context and Relation emissions either reuse a symbol through `as`, or introduce
or reuse a local symbol through a block label. Contribution has no named
predicate. Local definitions can document a symbol but do not emit facts.
[Fact emissions](https://docs.rootform.dev/language/reference/emissions/) owns the syntax.

## Prove the endpoint separately

Every emission declares `to`, `via`, `on_null` and `on_empty`.
`to` is a semantic requirement, such as a Concept or a specific Rule.
`via` names the source evidence path. It is not a target selector by itself.
Use an explicit `match` for value comparison, or a verified saved-plan reference
for direct endpoint resolution.

Several facts may use the same path when their declared meanings differ.
Neither a shared dependency nor similar resource names justify adding them.
If no emission exists, no closure exists for that claim: missing coverage is
not a declaration of absence.

Read [Evidence and target resolution](https://docs.rootform.dev/language/learn/evidence-targets/) before choosing
identity paths or external policy. [Policies over facts](https://docs.rootform.dev/language/learn/policies/) explains
why Contexts and Relations are queried from their source, while Contributions
are queried from their beneficiary.