feat: decouple intent/code from core so braid is a pure framework (ontology-declared SourceRoles)

[mroops0111]

Summary

braid is meant to be a pure framework where the ontology is a plugin (DDD is just the default). But the intent / code worldview is still hard-coded into core, which blocks any non-software ontology (e.g. a story-world ontology, a research-notes ontology) from being a first-class citizen.

This issue tracks decoupling intent/code from core so braid is genuinely "braid anything".

Evidence (where the leak is)

The single root of the leak:

• packages/schema/src/source.ts:5 — export const SourceRole = z.enum(['code', 'intent'])
  A core schema type pins the source taxonomy to exactly these two values.

Two downstream dependents build on that binary:

• packages/core/src/application/BatchService.ts — mode === 'intent' branch, buildIntentUnits(), and error strings like "has no intent or code sources to bootstrap from". Core knows about two privileged source kinds.
• packages/core/src/infrastructure/validation/EvidenceValidator.ts — encodes the intent↔code drift semantics directly (implementationMissing = intent-only, intentMissing = code-only).

Everything else is already generic: LoaderKind is a branded string (loaders are open), deriveUnits lets an ontology bootstrap its own extraction units, and the whole ontology axis is a plugin. So this is roughly the last ~20% that bakes the software-engineering worldview into core.

Proposed direction

1. Open up SourceRole. Turn the fixed ['code','intent'] enum into an ontology-declared, open set of roles. The ontology plugin declares its own roles (DDD: code | intent; a story ontology: canon | play; etc.).
2. Generalize the drift semantics. Replace EvidenceValidator's hard-coded intent↔code drift with a generic notion of cross-role corroboration / conflict between sources of different declared roles. The intent↔code drift becomes one instantiation, not a built-in.
3. Route bootstrap through deriveUnits. BatchService's mode === 'intent' special-casing should fall out once roles are ontology-defined and unit derivation already has the deriveUnits escape hatch.

Forcing function

Don't do this speculatively. The real driver is an external consumer project building a second, non-software ontology (ontology-story, a serialized interactive bedtime-story world) as a genuine third-party plugin via @braidhq/sdk. Its sources are canon | play rather than code | intent.

Acceptance criteria: a third-party ontology can declare its own SourceRole set and run the full extract → propose → HITL → model → view loop without touching @braidhq/core or @braidhq/schema. When that passes, braid is proven to be a pure framework, and ontology-ddd is just one example among several.

Notes

• This also makes ontology-ddd a cleaner reference: it would declare code | intent instead of relying on core knowing those names.
• Keep the HITL gate and evidence/provenance invariants — those are framework invariants and should stay. Only the software-specific source taxonomy is being lifted out.

---

🤖 Drafted via Claude Code while scoping the everstory project (a downstream consumer that will be the forcing function above).