Composition and plan explainability

UniversalToolchain uses one planning authority and one explainable result.

LanguageCompiler returns LanguageBuildResult: planning diagnostics plus an immutable LanguagePlan on success. The plan exposes resolved features/contributions, runtime-provider identity, backend artifact routes, runtime policy, LanguagePlanSummary and PlanHash; LanguageLockFile is the stable machine-readable projection.

A useful report should show language/version, entry artifact, exact package identity, selected features/contributions, runtime provider, one ordered route per backend, runtime policy, diagnostics and plan hash.

Authority boundary

• package registry and LanguageCompiler own planning truth;
• LanguagePlan owns selected execution truth;
• LanguageRuntime verifies and materializes exactly that plan;
• formatted explanations are projections only.

The former dialect-integration explanation/runtime-selection graph was removed in S13 rather than retained as a parallel source of truth.

What the plan actually establishes

Configuration-time checks can reject unsatisfied dependencies, conflicting single-owner slots, ambiguous providers and routes, invalid typed artifact connections, or impossible pass ordering. The successful plan fixes the declared selection and package provenance. LanguageRuntime verifies and materializes matching components; it does not prove that independently written compiler passes preserve meaning.

The package feature manifest is schema v5 (LanguageFeatureManifestSerializer). The language lock snapshot is schema v6 (LanguageLockFile). These two projections describe different identities; neither is a certificate of arbitrary extension semantics. See planning diagnostics and limitations.