Interpreted and compiled code

Status: implementation guide. The normative rules live in the Core-1 modalities specification.

Core-1 gives every message an explicit authoring modality. An interpreted message can use the open-ended Dynamic boundary; a compiled message belongs to the stricter materializable subset.

module signal;

public value Signal {
    public compiled algorithm message score(type: Class<Self>,
        input: Int64) -> Int64 effects {} {
        return input * input;
    }

    public interpreted algorithm message explain(type: Class<Self>,
        item: Dynamic) -> Dynamic effects { reflect } {
        return item;
    }
}

Authoring modality is separate from execution tier. Compiled-profile code may still run in the checked interpreter. Native materializers reject a dependency closure that reaches interpreted-profile code.

Adaptive conversion is an agent-assisted rewrite, not a syntax-only transpilation promise. The compiler sends a canonical checked-AST package closure to the adapter, recompiles the candidate, and admits it only when contract identity and replay agreement hold. A later divergence quarantines the exact body and falls back atomically.

Read the normative modality and adaptation clauses for the precise trust boundary.