Build, cache, and import#

The runtime path is intentionally transactional:

validate -> fingerprint/key -> per-key lock -> staging build
         -> validate artifact -> atomic cache publish -> import

Cache keys include source and runtime/toolchain information required to decide whether an existing artifact is reusable. Cache metadata is schema-versioned; unknown future schemas are rebuilt rather than trusted optimistically.

Use compile_and_load_result or package *_result APIs when automation needs structured evidence. Use force_rebuild to invalidate reuse for a particular operation, not as a routine substitute for understanding cache identity.