Compile vs run

What it is: the boundary between creating a RunSnapshot and executing it.

When it matters: whenever you need to reason about run_id, resume semantics, or runtime tuning.

What you provide: compile-time experiment inputs plus optional runtime overrides at execution time.

What Themis provides: immutable snapshot compilation and orchestrated execution over that snapshot.

Use this boundary map when it is unclear whether a setting freezes the run or only changes execution behavior.

flowchart LR
    A["Experiment definition"] --> B["compile()"]
    B --> C["RunSnapshot"]
    C --> D["run() / resume() / replay()"]
    E["RuntimeConfig"] --> D
    C --> F["Stable run_id"]
    D --> G["Execution state, artifacts, reports"]

Compilation decides what the run is, while execution decides how that frozen run is carried out.

What to inspect when it goes wrong: inspect the compiled snapshot first, then inspect execution state, stored events, and runtime settings.