16.3 Three validity layers

Simulator-grounded claims require three validity obligations that must not be collapsed:

1. Formal validity: the theory is typed, coherent, and compatible with its declared laws.

2. Simulator-relative validity: its realization reproduces registered behavior and predicts held-out simulator outcomes under declared interventions.

3. Empirical validity: independent observations or physical experiments support the interpretation of those mechanisms in the target system.

A passed obligation may contribute a positive certificate to the dossier. The enclosing admission record must also preserve failed and unresolved obligations; executable success cannot silently erase missing empirical validation. The first two can hold when the third is unknown or false. Conversely, an empirical regularity may be real even when the simulator omits its mechanism. Disagreement can therefore localize inside the theory, inside the simulator, or in \(R\).

A code-level replacement \(\Phi _i\mapsto \widetilde{\Phi }_i\) has intervention semantics only when the registered simulator interface treats it as replacing the target mechanism while preserving the stipulated remainder of the system. If that modularity contract is absent, it is merely a controlled code perturbation. Even a valid simulator intervention becomes evidence for a real-world intervention only after the interpretation of \(\Phi _i\) is independently justified.

. The registered rungs below test executable, simulator-relative counterfactuals inside their declared substrates. Where a rung is admitted, its verdict remains relative to that interface and observer. It is not a substitute for intervention on animals, historical societies, or the atmosphere.