Validation

Validation catches shape, dtype, and semantic errors at construction time — before a long recording loop turns them into a mysterious mid-run traceback.

Validation levels

• Structural — types, shapes, dtypes. Always on.
• Semantic — selectors match at least one module, seeds are actually applied. On by default.
• Strict — reject anything the adapters do not fully understand. Opt-in via NS_STRICT=1.

When it runs

Validators run at Recorder.__init__, at save(), and atload(). Reload-time validation is what protects you from cross-version drift.

Custom validators

import neuronscope as ns

@ns.validator("recorder.probes")
def probes_are_unique(probes):
    names = [p.name for p in probes]
    if len(names) != len(set(names)):
        raise ns.ValidationError("E2201: duplicate probe names")