bug: `SubPopulation` index isn't passed to `MultiSignalModel.sample()`

[KOVALW]

Goal

We want to pass a SubpopulationCounts observation process into MultiSignalModel, but on run the sampling fails

Context

Runtime fails because multisignal_model.py#L413 doesn't pass the subpop_indices to sample() as required

This isn't caught on build because validate checksn_subpops appropriately

Specifications

• Modify the multisignal model sample function to accept subpopulation indices if present
• Tests to reproduce behavior including a SubpopulationCounts observation process inside the run

Out of scope

• Any changes to how subpopulations connect with each other

[cdc-mitzimorris]

We reproduced the failure. MultiSignalModel.sample() can forward subpop_indices, but only when it is nested under the registered name of the corresponding observation
signal:

model.run(
    ...,
    regional_counts={
        "obs": counts,
        "subpop_indices": subpop_indices,
    },
)

Passing subpop_indices at the top level does not associate it with the corresponding observation signal. The current high-level API accepts that call without identifying the misplaced argument, does not automatically perform the available validation before inference, and subsequently fails when SubpopulationCounts.sample() receives no subpop_indices.

Therefore, although the failure is not caused by line 413 omitting the argument—line 413 forwards the contents of the signal-specific dictionary—the reported behavior is a genuine API and validation failure. Users should not need to understand the internal dispatch mechanism to diagnose it.

As an immediate correction, we will document the required signal-specific nesting, add an end-to-end regression test, and make incorrectly placed or unknown arguments fail early with an actionable error.

As a follow-up, we propose introducing a typed observation-data interface to make observation data and indexing metadata easier to specify, validate, and reuse across fitting and predictive sampling.