I have been testing France scenarios in OpenFisca, and I found that omitted salary inputs are treated the same as explicit zeros once the scenario is built.

Concrete local evidence

• Without salaire_de_base:
  • salaire_de_base = [0.0, 0.0]
  • salaire_super_brut = [0.0, 0.0]
• With salaire_de_base = 0:
  • salaire_de_base = [0.0, 0.0]
  • salaire_super_brut = [0.0, 0.0]
• With salaire_de_base = 2000:
  • salaire_de_base = [166.6666717529297, 0.0]
  • salaire_super_brut = [170.17333984375, 0.0]

That is convenient for fully specified simulations, but it erases the difference between:

1. The user intentionally entered zero, and
2. The user did not provide the field at all.

For screener-style or partially completed scenarios, that distinction matters. A missing salary field should usually propagate as unknown or incomplete instead of being silently normalized to zero.

Questions for maintainers

Would any of these approaches be idiomatic?

• A preserved missingness state on input holders
• An explicit valueState-style annotation for inputs
• A documented partial-input mode that propagates "cannot determine" rather than defaulting to zero

I am not proposing an engine-wide rewrite. I am asking whether there is an accepted way to distinguish missing from zero today. If not, I would be glad to follow up with a narrow proposal.

Runtime used for the cases: openfisca-core==44.7.0, openfisca-france==175.1.0.

Evidence packet (external): https://github.com/edithatogo/rulesandprocesses/blob/main/external/openfisca/MISSINGNESS_CASES.md