Data validation structures and storage

[cmutel]

An quantitative observation is located in a technological, spatial, and temporal context (observations from future-oriented models can also have a scenario). Such observations need automatic validation checks.

Models also need test data, and tests validating against known good results. Such unit or integration tests should follow standard practices and are outside the scope of this discussion, though the techniques used might be of interest.

Validation checks could include the following:

• Min and max of accepted values based on physical limitations
• Min and max of accepted values based on economic conditions
• Expected values for statistical moments, such as the average

We can also have broader, system-level feasibility checks. For example, the amount of ice cream produced in a country over a time period can't be greater than the amount of milk imported and domestically produced. In theory we should be able to introspect our models to automatically generate such feasibility constraints, though maybe we can just copy wholesale from the MFA community.