Comment by vpetrovykh

A few things come to mind:

1) It's a little unlikely that at the same time you have data where an {} is an error, but you are not making the property required AND with knowledge of that you still don't bother with other validation approaches. The point is that if you're aware that this property is potentially incorrect, you would want to check or restrict it. But yes, if this situation is completely unexpected, then "sum" won't notice any issues.

2) You have to remember that {} can arise from perfectly normal operations, such as filtering. So if you filter by a certain date range and there's no Payments there, then "(SELECT Payment FILTER .timestamp > <datetime>$date).amount" expression becomes {} even if the "amount" property itself is required. So doing a sum over it is simply a question of "What's the total amount in payments since $date?" and if there aren't any payments, the answer is 0, not {}. Plus what you certainly don't want is to get an {} from the "sum" here and do "{} + 100" to add some other charge (perhaps a sum from a different account) and still end up with {}, which now creates an error in a situation where there's nothing wrong with the data.

3) Rather than imbuing {} with special meaning to signal errors, a separate property would be more appropriate. Such as a boolean "valid" flag that gets set after the record checks out or can even be a dynamically computable expression that looks at the record (say, Payments from our example) and does something like "valid := EXISTS .amount". Then you'd filter things by the valid property before feeding them to "sum" like so:

  SELECT sum( (SELECT Payment FILTER .valid).amount );