The DISTINCT_SCAN optimization can only be triggered if the entire pipeline and indexes need to have a specific shape and reference specific fields in a specific order.

This can be used as an exercise to construct such pipelines in PBT:

• generate a list of field names
• create indexes over said fields
• $sort over that same list of fields
• optionally $match over that same list of fields
• use the first few columns for the _id field of the $group, and the remainder for the actual accumulators
• Add additional stages, e.g. $project, etc. that would also be using the initial list of fields

 
Then enforce the property that the indexed plan returns the same results as the collection scan. Given the way the $sort and the $group are generated from the same input list of column names, the output will be deterministic
 
This can be further extended by adding additional stages, e.g. $project