Sharding has several features that operate on all commands (or large swaths of commands):

• safe secondary reads
• shard and database versioning
• causal consistency

For several of these, we have declarative tests that define a test case for every command (from listCommands output) and check that the feature works correctly for that command.

As query moves towards pushing more operations into aggregation, we lose coverage in these tests, since they do not define test cases for every possible initial aggregation stage.

This is especially problematic because these aggregation stages often have special logic ($changeStreams, $listCollections, $currentOp come to mind), so just testing a simple "aggregate" is not enough.