Based on what we've found in profiling, there are a few really simple optimizations for per-column filters that we should do.

Here's a simple query on the bestbuy dataset which exercises the behavior. The filter does not match anything.

db.products.aggregate([{$match:{'type': 'asdf'}}, {$group:{_id:'$type', count: {$sum:1}}}])

We should change this to not copy values out of the column index in order to apply the filter to them. This means putting an unowned value into the slot used for filtering here.

These are only ideas for a starting point. We can also split these into separate tickets.