I'm using Atlas full text search to query a collection and highlight search matches. The search is executed against a search index using an aggregation pipeline with a compound filter/should query.

There seems to be "cross-pollination", for lack of a better term, between the filter and should stages. It's causing matches between the filter text and the values at the 'should' paths to show up as highlights. I wouldn't expect this, as in my case the filter path and should paths are disjoint sets.

It's a little convoluted to explain, but the forum post below has an example that I think will make it clearer.