With the Virtual Replicas, it is expected that the sorted results are not exhaustive. It does show the same amount of results with the facet filter, but they are a different set of records than without the facet filter. When you use a filter, it first filters the master index and then retrieves the first x number of records based on the paginationLimitedTo parameter value, then sorts those results based on the virtual replica configuration.
If I use the facetFilter on the master index, then I get different results for the first x number of records, which is the 'view' of the results that the virtual replica will grab to sort.
The "paginationLimitedTo" is set to 1000 by default and can be increased up to 20000.
That said, by default, the virtual replica retrieves the top 1000 items from the primary index and return them with the specified sorting.
Because the initial fetch from the primary index is based on the default relevancy configuration, it's possible that some items are not retrieved in this phase.
Increasing the parameter to 20000 allows virtual replicas to retrieve more items but it can also increase the latency.
The relevancy strictness parameter controls the threshold when fetching records from the primary index, but the "pagenationLimitedTo" parameter still limits the result.
If you need the exhaustive sort result with an index that has more than 20000 records, please consider using standard replicas instead.