My "preferred take on filtered search" (as the OP phrases it) is to follow Google's lead. They use Group results with great success. With group results, the filters (or "categories") are presented as examples, and after the search has been submitted.
From a previous answer of mine:
To solve this problem, I recommend my clients to make two main result templates, based on the market leader's (google.com's) behaviour:
- One Main results template where categories (top level filters) are presented as group results. This template is without facets (granular filters).
- One Category results template where results within a category are represented with facets.
This also solves how to show facets that only appear within certain
categories. The answer is to wait until the user has clicked a group,
thereby informing the service what category she is interested in.
This, again, is a design pattern derived from google.com.
If you do multiple test searches on google.com (1, 2,
3) you will oftentimes see that on the first search results page,
results from one type of source (images, videos, news, tweets, map
results, shopping, facts etc. etc.) are represented together in a
group. Some refer to these as Google's OneBoxes. The purpose is
intuitive disambiguation; to guide the user to the desired
repository by way of providing examples, as opposed to forcing them to
read labels and try to understand your information architecture.
As part of an open source project I have illustrated template #1
here. Below are three of the suggested group results (OneBoxes).
These groups all have potential, individual facets, and these facets
are shown only when the particular group is selected, taking the user
to template #2.
In the original template the results are responsive. Based on
Bootstrap.
![enter image description
here]11 
This design pattern should be very useful and intuitive for many user stories. However, it may be more expensive than other, simpler solutions.