how the heck do I model this?

I have the following case:

Based on a source_id plus n key/value pairs, where n is unknown (and theoretically unlimited), I need to look up 2 more values (let's call them site_datatype_id and interval). Obviously, the problem is that the 2 values to be looked up depend on multiple rows of lookup-data.

Take the following mapping table. It more accurately models the true relationship of the elements (which is a 1:m relationship between source/ site_datatype/interval and key/value). (I know it's not normalized: source, site_datatype, and interval are repeating in this design -- this is really probably a view on a couple of tables):

*source_id
*site_datatype_id
*interval
*key

key_value

It's easy to lookup a key_value based on PK of source, site_datatype, interval,
and key (and it is a requirement to go in that direction as well as the one described above). However, getting the site_datatype_id and interval from the source and multiple key/value pairs is not so straightforward... at least not that I can see. I'm sure that some sort of procedural kludge could be implemented to give the desired answer in this direction (e.g., get the site_datatype_id/interval pair that occurs most frequently (i.e, in the most number of rows) where source_id = <x> and ((key = <key1> and key_value = <keyvalue1>) OR ... OR (key = <keyn> and key_value = <keyvaluen>))