I don't do an ONS but I am an ardent splitter. However, my unscientific gut feeling is that
- (a) virtually every splitter will actually lump **some** source records;
- (b) the classic source records that we (including me) lump are BMD **indexes** such as FreeBMD or the GRO online indexes.
I have a single source record for all of the FreeBMD indexes, another for the GRO Online indexes, another for the Ancestry BMD indexes for England & Wales, one for CheshireBMD, one for LancashireBMD, etc, etc.
My reason was quite simple - I felt that the payload of repeated data in the source-record made it easier for me to reuse the same source record and the actual varying detail could easily go into the "citation" details. This gives only a couple of repeats for a citation for a single BMD index entry, unlike the number of repeats for a birth certificate, say, given that tends to support many more facts. (Normally I dislike repeats).