This seems like a question that should have been addressed before, but somehow I cannot find anything. There is a post on samtools SourceForge page from 2014 without any responses. At some point, bcftools introduced an alternate variant calling model:
In versions of samtools <= 0.1.19 calling was done with bcftools view. Users are now required to choose between the old samtools calling model (-c/--consensus-caller) and the new multiallelic calling model (-m/--multiallelic-caller). The multiallelic calling model is recommended for most tasks.
However, I could not find anywhere where the differences between the two are explained. The obvious one would be that the new one is able to handle multiple alleles. If it was only an improvement, why keep the old method as well? It's recommended for most (not all) tasks. Which tasks are those?
reproducibility of previous results
If reproducibility is the goal, wouldn't you just use the previous version of the tool?