Does dbSNP have different conventions regarding right-shift vs left-shift of variant positions when it reports for GRCh37 versus GRCh38?

Two examples are rs367896724 and rs1424506967. Both are:

• small insertions
• occur at the beginning of chr1 where the reference sequence has not yet diverged between the two genome references.
• have a repeated base in the ref sequence so there are two positions that could be reported, and still yield the same resultant sequence.

But in GRCh38, their web-page's reported positions are left-shifted, whereas for GRCh37, they are right-shifted.

I speculated that perhaps this was somehow related to execution of a liftOver, but I tried liftOvers in both directions (using chain files from UCSC) and those positions get mapped unchanged.

Is left-shift versus right-shift an explicit choice of dbSNP which is documented somewhere (I couldn't find anything)?

If there is a systemic difference in left-versus-right shifting in dbSNP, then I would think that could seriously impair attempts to assign RS#s to variants from e.g. VCF files which I believe generally follow a left-shifted convention.