23andMe results indicate SNP positions and alleles based on something called the NCBI human genome assembly. Both the raw data as well as site features and reports use NCBI Build GRCh37 assembly.
While the reference human genome has been “finished” since 2004, it contains a small number of regions of unknown and/or incorrect sequence. When these inconsistencies are discovered, the reference genome is updated to reflect the more accurate genome sequence. Each of these updates is called a “genome assembly.”
The strand reported in the raw data may be different than what is reported on a third-party site such as dbSNP; dbSNP has their own stranding convention which is not always consistent with the genome reference strand, because they have to be able to strand many types of SNPs, including those that don't align properly to the human genome or align in multiple places. In this case, 23andMe's reported data is stranded opposite of dbSNP, and this is expected behavior.