Why do SIFT scores differ when generated automatically versus manually?

SIFT scores are calculated based on multiple sequence alignments of protein orthologues (SIFT Aligned Sequences). The scores can differ between genome builds 37 and 38 because MSA (Multiple sequence Alignments) can differ depending on the genome used. Differences seen between the Alamut^TM Visual Plus in-house predictors and the predictor website can be because the MSA differs and/or because the algorithm version differs between the website and the in-house versions.