As you might have seen in the spreadsheets for the reference and for participants, I am rounding off the percentages to the nearest integer.

There is a reason for that. For one thing, there are lots of factors that can influence these results. If I choose somewhat different reference samples, the ancestral components as well as their proportions in different individuals would vary from the current case. This is especially true for minor ancestral components.

I am running admixture for project participants in batches of 10 along with all of my reference dataset. Thus I am be sure that the ancestral components inferred stay the same from one batch to the next.

While the percentages do not vary much for the reference samples from one admixture run to another (with different project participant samples), they do change a little. And I have seen a few changes by as much as 1-2%.

Therefore, these ancestry percentages are at most accurate up to the nearest whole number. There is absolutely no difference between 11.7% and 12.4% for example in my opinion.