Nothing in the best-worst sets get coded any differently. Anchoring works by the addition of the extra choice tasks (and sometimes the extra concept in each choice task) that capture the performance of the attributes relative to the anchor.
For example, on page 3 of Kevin's paper he describes the coding for the "indirect" method of setting an anchor. The first two coding matrices on that page show the standard MaxDiff data coding. But the next three small matrices (two on page 3 and one on page 4) show how to code the extra information provided by the follow-up anchoring question, depending on whether the respondent reports that some, none or all of the items in the task ate above the anchor.