After a merge, the corpus is re-segmented and the same rule runs again. The second round is computed from the post-merge rows, so the visible segmentation is the source for the next count.
highlighted = computed this step
Repeat the procedure
Round two starts from h ug and p ug. The same count-then-merge rule runs again on the new visible segmentation. Nothing about the procedure changes between rounds; only the symbols in each word have changed.
same deterministic rule
Round two counts
The pair (h,ug) has count 3 and (p,ug) has count 2. The selected merge is (h,ug) into hug because its integer count is larger under the same deterministic rule.
(h,ug)=3,(p,ug)=2
Summary
The second merge is deterministic because the counts are exact and ordered. The sequence so far is (u,g) to ug, then (h,ug) to hug. The ordered list is part of the tokenizer state.