A prior is the class share before the query features are applied. The lesson turns class counts into exact rational priors and keeps the count-over-total provenance visible.
highlighted = computed this step
The prior rule
The class prior is count(c) divided by the total count. For A this is 4/6 = 2/3.
P(A)=4/6=2/3
The other class
For B the same rule gives 2/6 = 1/3. The two priors sum to 1.
P(B)=2/6=1/3
Summary
The prior is each class's share of the shown data. No feature count has entered the product yet.