The study population is at risk, that is to say that its members should be able to contract the disease in the study.
In general, we look at the first occurrence of a disease in a person. In your example, the 3 cases diagnosed during the study are not at risk from the moment they are diagnosed. The total duration of follow-up for this study was 3*14person-days for those who do not have the disease (3 people times 2 weeks) and say that the study begin the first week on Monday. And the 3 peoples get sick on Monday, Tuesday and Wednesday of the week after, we have 7+8+9 person-days for those who have it.
The true incidence is $\frac{3}{(3*14)+(7+8+9)} = \frac{3}{66 person-days}$.