Slices shorter than bins

When the inactivity timeout $t_{inactive}$ is short or when the analysis is over long time bins (say hours), flow slices can be shorter than the bin size. It can happen that we have more than two records for the same flow within the same bin. For byte and packet counts we can just add the individual estimates for the different records and we get an unbiased estimator for the entire bin. For active flows we cannot get an unbiased estimate, not even with flow reconstruction. For flow arrivals, by using $\widehat{A}^{(1)}$ for the individual records and summing the contributions without any flow reconstruction gives unbiased estimates as long as assumption 1 is not violated. For a record started before the beginning of the bin, even if it has the SYN flag set in violation of assumption 1 we do not count it as flow arrival and thus have $\widehat{A}^{(1)}=0$.