hifiasm -3/-4 mode doesn't apply the constraints

[zhenzhenyang-psu]

Dear Haoyu,
While I was running hifiasm -3/-4 mode, I provided hifi data and two lists of hap1 and hap2 reads as input. To my surprise, the resulting contigs mix reads of hap1 and hap2 reads.
For instance, see from the 1st following figure, h2tg000202l, it is contributed by 36 hap1 reads and 58 haplotype reads. The assignment accuracy of the input reads is 99%. I wonder if a fraction cutoff like 0.01 could be used to prevent such cases.

Rplot-issue_range3.pdf

Rplot-issue_range2.pdf

Can you comment?
Thanks,
Zhenzhen