n3 defects map looks different from actual defects

Looking at raw.exposure plotting defects map (taken via butler) on the top of it (just using x0 and y0 ignoring heigh and width), there is reasonable matching in n2 and n4 but it looks that n3 defects are completely different from real feature. I thought first about the difference of the current master and old gen2 version
https://github.com/Subaru-PFS/drp_pfs_data/tree/master/nirDefects/sn3
https://github.com/Subaru-PFS/drp_pfs_data/tree/44e3a6b7e681cb162b3296db4d5d4188d1e70c56/nirDefects/sn3
but it seems that the content itself is the same (except for filename and header).

Please investigate that n3 defects are correct.