Yep. Defect pixels are those that cannot be relied on: being either too sensitive, too insensitive or just plain whacky in their behaviour. When a chip contains that many pixels its not surprising that a few are not perfect.
I also tend to create my defect map using the master dark. I measure the standard deviation and image background (Statistics option if using AstroArt) of the dark and then set the threshold at background + 5 x standarddev. The Starlight 694 I use doesn’t have many defective pixels and that usually cleans them up. My approach is overkill perhaps, but its adaptive and does lend itself to processing automation – I still sometimes use the Starlink CCDPACK image reduction system under Linux.
But if you always use the same camera temperature and binning, then a hardwired threshold derived by experiment should be fine – as Robin has found.