Have you looked at DynamicPSF, Rob? It appears to provide a reasonable method for estimating the PSF by measuring a sample of suitable stars from a linear image. It also appears that best practice of using a deringing mask, global deringing and masking low SNR areas them can control artifacts pretty well.
I'm convinced that, in theory at least, deconvolution can provide benefits that the other tools can't. Unfortunately, it's a fairly daunting process in PI. I'm sure it gets easier with practice. And that's what I'll be doing again tonight
Cheers,
Rick.