The spec and SDK are available here:
Yes, the lossy compressed images are in 8 bit. Whether or not you'll get posterization will depend on the circumstances of the original capture, as well as how strong of a correction you apply. As you say in your case, you tend to make strong color corrections, so you'll simply need to do some more testing to determine whether or not there will be any problems. Unfortunately I can't think of any other rules of thumb or shortcuts other than direct evaluation with your own eyes.
Thanks, just saw your answer.
Hm, 8 bit lossy will be okay for second grade pictures. Don't have a faw days to go and compare each picture one by one.
The thing is, 8bit log is nice as an end/delivery format, not as a "negative storage" format that is post-processed using live effects by software like lightroom.
Hope that something like 10bit (or better) lossy compression will make it through in a future version.
Even if we had used 10 bits, my advice would be the same: whenever there is lossy compression involved, the only way to know whether that will suit your needs and meet your quality standards & expectations is to evaluate with your own eyes. Personally, my recommendation is that if you want to archive images and not worry about any possible quality loss, use lossless compression.
If a raw file is initially processes to boost shadows, color balance, etc., and then converted to lossy DNG, are the corrected signal values then mapped into 8-bits? Seems this would improve the information content of the lossy DNG around the values of the ideal image allowing latter additional tweaking
At the risk of stating the obvious: Lossy DNG is nothing else than good old lossy JPEG - with all its up- and downsides. It's not even meant to be used for archival purposes, as suggested in the specification: "Also added in DNG 220.127.116.11 is a new lossy JPEG compression option that is particularity useful for proxy DNGs, greatly reducing their file size."
It will almost certainly exhibit visible color banding (posterization) when fiddling with the histogram AFTER the conversion into 8-bit data.
The DNG-specification itself restricts lossless JPEG to the usage of 8 bit data: "Lossy JPEG (34892) is allowed for IFDs that use PhotometricInterpretation = 34892 (LinearRaw) and 8-bit integer data"; where "Linear..." means de-mosaiced.
ps: Sorry, I didn't recognize the age of this question.