Alright, based on my testing and everyone’s feedback, my current conclusions are that a 16-bit TIFF and a Linear DNG can give reasonably close results:
- The biggest difference is in the color space handling. The linear DNG still contains scene-referred data, while a 16-bit TIFF will usually contain output-referred data. The impact of this difference can be mitigated by using a wide RGB color space such as DxO Wide Gamut or ProPhoto RGB. In practice, whether it matters or not for archival or subsequent edits depends on how much you may need to work on the image’s colors. If the white balance in the RAW’s metadata is very off, and you export that as a 16-bit TIFF without fixing it first, even in DxO Wide Gamut or ProPhoto RGB it might make that TIFF hard to use; while the linear DNG will not have this issue.
- Dynamic range seemed well preserved in both the linear DNG and 16-bit TIFF. My tests with very dark shadows seemed to show some limits in the TIFF initially, but after more testing it looks like limitations in the software I was using. Most photo software I tested seem to treat TIFFs and DNGs differently, with most having some issues applying extreme Exposure or Shadows corrections to the TIFF. Tools like a Tone Curve seem to have more predictable results with both formats. And at least one program I tried handled TIFFs better than the linear DNG.
- For archival, a linear DNG with some compression (old JPEG lossless or newer DNG 1.7 JPEG XL lossless or lossy compression) will probably fare better than a 16-bit TIFF, and since the linear DNG remains closer to the original RAW than the TIFF when it comes to color space, the linear DNG probably makes more sense… provided your current and future software will handle the format well.
- As an intermediate format between different pieces of software, a 16-bit TIFF might have an advantage if the software you’re sending your image to handles TIFFs better than linear DNGs, especially if it’s software that works with output-referred RGB data only (and optionally forces you to use a “Develop” module to develop the linear DNG data into an output-referred RGB layer, like Photoshop and Affinity do, and Pixelmator Pro seems to do in a less obvious way).
PS: for these conclusions I had to read up on what “scene-referred” and “output-referred” (also called “display-referred”) data means, and I think I get it, but I’m a bit fuzzy on the details or the practical applications. Looks like keeping data “scene-referred” keeps your options more open for later editing, but I don’t know how true that would be in practice.