I know that this I am asking an unanswerable question because the answer probably depends on the image and what we are going to do with it in post-processing.
But I'll ask it anyway. What do you think is the best sharpening workflow?
Formally, from the point of Digital Signal Processing (DSP), the Capture sharpening/deconvolution should take place earliy in the chain of postprocessing steps, when the captured signal still has a linear relationship (except for the capture blur and conversion noise) with the scene's signal or exposure. That will allow to extract the most of the original signal from the captured signal.
So that would plead for deconvoluion capture sharpening at the earliest moment of Raw conversion, before noise reduction, before tonescale adjustments/tonemapping.
Since Raw converters can use metadata from the Raw files to streamline part of the Capture sharpening process, it would be relatively easy to implement such a facility, but the software engineers are apparently somewhat oblivious to the opportunity. As I've explained on other occasions, the lens quality and the aperture used for a given sensor are the main deciding factors for the default (Gaussian sigma) radius that comes closest to a perfect setting of that important Point Spread Function (PSF) modelling parameter. Yet most sharpening dialogs start with "Amount" instead of radius, which is backwards and telling...
Assuming Lightroom and Photoshop; and that there will be either upsampling or downsampling for output; and that we have a well-taken image with a good lens and camera.
I will understand if you are too weary of the subject to answer
Well, it's being made unnecessarily difficult by the programs you mention, but unfortunately they are no exception. It can be easily shown that the PSF shape required for deconvolution is variable, but that the aperture value that was used when shooting is a driving force. Good lenses usually require approx. a 0.7 radius for shots taken at 'optimal' aperture values (usually something like 2 stops down from wide open) in the focus plane. Wider apertures may suffer from some residual lens aberrations and demand a larger radius (how much that is depends on lens quality and widest aperture), and narrower apertures will be diffraction affected which also increases the required radius, perhaps to something like 1.0 to 1.2, depending on lens and aperture shape.
Defocus will require a somewhat different PSF shape, more resembling a flattened Gaussian shape, but the Gaussian shape remains pretty dominant overall.
So strictly speaking (and for a smooth workflow), one should attempt to repair Capture blur, with deconvolution during the Raw conversion. Unfortunately, the Raw converter sharpening tools/dialogs have a rather mediocre implementation of Deconvolution sharpening. Therefore, the attempts to do it properly in the Rawconverter tend to create substandard results.
That is why it may be beneficial for image quality to postpone the Capture sharpening to a later stage, although that also creates a less than ideal situation for the Deconvolution tools. And it makes for a relatively clumsy workflow, having to render the image, resize it and only then do the thing that needed to be done first. One benefit of the process though, is that while things like scaling, distortion correction, etc , all add new blur to the image, the blur PSF tends to become more Gaussian in shape again, so we can address the combined blur with a relatively simple model, that can also be implemented much more efficiently in software as two separable linear (de)convolutions rather than one 2-dimensional (de)convolution. Doing it later in the workflow also means that we have to worry less about artifacts that cumulate, due to rounding errors and sub-optimal settings early in the chain of events.
So, to make a long story short, in my workflow I usually postpone the capture sharpening to a later moment in finishing the final output. Since I use Capture One as my main Raw converter, that's easy because I can just select to disable the sharpening on export with one checkbox in the output recipe. If I need a faster workflow, I keep the sharpening settings that have an adjusted Radius setting based on aperture value, and an amount that matches the output requirements (more for printed output, less for other output).
Cheers,
Bart