They say sharpening is to the image what spice is to food - a little goes a long way and too much ruins it. To keep me from the temptation of over-sharpening I use the 'sharpen for output' rule.

My 'original' tiff file is saved immediately after RAW conversion. From there I duplicate the original and process depending on output.

Web based images get a whole different workflow from print images. I typically sharpen for print tiffs three times during the workflow. I sharpen for the web jpgs once only, and then usually not globally but selectively using a sharpening brush. In both cases sharpening is the last thing I do before saving the finished file.

Also, if the subject is 'at risk' for over sharpening I will often downsize and resample for the web using bicubic rather than bicubic sharper. Initially the image is too soft but if image data has been properly preserved and treated up to that point, then I can brush sharpen selectively with good results. (There is no chance for good final sharpening of a jpg file if there's been too much and/or the wrong kind of data loss between the camera and the point of sharpening.)

It's a deep subject of course, and many variables are involved. The more complex sharpening formulas are more complex by virtue of attempting to effectively address a greater number of relevant variables in hopes of a better result.

For me at least, since every image is different, every sharpening protocol is thus different - usually just a little different but sometimes a lot. Very few areas of post-processing benefit from experience like sharpening does. Eventually less becomes more.