Maybe I can share my thoughts on the subject. Some of this will be about sampling in general, and you might already be familiar with that stuff.
I think a good way to start a project is to assess how many samples you are ready to record and process. It is also useful to think about the articulations because some articulations are harder to execute than others. The practicalities of sampling can be the biggest challenge. A well planned sampling session can save you a lot of time and energy.
Samples inhabit a 3D space:
1. notes on a keyboard,
2. velocity layers,
3. sequence positions (round robins).
It makes a huge difference which dimension you decide to emphasize. And the workload will practically force one to make compromises here.
Let’s say you want to make a library suited for heavily distorted metal music with lots of repeated notes and not much dynamics. In this case I would prioritize round robin samples over velocity layers.
With RJS Guitar I did the opposite. I prioritized velocity layers because I like instruments that respond well to dynamic playing. For that I sacrificed range and rapid fire repeatability of notes. Each string patch has 210 samples in it. Some of the samples are shared, so, in total I recorded something close to 400 samples. My success rate for recording a note was maybe somewhere around 70-80%, meaning that for 7 or 8 usable samples I had to record 10 attempts. The success rate is a rough estimate, it could actually be a bit worse. Anyway, I think the performance was pretty good. So, I probably recorded something between 500 and 600 samples. And I also ended up chopping a lot of samples that were eventually discarded. It’s tedious work. Poor choices made in the beginning can cost a lot later.
Processing and mapping samples by hand takes a lot more time than recording them. (At least if you don’t have to design and arrange an acoustic recording setup.)
Regarding your questions:
I think every articulation you list could work. Bends might be challenging because they imply rhythm. Vibrato is a little bit in that category too, but there seems to be some tolerance for vibrato that goes against the rhythm of the music. RJS Guitar has vibrato and it has never really bothered me that much. I didn’t know if it would work at all when I started making the library.
About circle of fifths etc: I think sampling every other note (2 semitone spacing) is really good. 3 semitone spacing works ok. 4-5 semitone spacing is probably where the problems start to show up. If there are no round robin samples nor a significant amount of velocity layers then playing adjacent notes will reveal that they are the same sample. Repeating timbral qualities start to poke your ears in a toy-piano-like fashion. And connecting samples that are far apart might be awkward because of the changes in timbre (caused by pitch shifting and the guitar itself). Somehow these effects are not equally detrimental to pianos.
About processed vs dry samples: I tried processed samples, and they didn’t work for me. It could be possible to make them work. If you make a large sample set, don’t commit to a printed sound unless you have done extensive experimenting and are really sure about using the printed effects. Of this I’m pretty sure: Pre-printed distortion won’t mix convincingly in chords, if the aim is to sound like a single guitar. If you record dry guitar straight into your computer and organize your project well then you can easily experiment with printed effects by re-rendering samples. Old sample files just get overwritten so you don’t have to change anything in your Kontakt instrument.
About release samples: They probably would do some magic as a final touch, but, are they a good investment? Maybe. The resources are limited. How to spend a sampling budget?
If you happen to be a Reaper user then you might want to check out the automation scripts I created after chopping those 500 samples. 😉
I hope this was useful! 🙂