Your observations are mostly correct, Nizar.

However, please note the following:
1) The velocity factor setting affects only the reported length measurement, and nothing else. And setting the velocity factor to 3 significant digits is, in my opinion, never needed. This is because manufacturing tolerances of coaxial cables result in velocity factors across a much larger range than that. Even from the same manufacturer, two batches of cable of the same type can vary in velocity factor by a few percent. So if you need accuracy better than a few percent, you must physically measure each cable length to determine its precise velocity factor. Which means that the nanovna measure cable function is excellent for getting an estimate of the cable length within a few percent, but can only get better than that if you physically measure the cable anyway, which defeats the purpose of having the nanovna measure it for you.

2) As a side-note, when you require a precise length measurement for something like a tuning stub, the nanovna can do that with high precision, since you measure and trim for the electrical length, which is not subject to velocity factor variations.

3) While you may achieve the highest precision by adjusting the frequency range to make just over a half-circle on the Smith chart, please note that the nanovna 'measure' firmware does an excellent internal (cubic spline) interpolation between those 401 measurement points. So even if you set your upper frequency such that 3 or 4 full circles are drawn on the Smith chart, you will still get a highly accurate data. Just try it. Look at the data measured with one half-circle, then change it to several half-circles or full-circles and compare. It does so well at the interpolation that I rarely change my upper frequency (which is typically 30MHz for HF use) unless the Smith chart draws many circles and I need to verify the accuracy. (This is true at least for cable lengths of a few feet or more, which can be measured with a 30MHz upper frequency - I haven't evaluated this for very short cables - but the interpolation is the same, so I'm sure it will still achieve good results).

4) As for the loss measurements, early versions of the 'measure' firmware used a simple calculation: 1/2 of the measured two-way reflection loss of the open-ended cable, at the cursor frequency. This is the same measurement you would get with the normal S11 logmag. This works quite well, but for some scenarios, a more accurate loss measurement can be made by doing both an 'open' and a 'short' measurement, and averaging the two, or by connecting both ends of the cable to the nanovna and doing an S21 'through' measurement. Recent versions of the 'measure' firmware were updated to use a more complex and accurate loss calculation (discussed a few months ago in detail on the nanovna-beta group). The nanovna-H4 implements the first (and dominant) quadratic part of this calculation, while the V2 devices (with more firmware/memory space) include the full calculation for even greater accuracy. This method again is based on the S11 reflection measurement, but uses an advanced mathematical model on what loss that represents.

I hope this information is useful to you...
Stan KC7XE