In general, contrast-based autofocus systems as found in mirrorless cameras and DSLRs in live view are identical, and have the potential to be the most accurate.

Phase detection tells the system which direction to go, though, so it can be very fast — contrast-based systems have to seek around to find out. The main downside comes from using a separate sensor, which means that there's problems with alignment, and even when alignment is adjusted perfectly, problems with different light paths. Contrast-detection reads from the main sensor, and therefore avoids these problems.

The accuracy of contrast-detection AF comes down to a) whether you're actually focusing on exactly what you meant to and b) when the system decides it's spent enough time and gives up.

In low light, you'll likely run into both of these, as good contrast might be hard to come by in the scene. It's important to recognize that this means good contrast in at your desired focus point — the pools of light and shadow from streetlights you mention may be completely irrelevant if you didn't want to focus on one of those edges.

As for hybrid AF, think of it as contrast-based plus hints. Ideally, it should seek to the perfectly accurate focus in the end, but again may give up on the interest of getting you to "good enough" in a reasonable time.

So, going back to your symbols, it's roughly

(1 = 2 = 3) > 4

assuming enough contrast to work with and that the system doesn't give up.

That's the theory, by the way, behind this method of micro-adjusting phase-detect autofocus systems.