I want to give a brief history of the early roguelikes. These games are the branch of the roguelike family tree descended from the original Rogue. The other four classic branches (Hack, Moria, Larn, Omega) are games inspired by Rogue, written sometime during the 1980's. In contrast, the early roguelikes are modifications of Rogue, based on the same code.
Development of the original Rogue passed through three phases. These are often called V3, V4, and V5, but the most common version numbers from each phase were 3.6, 5.2, and 5.4. They were apparently released in the spring of 1981, 1982, and 1983 respectively.
All of the early roguelikes are descended from the V3 phase. As this @Play article explains, getting hold of the Rogue source code was difficult. All the early roguelikes can be traced to a single copy of the Rogue 3.6 sources that was sent to AT&T's Bell Labs.
Super-Rogue was the first roguelike made by modifying this code, beginning in 1982 and culminating two years later. While deepening the dungeon, adding more monsters, and inventing more items, its designers avoided making drastic changes.
Advanced Rogue appeared soon after, created by a different group at AT&T. It shares some code with Super-Rogue, but I believe only a few features were borrowed, and Advanced Rogue should be considered a separate fork of Rogue V3. It is a more complex game with heavy additions, notably character classes and magic.
A surviving version, Advanced Rogue 5.8, was released at the beginning of 1985. About this time, UltraRogue was forked from it.
Advanced Rogue continued to develop until version 7.7, released mid-1986 with many further additions. XRogue forked away sometime near the end of this process, making small changes to game balance and user interface.
Summary: Super-Rogue and Advanced Rogue are forks of Rogue V3, developed c. 1983-86. UltraRogue and XRogue are forks of Advanced Rogue, developed c. 1985-91.
The early roguelikes faded away quickly. By 1987, NetHack was in existence, and Moria had been ported to Unix systems. These games had more developers and newer, more easily modified codebases. I suspect the authors of early roguelikes found it too difficult to add new features fast enough to keep up.
For example, Rogue uses a player-centered time system. This is simple to write, but not good at handling monsters with different speeds. Complex games usually use a world-centered, tick-based time system. (This @Play article explains the difference.) Advanced Rogue 7 was the first Rogue variant to convert to a tick system.
But there's a quirk to it. In most roguelikes, the player presses a key, the character acts, and is then unable to act again until the proper delay has passed. In Advanced Rogue 7, the delay happens before the action. When the player presses a key, the action is merely scheduled for some time in the future. In between, the monsters often get their turns to move. Sometimes, you have to take a swing at the empty space where a monster is going to be.
I've studied the source code, and the developers seem to have deliberately intended the game to work this way. I can't guess why. It's not hard to imagine similar confusion among roguelike fans of 1987, resulting in switches to NetHack.
Whatever the cause, the early roguelikes disappeared. In fact, they are sometimes called lost roguelikes, because of the difficulty of finding a playable copy. But thanks to the Roguelike Restoration Project, they are no longer lost.
Now that they can be played again, which should you start with? That depends on player preferences. Most of the early roguelikes piled on features, starting a tradition usually associated with NetHack. If you don't mind the need for spoilers, try XRogue, which I hope to have online soon. If you prefer balanced, straightforward roguelikes, stick with Super-Rogue.