Off the top of my head, Sakurai's criteria:
1. The character must originate from a video game
2. The character must make people want to play the game.
3. The character must bring something new to the table.
4. The character must contribute to the game's balance.
Those are paraphrased from his speech at the GDC 2008, and are probably still applicable to his character decisions today, even though people on forums really only debate 2. and 3.
I look at relevance the same way I look at moveset potential, in that it can help a character get one foot in the door, but it's not the free pass so many people believe it to be. What's more, it borders on intentional ignorance to distill a process that Sakurai describes as 'stressful to the brink of death' down to 'determine the most recent characters and lock them in, whoever they are.' A newer character isn't inherently better than an older one, and an older character isn't inherently worse than a newer one.
Older characters aren't even actively hurt by missing appearances, they just remain stagnant, allowing newer characters to develop credentials of their own...which is arguably hurtful, but even then, only if you believe in the concept of 'representation,' a whole new can of worms.