Sleep is a badly scaling save-or-lose

The Sleep spell has a long D&D pedigree, it was the first spell for many, many Magic-Users and Mages back in the days before 3rd edition. But, design-wise, it's pretty terrible.

The platonic ideal of a D&D encounter is one where the party wins, but comes out of it with close to no spells left, everybody down a ton of hit points, but triumphant.

The sleep spell is the opposite of that. At first level, sleep can solve a good number of encounters. Cast spell, greenskins fail their saves, PCs win. By third or fourth level, you cast it hoping that one of the greenskins maybe fails a save. By fifth or sixth level, you just don't cast it. And if you have limited Spells Known, you're book-diving for when and how you can trade it for something else.

Pathfinder 2E says that they're replacing save-or-dies with spells that scale more elegantly--replacing success/failure with critical success/success/failure/critical failure. But they're basing it on a single d20 roll, probably +/- 10. That still means that, realistically, there are two expected outcomes--because the other two are mathematically unlikely. (There's a "sweet spot" where you have three plausible outcomes, 1-5 6-15 16-20. But that's a fleeing moment in the campaign).

Without importing 5E's advantage mechanic wholesale, I think that rolling multiple d20s is an attractive way of resolving multi-level success-and-failures. Roll 2d20 vs target, you can succeed 0, 1 or 2 times. In the case of a re-worked sleep spell, we can use the "dazed" condition. 0 saves means "deeper daze" (good name for the spell) for 1 minute per caster level (same duration as sleep), 1 save means dazed for one round. And we can port the same clause that any ally can break the daze (just like waking up a sleeper) with a full-round action. (Or even a standard action). In a stealth situation, you could spam it on a single target, making sure that they were out of action before you sneak past. And we can easily extend the spell chain to 3rd level with mass daze, a 20' radius burst.
We could specify that "deeper daze" is broken by taking damage. That creates more interesting choices for the players. A "deeper dazed" opponent is out of the fight, but is still a future threat and may have loot.

This setup would also let "color spray" stay useful at higher levels--instead of scaling by target Hit Die, roll 3 Will saves for stunned (fail 1), blinded (fail 2), unconscious(fail 3).