Stealth gameplay is broken.
It sucks. It always sucks. Stealth is the game element most likely to ruin any game that includes it.
Metal Gear Solid is a huge success. Splinter Cell and Thief and Tenchu have all done more or less okay. They have not succeeded because of their stealth gameplay; they have succeeded in spite of it.
The reason is this: stealth gameplay is broken. It is inherently bad gameplay.
Stealth avoids conflict
Avoiding conflict is not fun. More importantly, it is bad drama. The key to drama is conflict born out of the natual drive of the narrative; good story happens as a result of how that conflict changes those who participate in it.
You could say that the conflict in stealth is the player's desire to not get caught versus the desire of guards to catch the player; however, in most stealth gameplay the guards are not actively looking for the player, meaning they are not dynamic participants in the conflict. Neither protagonist or antagonist are changed by the activity.
Games that succeed in spite of stealth use stealth as a means to build tension, leading up to cathartic confrontations against boss characters or set-piece action scenes.
Stealth does not offer resolution
Stealth does not resolve problems. When you sneak past a guard, the guard is still there. The problem still exists, and may even complicate your further advancement. Your activity has made your situation more dangerous as a result of your activity, rather than increasing your mastery of the situation.
Games that succeed in spite of stealth use stealth as an element of assassination, rewarding your successful stealth activity with the elimination of the guard and a correspondingly less fraught game space.
Stealth removes the ability for the player to pace gameplay
Stealth gameplay proceeds at the pace of the guards. Your motions are dictated by the patrols of non-player characters. This disempowers the player and removes their ability to customise their game experience to their tastes. It creates periods of downtime in which the player can take no meaningful action without causing mission failure and thereby reduces the player to the position of an audience to their own story.
Games that succeed in spite of stealth offer alternative routes for navigating an environment, with each route offering a different pacing and a different level of risk. They also provide meaningful things that a player can do while waiting for guards, or offer a level of environmental detail that is inherently enjoyable to experience during enforced pauses.
Stealth mechanises NPCs
Stealth activity involves the player moving through the gaps in the enemy guard pattern. To successfully engage in this activity, the player requires three types of information: the location of the enemies, the cone of sight of the enemies, and the manner in which the enemies move. This is, obviously, a significantly higher level of information than the enemies have available.
In the most poorly designed games this information is not well revealed to the player; however, when it is available, it renders the guards no better than machines, bound to understood rules of behaviour and possessed of unfeasibily low levels of awareness and intelligence. It can be difficult to respect such enemies and take them seriously. Contemptible opposition isolates a player from the game world and inhibits their ability to become immersed and empathise with their avatar and other characters.
Games that succeed in spite of stealth take steps to equip their predictable NPCs with personality quirks, idle animations, a wide range of context-appropriate spoken dialogue, and patrol patterns that are believable and effective in relation to the environment. These games have NPCs who are inherently intelligent and effective, whose ability to spot the player is limited by the nature of the level design and the competencies of the player character rather than their own inadequacy.
Stealth is binary
You are seen, or you are not seen. You are alive, or you are dead. Where shooting games can include health metres and driving games include finishing times, there is no "almost stealthy". In a worst case scenario, being detected results in death, and you move back to the last checkpoint (or, worse, the start of the level). In a best case scenario, being detected sends guards into an "alert state", where they deviate from patrol routes and possibly even actively search for the player. The player must hide (which involves standing around in an isolated spot and doing nothing, a form of punishment itself.) If the guards find the player, though, we return to that problem - being found equals failure. Not "a bit of failure" or " a portion of failure" - just failure. If you're lucky the stealth game ends and you're now in a fighting game. If you're unlucky, it's back to the checkpoint.
Games that succeed in spite of stealth attempt to make failure non-binary by including multiple alert levels and allowing players to "fight their way free" of guards, effectively making their health meter a substitute "stealth failure" meter. They also make the process of hiding from alert guards a game in and of itself and minimise the "waiting for the all clear" downtime.
There are good stealth games; there are good stealth segments inside some games. But it's fighting an uphill battle. You're trying to bring fun to the player, instead of bring the player to the fun. Stealth is inherently broken; don't treat it as a gameplay staple, don't go there unless you know what you're doing, and for heaven's sake don't mix it into an otherwise un-broken product.