['Defying Design' is a bi-weekly GameSetWatch-exclusive column by Jeffrey Matulef analyzing gaming conventions and the pros and cons of breaking them. This week's column takes a look at games that dramatically alter their mechanics in the third act.]

I loved Splinter Cell: Conviction until I hated it. For most of the game it was an accessible (if divisive) take on the stealth genre, mixing tactics and shooting into a quick-paced adventure.

Levels were somewhat open with multiple routes inviting players to shimmy around ledges and crawl though windows before silently taking enemies out from behind. Silenced pistol ammo was an unlimited resource, so shooter veterans could go about popping enemies in the head if they were alone. Even if Fisher was spotted, he could slip away in the shadows and take out the opposition by force before hiding again.

About a two-thirds of the way the game everything changed. The level design shifted to aggressively linear paths with no room for deviation. Enemies traveled in groups, so there was no way to take out more than a few at a time without alerting the rest of the squad.

Worse, none of the lights could be shot out -- something absolutely essential for sneaking. These levels didn't even make sense as Fisher, an agile soldier adept at climbing, couldn't crawl over waist high shrubbery. It was no longer an action/stealth hybrid, but rather a straight up third-person corridor shooter, and not a very good one because Fisher would go down after a scant few shots.

This is by no means an isolated case. Lots of games switch up their focus too much in the third act leaving a sour aftertaste.

Cult classic Mirror's Edge was notable for being a first-person game that wasn't a shooter. In fact, the entire game could be beaten without firing a single bullet (there's even a "pacifist" achievement for it).

The early levels of Mirror's Edge barely had any enemy soldiers and the game seemed like it was going to be something special. But as it wore on the enemy count grew denser culminating in one particularly awful sequence where Faith is trapped in a small parking garage with four soldiers. The only way out is a locked door that requires turning a crank several seconds. 

There are only two ways to get past this: disarm a foe and use their gun to shoot everyone else, or isolate the soldiers one by one and take them out with melee attacks. The former turns the game into an unremarkable shooter, and the latter is an exercise in frustration. Either way ruins the game's best quality; its sense of speed from free-running.

An emphasis on combat in the conclusion also plagued Tomb Raider: Anniversary. Most of the game was a lovely throwback to early adventure game design, focusing on the raiding of tombs rather than the ass-kicking of soldiers.

Throughout most of the campaign Lara would encounter deadly creatures sparingly, just enough to punctuate the slower, thoughtful exploration and puzzle solving that made up the vast majority of the game. Though in its final couple levels, the ratio of exploration to combat flipped and it became a poor action game. A pity as combat was always the Achilles heel of the series.

While an over abundance of combat bogs down games about sneaking, running, exploration, and puzzle solving, in a game that's explicitly about fighting, the former can be just as damaging. 

God of War was an exceptional hack-and-slash title until it got to its penultimate level, the aptly set Hades. This prolonged section forced players to navigate Kratos over a series of raised platforms over the river Styx.

It was a great premise and I love platforming, but that was never Kratos' forte, and these sequences were incredibly frustrating due to an unhelpful camera pulled way too far back making it difficult to gauge depth. With better execution and more foreshadowing, this could have been a nice respite from the action before the final boss, but instead it was an exercise in tedium.

Switching things up isn't always a bad thing as when done well, it can keep players surprised while still staying true to the game's concept.

One of my favorite examples of a game that does this well is Metal Gear Solid 4. The first half of the game was relatively traditional MGS "tactical espionage action" (i.e. it's a stealth game) with a few modest tweaks like being able to move in first person or fighting alongside a warring faction. Midway through that all changed and there were only a couple more conventional stealth sequences the rest of the game.

Instead, the second half focused on boss fights (of which there was only one in its first half, yet six in its second), on-rails vehicle sequences, and an interesting variation on stealth where players have to sneak past robots that behave very differently from traditional soldiers. 

These changes worked for me because they built upon mechanics used earlier in the game. There was already one vehicle mission in the first half and one boss fight, so these elements weren't completely our of nowhere. And while sneaking past robots is very different from enemy soldiers (your camouflage doesn't make a difference for them. They either see you or they don't), it's still stealth based and forced players to concoct new strategies to get past these parts undetected. 

The progression from sneaking past human (albeit nano-machine enhanced) soldiers to robots is also indicative of the game's themes that there's no longer a place for Foxhound's aging soldier. This switch in tactics might be a bit overkill (since the first half was so good that it leaves us wanting more, then doesn't deliver), but there's a reason for the game to switch (metal) gears as often as it does.

"Wow 'em in the end" was screenwriter, Robert McKee's advice to a fictitious Charlie Kaufmann in Adaptation. This advice rings true of games as well. Part of being wowed is doing something new, but there needs to be a buildup to this. When done well these transitions can be a welcome change of pace, but when done poorly they feel jarring and don't play to a title's strengths.

Part of the fun of games is honing one's skills at something, be it sneaking, jumping, or fighting. So when this knowledge is no longer needed it can feel like taking a final exam where most of the material was never covered in class.

[Jeffrey Matulef is a freelance writer whose work can be found at G4TV.com, Eurogamer, Paste, Joystiq, and Kill Screen among other places. He's also a regular on the Big Red Potion podcast. You can contact him at jmatulef at gmail dot com.]