Uncharted 2 has my favorite cold opening in all of gaming. Drake, woozy from blood loss, wakes up in a train car dangling over a cliff. The player's first task is to see Drake through the dizzying ascent, hand over hand, as the squealing metal of the railcar begins to deform and strip. It's the perfect sweet spot of gameplay and cinematics, but there's something most players don't realize: Drake is never in real danger. Sure, it's possible to die during the opening of Uncharted 2, but it's designed to be difficult to fail for one reason: if Drake dies during the climb, it kills the pacing. As a result, developers use audio-visual cues to give the player a sense of threat, but make sure the obstacles themselves aren't overwhelming. This is the challenge for action-adventure games: to inject players into death-defying Hollywood stunt sequences and make them feel like they've cheated death. The various tools for this have evolved over the years, from chase sequences to QTEs, but it's always the threat of death - real or perceived - that drives the action. In the case of Tomb Raider, it's even redefined the genre as a whole.
Much of the problem with agency vs. pacing stems from the relationship between games and the Hollywood blockbuster. A century of adventure serials, James Bond films, and summer popcorn-munchers have cemented the spectacular stunt as a necessary component of the action-adventure genre. Put simply, it's thrilling to watch someone defy death. When we watch Indy run from a careening boulder in Raiders of the Lost Ark we get excited not because we think Indy's going to die, but because we know he'll somehow make it out intact. The reason we implicitly know Indy will escape is because of the nature of film: movies are a controlled environment. On some level we understand that we as the audience only see the best result of the action scene, where everything works perfectly and looks good. If Harrison Ford dodges left instead of right during filming or a stuntman misses a jump between two speeding cars, you just cut that film, reset, and do it all again (presumably with a different stuntman). Games are different since, to a certain extent, players take on the role of actors, stuntmen and audience. If a player misses a jump in a game, the audience sees the mistake and do-over, killing the momentum of the sequence as the game reverts to a save point. Tension in action films doesn't come from the character dying, but from nearly dying, and when we see our avatar die it breaks both immersion and the sense of pacing.
Winston Churchill famously said: "Nothing in life is so exhilarating as to be shot at without result." That feeling, the sense that you're barely escaping with your life, is what action-adventure games try to capture. However, basic game mechanics throw a wrench in the gears, since to preserve a sense of challenge, players need to be put in danger of actually failing. In short, developers need to create the feeling of being shot at without result, while still making the player aware that the bullets can hit them. It's a fine line to walk, with developers having to make the player feel like they're in mortal danger without killing them. Really, it all boils down to how much agency a developer wants to give the player - they can either put the player on rails to ensure death doesn't destroy the pacing of a stunt, or they can give the player more leeway and risk repeated failures destroying a player's the sense of awe and excitement.
Uncharted 2 created a novel solution to this problem. While Drake is in little real danger during his ascent, all of the visual and auditory stimuli carries a sense of threat. There's the yawing emptiness of the chasm, the squeal of rusted metal, and the wind buffeting him on his handholds. At predetermined points, pieces of the car unexpectedly separate and drop him, or send him swinging into space. None of this can actually kill Drake, but that's immaterial. When you wrap players in such a compelling experience that their brains manufacture the sense of danger, there's little need to actually try to kill them. It turns a simple climbing tutorial into a fingernail-gnawing experience. But the tradeoff is that it's on rails, and the player doesn't have a lot of agency - essentially, the sense of threat only works if the environment remains compelling enough to mask the scripted nature of the experience.
As a contrast, consider the much-maligned Quick Time Event, where devs replace the audio/visual danger cues with very real danger. QTEs have become the bête noire of the gaming community of late, but developers still rely on them when they need characters to perform an action that isn't supported by normal gameplay mechanics. There are two main problems with QTEs when it comes to portraying stunts or other dangerous actions - the first is that they're usually not properly integrated into the experience of the game. God of War worked because it made button combinations a mainstay of combat, mostly serving as finishing moves or memorized combos rather than saving the player from insta-death. Far Cry 3 and the original Uncharted, on the other hand, throw in instant-death QTEs with no warning, some of which only exist in one section of the game. What's so frustrating about using QTEs this way is it feels like cheating. The developers are literally changing the rules of the game with little or no warning, and expecting you to learn on the fly. And because QTEs are tricky and precise button-press combinations, the player often fails on their first few tries, taking them out of the action. That's the second problem: QTEs are so game-like that at best the player's reminded that they're pushing buttons, not fighting aliens or running from an explosion, and at worst they're killed, destroying the pacing. The result is that most QTEs actually undermine their original intent by creating a sense of tunnel vision - players can't admire the incredible visuals because they're only looking at the button prompts that keep them playing. It's the worst of both worlds, removing the player's sense of agency while instituting an immersion-breaking penalty for failure.