The Cave of Time problem
The issue of story sprawl, or plot-divergence, where the player has 1 common starting point and each decision results in a new branch leading to a different ending, is usually referenced by the Choose Your Own Adventure™ book called The Cave of Time.
Cave of Time is shallow but diverse, lots of story branches are left unseen, but the meta-purpose is replayability because each read-through is short. Some games use a timeloop as a similar mechanic. No set path and keeping each playthrough short, but requiring multiple plays to accumulate the knowledge to win the game in a single play.
Your rpg could imagine these shallow branches as side quests, experienced in no correct order but the knowledge eventually accumulating, allowing the player to deduce their significance through environmental storytelling and worldbuilding.
But your survival gameloop may suffer because it's in direct conflict with a player uncovering all the necessary parts of the meta story.
A 'fix' to this ever-geometrically branching structure is called branch-and-bottleneck. Essentially the story is allowed to branch within a section, but bottlenecked by a major plot event that would be the equivalent to an act in theater/film – a plot point that is so important that it fundamentally changes the goal of the protagonist.
Branch-and-bottleneck allows for character freedom within the acts, but resets the story at each bottleneck to stay on course for a common ending. Stats are usually involved that refer back to the previous sections, allowing for tailored resolutions after the shared (inevitable) climax.
Notice in the second diagram, there are two completely siloed paths through Act 3 (presumably dictated by earlier choices). Any section like this has to be fully produced and packaged in the game despite a percentage of players never encountering any of it. Worse, a player would need to restart the game from a saved checkpoint to see the 'full' content.
'Strong Narrative' means character growth
A good story puts the protagonist through the ringer. They start with an unrealistic desire (an unearned 'want'), and over the course of the story, through their own actions, they sacrifice that ideal but generally get something that resembles the original want but it's not the naive (unearned) thing they imagined.... They emerge from the ordeal as changed person who's gone through a growth arc. They didn't 'live happily everafter', they didn't have all their wishes granted. Their naive want gets discredited in favor of an earned reality which (thanks to us following the protagonist's arc) is also deeper and more meaningful to the reader.
A typical fight and shoot game protagonist is (in narrative terms) a Mary Sue who keeps leveling up and is always the center of action. This character doesn't have an 'arc', just a trajectory. After defeating the antagonists, the goalposts are reset for the next level.
It's a genre clash between an 'adult' story about consequence and growth vs an episodic wish-fulfillment power romp.
There is no common ground in which these overlap, leading to the much-discussed ludo-narrative dissonance where a game's play loop is the episodic power fantasy, while cutscenes tell a tonally different story on rails.
Notice how this issue is only exacerbated by branch-and-bottleneck. It doesn't matter what the player chooses in (open world) Act 1, it always lead to the same bottleneck that resets the plot for Act 2.
Decide what the player can control
Is the player making choices for the story, or for their character?
The option to let the deurtagonist to tag along. She plays a MAJOR
Player can TRY to prevent deurtagonist from coming, but can they prevent a plucky self-determined strong character from following?
The sidekick needs her own reasons for going to Mordor, and her own reasons for sticking near the MC despite him not wanting anything to do with her.
Assuming the player wants to play as a drunk loner, it necessarily changes the motivations of this sidekick. Not so much "Boo hoo, he was mean to the sidekick and now she is sad and can't go...", more like "She is revealed to not be looking for a hero, she is looking for a bull-headed fool who will kick down the door and take all the bullets, something she can't do on her own.
Turning the MC into a selfish p.o.s., turns other characters into selfish p.o.s. to compliment. You can preserve the plot bottleneck – they both end up in Morder together, but the player's behavior influence the way other NPC are characterized. She might help him along the way, but only until he gets the door open for her, then he's expendable or the fallguy.
This can be within the range of the character, for example: she can come on as starry-eyed looking for a hero. If the player refuses that role she follows anyway giving the player opportunity to 'correct' his behavior and allow her in, 'sad puppy' but now with caution. If he continues to rebuff her, you need a dark fall-back reason that motivates her to follow. And presumably these are just on-going character stats so they can have many fails and resets within that relationship arc as events happen.
Character archetypes play with these trope shifts. You can probably imagine any achetypal character along a trope spectrum. It's equally interesting that she starts out with false intentions, but he's nice to her and she feels conflicted..., and later her false motives exposed through a plot twist, or keeping her on the fence throughout.
The character growth of the protagonist from being a miserable drunk
to a full on hero (though he is more of an anti-hero)
His friendships have to be earned. The reward is not more character points but unlocking unique story interactions with other characters – story moments that are also clues and hints, or earned tokens that can pay-off later.
Extreme actions get extreme reactions, and deception and coercion, thus the player is building their own character's arc that is working for or against him (positive or negative arc). Stats will keep track and adjust, with alternate dialog covering each stat possibility. A scared village could become an angry mob under the right stats, and the player's ability to influence those stats through their behavior could lead to very different gameplay.
'Nice' players may encounter a nicer world where social transactions lead to rewards. Crash and burn players get a more hostile world.
The arcs of multiple characters
Were getting into game mechanics territory. Does the player control one character, or multiple characters? What exactly are the 'levers' that trigger a stat change? Also what is the game loop?
Is each situation a different puzzle? Are there factions? Are there transactions? Try to think beyond fetching someone's frying pan in exchange for an infodump. NPC are multi-dimensional characters with a range of wants and needs. No interaction should result in a binary yes/no. Each situation should be plotted on a range of possibilities within this storyworld, where both player and NPC have other levers to negotiate with, and secondary motives they're willing to settle for.
Certain events that occur with the storyline
This reminds me of the classic time travel paradox. Does the world 'right' itself when you attempt to alter history? In this case, can some other character do the the thing a player is refusing to do? Can your story 'right' itself?
Or is the player in control? What are the stakes if they don't want to play your story?
Can't really address all situations in a branching narrative. Think of your genre and treat a setup as a foreshadowed event that – despite the MC having a change of heart – happens anyway, and they are unable to stop it. It's just a tonal shift whether the MC causes a disaster or is unable to prevent a disaster they have some responsibility for/connection to.
Also, I think this is closely related to can the player die. If the MC is stubborn, make them an offer they can't refuse. Kill them until they do it right the next time. If they make enemies, let those enemies pay-off.
Side characters who play a major role
In my story game I settled on 'head hopping' 3rd-person. I had an ensemble cast, and you made choices for only one of them at any time – losing narrative control over the others. Agatha Christie where you get to play all the characters.
Choices while playing a character altered personality stats when that character was an NPC. Player had a broader awareness of what a character is capable of, having seen prompts for choices they didn't pick. Essentially the player had the option of ramping up how extra or how moderated the character behaves, how honest or suspicious. I tried to use it as a character device rather than a branching plot device.
Also the story isn't controlled by any one character, and if that character needs to do something for plot reasons, they could do it as NPC.
I think TellTale was just bad