The nit I am going to pick is slightly tangential to your point, but not to my participation in this thread: this claim is true only if the range of events and actions over which choices extends is rather narrowly constrained.
This is quite fair: the set of choices that "matter" in a mechanical sense is constrained. I dunno if "rather narrowly" is necessarily true, but that's certainly an area where objective statements are dubious.
For instance, in "classic"/traditional approaches to D&D: a player's choice of his/her PC's attire (other than armour) is not supposed to matter. Nor choice of romantic interest. Choice of religious affiliation is meant to matter in something like, in the real world, choice of trade union membership matters; but it is not supposed to matter in the way which, in the real world, private/inner religious conviction matters.
Well, I'd say the "other than armor" caveat is rather an important one, but even without that, clothing and self-presentation
do matter (or at least they have in several games I've played, including the one LL game). Wear your muck-covered, bloodstained armor to an audience with the Queen, and you'd better hope she's a warlord at heart or you'll almost certainly take some mechanically-relevant penalties for that choice. (In my DW game, I could
easily see this requiring a Defy Danger CHA to avoid losing some clout in the Sultana's court due to not observing etiquette). I'd also argue that religious affiliation in the latter sense matters...but only for some characters (
mostly Divine-source classes, but consider the significance of religion in a campaign setting like
Zeitgeist, or Eberron.)
The choices that are canonically supposed to matter are choices that pertain to travel, architecture, and certain sorts of resource expenditure. But as this thread has brought out, the parameters within which these choices matter can be surprisingly narrow: eg while choice of (say) class or skill, and choice of which PC ability to bring to bear, is generally supposed to matter, does it always? Eg does it matter if a PC engages a NPC leading with CHA (Persuasion) rather than CHA (Intimidation)? Sometimes it does, but I think that sometimes it doesn't.
I would find it unusual if the choice to persuade vs intimidate even
sometimes didn't matter, but then again, I'm a fan of both Dungeon World and 4e, so I'm one of those "weird" people who values both consistent fiction
and mechanical rigor. Change the fiction, change the result, IMO. And I'd also say that a common area where the skills
aren't in practice treated differently is a good counter to this argument, as both the rules themselves and "best practices" from 5e DMs say that the common behavior is incorrect: Perception vs Investigation. The former is, in theory, supposed to be purely for the physical act of observing--can your eyes see the lettering carved in the wall, can your nose discern which specific
kind of incense it smells, etc.--while the latter is supposed to be used for integrating a set of observations into a meaningful conclusion (e.g. Perception determines whether you notice that one of the books on the shelf is slightly less dusty, while Investigation determines whether you realize that the room is shorter internally than it looks externally, and thus has a secret chamber.)
But there are some other areas I think you're leaving out. Anything to do with academic knowledge is relevant--it's why there are several academic skills. Athletics vs Acrobatics is another meaningful choice area, with the two being
sometimes interchangeable but never
equivalent. Comporting yourself toward others has been relevant in D&D since forever, it's why Charisma was important for Reaction rolls, and as the "story about adventurers" aspect of D&D grew and the "ruthless mercs looking to score big in a murder-hole" aspect waned, I'd argue it's only gained importance, hence why a "Face" is an important thing for any party to have.
The idea of "fixed" setting elements that a player can learn about and make reasonable causal extrapolations in respect of works well for: treasure maps, architecture, certain very simple (simplistic?) social structures (just to give some examples). It works poorly for: dynamic social situations, alliances and betrayals, complex human geography, truly magical places (just to give some different examples). It can even work poorly for contests between reasonably evenly matched opponents - eg races, chess games, etc.
I think I gave an incorrect impression here. I don't mean these things are fixed
for all time, never to be changed or revealed as different. What I mean is that if something is evidenced within the game, then either it should be true, or the players should have the
chance to learn that it is not true. Maybe they fail, or maybe they simply don't bother trying to find out. Many of the areas you refer to as exceptions...aren't things that have been put into evidence yet, as it were. The outcome of a race is probably uncertain, not because there's no evidence, but because dice rolling is likely to be involved. Dynamic social situations are ones where the evidence itself demonstrates that what is true
today need not be true
tomorrow--unless the DM decides midway through that what
wasn't a dynamic situation initially now IS a dynamic situation.
That, truly, is what I'm talking about. Shifting alliances are fine. Alliances that
in-character secretly shift around are also fine, because it is perfectly reasonable for
characters to tell lies or deceive others. The example I like to give here is a murder mystery. Let's say that the Duchess is the one who murdered the Count, but both she and the Baron are suspects. The party does something clever and unexpected, which makes the party
extremely sure that the Duchess is the guilty party...but this would derail the intended plotline and prevent the "Baron begs the party for help in desperation" scene the DM had planned. Instead of adapting around the players' choices, the DM decides that now the Duchess is innocent and the Baron really IS guilty (simply switching who plays which role)...even though up to this point it was literally impossible for the players to learn this
because it wasn't true before. Now they have evidence which, at the time they got it, was perfectly correct and sound, and they have no real way to know that that evidence
became unsound, as that would reveal the DM's hand and probably piss off the players.
That is what I mean by something that shouldn't change unless the players at least have the
possibility of learning how and why it changed. Another example would be deciding that one of the party's close, trusted allies--someone who
genuinely passed efforts to vet them--is now suddenly the traitor in their midst, even though the party literally couldn't have learned that because that ally WASN'T the traitor up until the moment the DM decided they were. (Other examples include altering a monster's statblock mid-fight, changing the rules of engagement, and violating established setting elements--up to and including the nature of magic, even though that can be a fickle thing!)