The game posits an unmediated access to consistent effects, which get labelled as "good" and "evil" in the metagame--but there's no need for a PC or even a player to accept that label.
I agree at least with the point in bold, and I've been saying something very much like it for ten pages or so.
What, then, do you think that Detect Good or Detect Evil (pre-5e version) is detecting?
It's not detecting value per se - because you deny that the torturing villain, or succubus, is obliged to accept the deliverances of the spell as stating evaluative truths.
[MENTION=2067]Kamikaze Midget[/MENTION] suggested, upthread, that it is detecting X, Y and Z - but then characterised X, Y and Z using evaluative language. You seemed to do the same thing when you referred to "petty reasons".
And when you recharacterise your villain as ""Yeah, I probably shouldn't have lost my temper and tortured that lady for losing my laundry. Oh well, we all have off days." you have a villain who is not the hero of his adventure at all! He's conceding error. (Though is failing to manifest contrition.)
I don't know that the language analogy is helping
It's not an analogy.
You said that, in PS, the truth of your evaluative Xs, Ys and Zs (say, "beauty" or "doing right by others") is determined by majority belief. Beliefs are mental states with propositions as their contents. So we can't talk about shared beliefs until we have an account of proposition individuation, which will require an account of concept individuation.
In the case of chairs and tables, the account of concept individuation is fairly straightforward: my word "chair" and the French word "la chaise" are synonyms because they are both introduced into their respective languages by reference to things of the same kind (namely, chairs, with which we interact in all sorts of mundane and non-spooky ways).
In the case of evaluative language, the account of concept individuation is hugely contested, and is perhaps the longest ongoing intellectual debate in human history. Some people think that you work out the French word for "good" by seeing what word, in the French language, serves as a general term of praise and commendation (which is the basic use of the English word "good"). But in that case, to judge something good is to judge it as praiseworthy - which won't work for the framework you're trying to set up, because you want the demon to be able to agree that the Seven Heavens is X, Y and Z yet deny that it is praiseworthy.
An alternative approach would be to identify some non-evaluative X, Y and Z to be a supervenience base for the evaluative terms. The debate about whether or not this is possible for the social sciences is over 100 years old and still going strong. I've certainly never seen anyone even try and do it for D&D alignments, and even if you did you would likely get a contentious list of factors which - whether or not it was accurate as a matter of fundamental analysis - would not necessarily work for play purposes (because it is contentious).
Therefore, the coherence of the position that (a) Detect Good/Evil delivers uniform results, because (b) the X, Y and Z are intersubjectively accessible features of the scanned entities/places, because (c) the X, Y and Z are determined by majority belief, is not easily demonstrated. You need an account of concept individuation for the Xs, Ys and Zs that doesn't depend upon a presupposition of being admirable/undesirable. That's not easy to do.
If we put to one side the worries about concept individuation, though, then it's fairly easy to see how we get (d) anyone is free to deny that the results of Detect Good/Evil has no necessary bearing on his/her self-evaluation. Because all that amounts to is rejecting the majority consensus in respect of <something - not sure what, because I've put the issue of concept individuation to one side>.
Which gives rise to (e) can evaluative judgements, in this system, be in any way non-arbitrary or can play some sort of justificatory or reason-giving role in a person's life? I don't see how. And given that Bahamut has a 20+ INT, presumably he can reach the same conclusion. How does he differentiate the killings he urges his paladins to undertake from the murders by demons that he condemns? As I posted upthread, the whole thing is very much in danger of collapsing into teams, with no moral asymmetry at all. Which, again, Bahamut, with his 20+ INT, can notice.
Part of the issue is that it is not the case that items (i)-(iii) of your definition necessarily warrant admiration. In example, any economist would balk at anything happening without reciprocation
That's not true of "any economist" - not all economists subscribe to psychological egoism.
But that to one side, of course it's not self-evident that (i) through (iii) are admirable: Ayn Rand may not be correct, but she hardly failed to notice a self-evident truth. My point is that if you don't think they're admirable then it is not at all clear that you think they are markers of generosity, given that to decribe something as "generous" carries with it an implication of worthiness of admiration. But if you don't think they're markers of generosity, then how are you going to participate in the great social consensus that establishes the meaning of the Xs, Ys and Zs?
They can accept the moral aspect of that definition or reject it as they see fit, and of course this means that the social context will judge them as well and define what they are
What does "define" mean here? It can't mean
make true - because in that case they
couldn't (rationally) accept or reject as they see fit. (Contrast: the positions of stellar and planetary masses "define" the shortest distances along which light rays travel - as with [MENTION=22779]Hussar[/MENTION]'s example of objective gravity and boiling point of water, light rays
aren't free to accept or reject the consequences of general relativity as they see fit; and rational physicists are obliged to bring their beliefs into conformity with reality.)
(unless they can change it)
<snip>
Change the opinion, and reality will work differently.
What is at stake in changing anything? I mean, if I make everyone believe that shiny armour is ugly and saving princesses is ignoble, and that acidic corrosion is beautiful and wanton killing noble, paladins will still go to the Seven Heavens when they die, and the Abyss and the Seven Heavens will still be at loggerheads. Perhaps, because the X, Y and Z for good and evil are held constant, Detect Good will now ping on demons and Detect Evil on paladins, but I don't see how that is of any great interest.
I certainly don't have any sense of what it means to say that paladins used to be good but now are evil, when
no behaviour in the world has actually changed, and
I haven't changed my mind about anything either.
The characters in a PS game exist in a setting where social consensus has defined alignments and what they mean
<snip>
Characters are free to dispute that consensus, and it is assumed that they will, at least in some form or another - that's the "things that need changing" in the 3-act structure of a PS campaign.
I'm not confused about what PS asserts. I'm expressing doubts about its coherence - both conceptual and practical.
If we solve the conceptual problem of even establishing what the global consensus is - far from trivial, because of the complexities and disputes around concept-individuation for evaluative terms - we get to the practical question: who cares what the consensus is? If I'm a paladin, and my goal is to relieve suffering and vindicate the innocent, why do I care what labels the majority of people use to describe my conduct (unless I'm guilty of the sin of vanity)?
And to leverage this practical question back to a conceptual one: if it is not per se irrational to value things differently from the global consensus, then why does the content of the consensus have any relevance to anyone? If there's no reason per se to have regard to it in one's own practical deliberations then it is arbitrary, and hence not a rational object of concern.