Well, part of that assumes that "state a goal and approach" was intended as a hard rule-like process of play and not just good guideline for how to encourage good interactive RP. The fact that you've reimagined at least one skill - insight - to make it more active and less passive suggests that what you call "in line with the other skills" is a more ubiquitous problem. For me, it's a non-problem. It's only a problem if you are insisting in applying a validation filter on all player propositions that they must "state a goal and an approach" and that otherwise it is not a proper proposition. Attempting to control the processes of play to achieve a particular game experience is a very post-Forge Indy like approach to the game, but I'm not sure that it is a necessary one. It's more of a preferred style.
The danger is that you end resolving a non-trivial number of propositions by fiat. You've basically forced yourself to adopt a process of play were all knowledge is either known (100% chance of success) or unknown (0% chance of success). But this does not account for PC background or PC downtime. It's entirely consistent to setting for a character that spent 8 years in some sort of apprenticeship, and maybe a couple of years practicing a profession before starting play to have acquired a large but not comprehensive body of knowledge about the setting that the player themselves simply doesn't have. One thing you might have gathered that I don't like is a situation where I'm effectively choosing before or during the session how to resolve doubtful propositions based on my meta-game knowledge of the PC's stats. For instance, I objected to passive perception versus static DC's, because - since I know the Pc's passive perception score - then whenever I place a challenge I'm in the position of deciding how I want the challenge to play out. And for me, knowledge is exactly the same issue. Sure, there is a certain amount of that going on, since I could always set the DC to be trivial (DC 5) or ridiculous (DC 45), but at least in that case what I'm supposed to be basing my decision on is whether it is knowledge 'known to practically everyone' or 'known to the gods, and maybe not all of them' and I can make that sort of estimation with confidence.
For example, when the party encounters a monster for the first time, it's often a homebrew monster, but a character with the right knowledge still might recognize the monster and know some facts about it. I would also allow a PC to go to the library (approach) to discover something about a monster (goal) using the same skill, but I don't automatically assume that in researching vampires last month the player didn't skim through some facts about mummies that weren't at the time relevant but which comes to mind now, or that their mentor during their apprenticeship didn't force them to write a 5 page essay on each of the common sorts of corporeal undead.
And the same general approach applies to all sorts of knowledge.
My general approach to this is whether the information could have ended up in a book or other body of common lore. So you can't but through play learn where a murder hid a murder weapon in an unsolved case, but pretty much anything that is known to someone and has historical or scientific significance could be known to you. Knowledge after all is testing your mastery of a body of lore.
I don't think they problem is so much finding 'good applications'. The problem is that you can no longer find immediate applications. And by removing their immediate utility and limiting them only to matters of research and learning, you are pretty much making them useless except in very slow deliberate investigative campaigns that are quite different from the usual 'quick the doors down, fight the monsters, and take their stuff' assumptions of play that the D&D rules have traditionally primarily supported. As such, you are basically eliminating the worth of 'the adventuring scholar' which might otherwise be a valuable archetype, and relegating knowledge skills to the sort of NPC sages described in 1e AD&D.
I never tell the players that they believe an NPC either. I tell the player that they are confident that that an NPC believes what they are saying, or doesn't believe what they are saying, or generally what emotion that the NPC is feeling and may be trying to describe, or broadly what motive seems to lie behind the NPC's actions. Rarely, an exceptionally unperceptive character may be deemed to have fumbled that their check and I tell them they are confident of something that is false, but that's not really a required process of play.
That is of interest to me too, but such a scene would be deemed at least as important as combat and would never be resolved with a single roll based off a single agenda and approach. That scene would be RPed out, with different rhetorical feints and parries and probably a raft of rolls contested or passive skills across a wide range of different proficiencies. And if the goal was 'expose them to some NPC' - convince the king his councilor is a scoundrel - then that might be determined ultimately by something cumulative to all the action. Just as an important combat might take an hour to resolve, so to we might spend an hour on the role-play of that scene, with dialogue ideally that could be turned into a screen play or novel of the scene that was so played out.
Yes, but ultimately that is a test of NPC vs. NPC that depends on the relationship between those NPCs and their perceptiveness is only influenced by the behavior and success of the PC. So a PC might could induce a reaction test between two NPCs at some modifier on the roll, or the PC get an NPC to confess to an NPC, but just because you've made a very convincing argument it doesn't force the NPC to recognize how convincing it was. Indeed, multiple NPCs hearing the same argument might behave differently. One NPC implicitly trusts their spouse regardless of what arguments you bring forward. Another NPC is a fool and believes the spouse for flimsy and illogical reasons. And a third might realize you've presented really damning evidence, while a fourth might believe that though the evidence isn't compelling, they recognize the subtle cues that the spouse is actually lying.
Define 'real' and we may be on the same page. I'm aware of the Forge meaning as well, but I won't discuss it because I can get myself in trouble when I discuss public figures in the gaming community critically. Suffice to say that your default, problematic though it is, is probably a more coherent definition than the Forge one. I think the Forge conversations were extremely valuable for the seriousness with which they examined game design as an intellectual exercise, but I think that much of what came out of GNS as something explored in and of itself is pretty worthless. By "simulationist" I mean really a game that treats exploration of setting as one of its principle aesthetics of play, and as such attempts to be internally consistent and tends to want to treat only the setting information as relevant to the adjudication of fortune.
Now, we could really digress here but one of the big ways that GNS goes wrong is it assumes that the three aesthetics of play they identified are mutually exclusive and so systems need to be designed to deliver that one aesthetic of play. However, I would argue that any game that attempts to do that very rapidly ceases to be an RPG (for which, I'd have to define an RPG). But, for example a purely simulationist game would not have players, and thus not have player characters, but only designers and observers. It would in fact be a model (or simulation) and not an RPG, because by the definition I provided a game that only and 100% was designed to accommodate that aesthetic wouldn't allow the player to influence the action because by the definition the player is outside of the in-game world and thus not a part of the adjudication of fortune.
I use different procedures of play to deal with those problems. But from a purely simulationist perspective there is nothing to "derail" and the game can't "grind to a halt". The story can grind to a halt and play can become uncompelling, slow paced, and uninteresting but properly simulation doesn't care about that. To solve those problems, I don't alter the rules, but alter the refereeing - the conditions of the model or fiction as it were. The game engine remains simulationist but the story teller alters the fiction to deal with non-procedural problems like weak drama, slow pacing, etc.
