Well, let's dive into that then.
I find the "map and key" phrase reasonably constructive. In many cases, the play is rooted in what the DM already, secretly, knows to be true, but which the players are ignorant of. They must provide inputs until they, too, know what is true and can thus make a properly informed decision. Since they cannot know how much they don't know, this is analogous to making a map of an area as one explores it. Just as with an actual map, there is both a fact of the matter (the territory which the map summarizes) and a range of relevance (one does not need the whole globe to navigate a single ruined city or spelunk a single cave.)
To add further interest, there are elements which are not what they superficially appear to be. The players cannot simply coast on GM narration; they must ask good, probing, effective questions and gain a full understanding of the place being mapped, not just a superficial awareness thereof. This is why the key is relevant, the coded parts of the map that grant full understanding of the location. Without the map, the key is useless, just symbols that don't point to anything; without the key, the map is incomplete, just a superficial description.
The great strength of this approach is also its great weakness: player ignorance, and the process of changing that to player knowledge, is the driving force of play. Firstly, it is dependent on players enjoying this back-and-forth process of figuring out what the right question(s) to ask would be for each case, and then making wise decisions based on the answers. (To be clear, I think lots of people DO enjoy this; I'm not trying to imply that that's weird or a bad expectation. Just that it is a prerequisite, and is not always true.) Secondly, and more relevantly for my above "look at technique and execution" post, "map and key" play depends on striking the right balance between discoverability (figuring out the right questions to ask must be practically achievable by the players) and difficulty (the right questions cannot be trivial.) This is a tricky line to walk! That it can easily fall prey to either extreme is one reason why designers might look to other approaches.
And we can again use video games as a good guide here for places where execution can be better or worse. Classic adventure games, e.g. Sierra and LucasArts adventure games, are effectively map-and-key games played with a preprogrammed story and automated DM. Some of these games are overall really, really well-made, e.g. King's Quest VI, The Dig, or Indiana Jones and the Fate of Atlantis. Some of them are notoriously bad either with specific puzzles (e.g. the TVTropes "Soup Can Puzzle" page examples) or with their general structure (relying on "Moon Logic" or even "Insane Troll Logic.") By examining these things, we can learn what techniques or features are shared by effective examples (or at least what considerations matter for building them), and how ineffective examples fall short.
And yet, as I said, we have what clearly appear to be knowledge claims regarding the effectiveness of game design decisions. We have (for instance) the widespread recognition that GP=XP achieves the "fantasy heist" intent of early D&D in a particularly deft way, or the less widespread but still common appreciation for how elegant and effective 13A's Escalation Die is for addressing the known issue of "nova" strategies being excessively dominant. How do we reconcile these (seemingly) blatant knowledge claims with the idea that it is impossible to achieve even the smallest amount of knowledge regarding game design? If it is truly impossible to learn anything at all about game design, why is it so like things we can make knowledge claims about?
