Whether we think of these elements as game mechanics with attached story or story elements with attached mechanics, I think the main thing is that, at a basic level, we often don't have a design pattern/philosophy which would help elevate the combination of the disparate parts to an emergent and coherent whole. At some level I don't even care what that pattern is if it is discernible, and we recognize the tradeoffs different patterns make. The question is then which set of tradeoffs best suits the intended use?
5e looks like it will have the following basic elements: ability scores, race, background, class, subclass, feats (optional), and skills (optional). What are the ways we can elevate that structure such that it contributes something to the game a pure character-point system (for example) might not be able to easily achieve?
Consider, for example, the the one-way vs. many-ways approach to building a character toward some archetype. I don't think either is necessarily better, but I do think we should consider their implications. If we use a many-ways approach but lack a way to express whether a given element is more appropriate as a feat or a class feature (using whatever criterion) we weaken the entire structure (because what is the point of having two structures if they are indistinguishable?) even though there is greater possible richness in principle. A one-way approach can have more inherent clarity, but loses potential richness. So, a many-ways approach should probably enforce some stronger guidelines on what game elements "mean" compared to a one-way approach so it doesn't collapse under its own weight.
This is where feats, in particular, are such a tricky bit. In the L&L from a few weeks ago Mearls mentioned that the catch-all nature of feats was a feature, not a bug. If we embrace that I think it really behooves the designers to tighten up the "meaning" (whether in terms of mechanics, story, or both) of the other elements so we avoid having multiple catch-alls. On the other hand, if feats are given a narrower purpose the meaning of class can probably be loosened somewhat. The fact that feats are meant to be optional just makes it trickier.
Assuming my thoughts above are not a load of crap, I think I might try for a breakdown somewhat as follows.
1. Ability scores primarily represent unchanging natural talent plus non-specific competency. If skills are not used this is sufficient since broad-based competency is all that is represented in play, while if skills are used this is still sufficient because the skills represent specific competency. The fundamental mechanics of ability scores are fixed and universal, although other game elements may leverage them in unique ways.
2. Race represents inherent in-born ability and ubiquitous cultural influence. (I'd personally like to separate in-born ability and ubiquitous cultural influence, but that's another discussion.) Because the scope and story of race is quite clear, and because one can only have a single race, the mechanical specifics are fairly unconstrained except for balance purposes.
3. Background represents vocation, non-ubiquitous cultural influence, and/or social station. A character will in general have only a single background, but the scope, story, and campaign dependence is highly variable. They speak more to what one does than how one does it. All backgrounds give some basic mechanical proficiencies and a special benefit, but they carefully avoid entanglement, overlap, and/or dependence with the mechanical abilities of other elements like race or class. They can, however, complement these other elements.
4. Classes represent the fundamental way one interacts with the broadest levels of reality (the gods, the nature of the mind, physics, the planes but not a plane) and introduce a small number of fundamental mechanics which form the basis for these interactions from start to end. Classes that mix these broad elements (e.g. the paladin) need a wholly unique mechanic or a unique interaction of existing mechanics. Class doesn't tell the story of a character, it shows the type of story for that character. Although the focus is on mechanics and broad story elements, it bears remembering that exactly what counts as a distinct "level of reality" depends very much on the kinds of stories told in that world and the cosmology in which they occur. D&D makes a strong distinction between arcane and divine magic, for example, but in another game it might not be so.
5. Subclasses represent the most distinguishing nuances in how a class interacts with reality. They reflect specific deities, iconic magical traditions/communities, schools of thought, fighting styles, single planes, civilizations, countercultures, and more. They don't simply add to or modify the fundamental mechanics of a class, they suffuse it. Mechanics independent of the core class mechanics (and especially those without similarly long-term scope) are strongly discouraged.
6. Feats are optional and largely self-contained story packages with limited mechanical scope (in terms of interaction with other elements, not necessarily power). They change how one plays the game, but they do not usually change how one plays a class. Feat chains are discouraged: if the feats do not share mechanics they should tell different (if complementary) stories, and if they do share mechanics a new class or subclass may be the proper place for the idea. Feats that rely on or modify a class' fundamental mechanics are the would-be class or subclass features which do not sufficiently suffuse that mechanic. Feats are an appropriate way to meaningfully expand race, background, and skills.
7. Skills are an optional element that fine-grain the scope (both mechanically and story-wise) of ability checks, and like ability checks work the same from player to player when in use. Other elements may leverage skills in the same way they leverage ability checks.
For something like the assassin I think the above gives sufficient guidelines about when or if something is appropriate. Suppose we envision the death attack as a special observation-based sneak attack, such that the assassin should be a rogue subclass. An assassin background might emphasize social connections with appropriate elements (whether criminal, national, or religious). A feat might introduce very flexible poison use, which is useful to rogue(assassins) and others as well without overlapping core identity. A different feat might allow selecting multiple backgrounds for the purpose of creating cover identities, useful to many characters that want to practice subterfuge. For abilities that utilize the subclass' death attack we have guidance as to whether it is better as a subclass feature (usually) or a feat (less often). For example, an ability like "Always Watching" might grant the assassin significant flexibility on what counts as observation for the purposes of the death attack, and since this would likely come into play all the time it would be an appropriate subclass feature. An ability like "Shadowing" might grant the assassin the ability to designate and stalk a specific target over a longer period of time (i.e. days) for some benefit when maneuvering into the critical position or actually executing the hit, but because not all assassins are assumed to be stalking a single creature over the long term this might work well as a feat. Such a feat could be powerful, even opening up new ways to play, while remaining secondary to the assassin subclass' core identity. Finally, a later class might define a whole range of assassiny types starting from a broad story and mechanic separate from the observe-you-to-death rogue(assassin). Existing backgrounds and feats independent of rogue(assassin) mechanics would still be perfectly useful, but the rest could stay in their own story and mechanical space. Hopefully something like this would encourage the richness possible in the game without leading to the feat or class feature free-for-all of past editions.
5e looks like it will have the following basic elements: ability scores, race, background, class, subclass, feats (optional), and skills (optional). What are the ways we can elevate that structure such that it contributes something to the game a pure character-point system (for example) might not be able to easily achieve?
Consider, for example, the the one-way vs. many-ways approach to building a character toward some archetype. I don't think either is necessarily better, but I do think we should consider their implications. If we use a many-ways approach but lack a way to express whether a given element is more appropriate as a feat or a class feature (using whatever criterion) we weaken the entire structure (because what is the point of having two structures if they are indistinguishable?) even though there is greater possible richness in principle. A one-way approach can have more inherent clarity, but loses potential richness. So, a many-ways approach should probably enforce some stronger guidelines on what game elements "mean" compared to a one-way approach so it doesn't collapse under its own weight.
This is where feats, in particular, are such a tricky bit. In the L&L from a few weeks ago Mearls mentioned that the catch-all nature of feats was a feature, not a bug. If we embrace that I think it really behooves the designers to tighten up the "meaning" (whether in terms of mechanics, story, or both) of the other elements so we avoid having multiple catch-alls. On the other hand, if feats are given a narrower purpose the meaning of class can probably be loosened somewhat. The fact that feats are meant to be optional just makes it trickier.
Assuming my thoughts above are not a load of crap, I think I might try for a breakdown somewhat as follows.
1. Ability scores primarily represent unchanging natural talent plus non-specific competency. If skills are not used this is sufficient since broad-based competency is all that is represented in play, while if skills are used this is still sufficient because the skills represent specific competency. The fundamental mechanics of ability scores are fixed and universal, although other game elements may leverage them in unique ways.
2. Race represents inherent in-born ability and ubiquitous cultural influence. (I'd personally like to separate in-born ability and ubiquitous cultural influence, but that's another discussion.) Because the scope and story of race is quite clear, and because one can only have a single race, the mechanical specifics are fairly unconstrained except for balance purposes.
3. Background represents vocation, non-ubiquitous cultural influence, and/or social station. A character will in general have only a single background, but the scope, story, and campaign dependence is highly variable. They speak more to what one does than how one does it. All backgrounds give some basic mechanical proficiencies and a special benefit, but they carefully avoid entanglement, overlap, and/or dependence with the mechanical abilities of other elements like race or class. They can, however, complement these other elements.
4. Classes represent the fundamental way one interacts with the broadest levels of reality (the gods, the nature of the mind, physics, the planes but not a plane) and introduce a small number of fundamental mechanics which form the basis for these interactions from start to end. Classes that mix these broad elements (e.g. the paladin) need a wholly unique mechanic or a unique interaction of existing mechanics. Class doesn't tell the story of a character, it shows the type of story for that character. Although the focus is on mechanics and broad story elements, it bears remembering that exactly what counts as a distinct "level of reality" depends very much on the kinds of stories told in that world and the cosmology in which they occur. D&D makes a strong distinction between arcane and divine magic, for example, but in another game it might not be so.
5. Subclasses represent the most distinguishing nuances in how a class interacts with reality. They reflect specific deities, iconic magical traditions/communities, schools of thought, fighting styles, single planes, civilizations, countercultures, and more. They don't simply add to or modify the fundamental mechanics of a class, they suffuse it. Mechanics independent of the core class mechanics (and especially those without similarly long-term scope) are strongly discouraged.
6. Feats are optional and largely self-contained story packages with limited mechanical scope (in terms of interaction with other elements, not necessarily power). They change how one plays the game, but they do not usually change how one plays a class. Feat chains are discouraged: if the feats do not share mechanics they should tell different (if complementary) stories, and if they do share mechanics a new class or subclass may be the proper place for the idea. Feats that rely on or modify a class' fundamental mechanics are the would-be class or subclass features which do not sufficiently suffuse that mechanic. Feats are an appropriate way to meaningfully expand race, background, and skills.
7. Skills are an optional element that fine-grain the scope (both mechanically and story-wise) of ability checks, and like ability checks work the same from player to player when in use. Other elements may leverage skills in the same way they leverage ability checks.
For something like the assassin I think the above gives sufficient guidelines about when or if something is appropriate. Suppose we envision the death attack as a special observation-based sneak attack, such that the assassin should be a rogue subclass. An assassin background might emphasize social connections with appropriate elements (whether criminal, national, or religious). A feat might introduce very flexible poison use, which is useful to rogue(assassins) and others as well without overlapping core identity. A different feat might allow selecting multiple backgrounds for the purpose of creating cover identities, useful to many characters that want to practice subterfuge. For abilities that utilize the subclass' death attack we have guidance as to whether it is better as a subclass feature (usually) or a feat (less often). For example, an ability like "Always Watching" might grant the assassin significant flexibility on what counts as observation for the purposes of the death attack, and since this would likely come into play all the time it would be an appropriate subclass feature. An ability like "Shadowing" might grant the assassin the ability to designate and stalk a specific target over a longer period of time (i.e. days) for some benefit when maneuvering into the critical position or actually executing the hit, but because not all assassins are assumed to be stalking a single creature over the long term this might work well as a feat. Such a feat could be powerful, even opening up new ways to play, while remaining secondary to the assassin subclass' core identity. Finally, a later class might define a whole range of assassiny types starting from a broad story and mechanic separate from the observe-you-to-death rogue(assassin). Existing backgrounds and feats independent of rogue(assassin) mechanics would still be perfectly useful, but the rest could stay in their own story and mechanical space. Hopefully something like this would encourage the richness possible in the game without leading to the feat or class feature free-for-all of past editions.
I'm not sure what you're trying to say here vis-a-vis the naming of subclasses.
