In general, there's necessarily a right answer for selecting one over the other in Godot. There's probably a measurable typically negligible performance impact to adding an extra layer of indirection in the scenes. But to my in mind, selecting one over the other is largely a matter of selecting the relationships that make it easier to manage the complexity of developing the game.

To me, using the primary "idea" as the parent node type is represents an "is a" type relationship whereas including one node as a child of another represents the "has a" relationship (also known as composition). And keep in mind that TileMap and Sprite2D (and any other node of the 2D category that I bothered to look up) inherit from Node2D. Since they are already specialized versions on Node2D, they have a built in "is a" type relationship with Node2D.

In Godot, I almost always lead with the main idea of a thing to select the "is a" relationship for the parent node. For the player and enemies I used KinematicBody2D, for hit boxes and hurt boxes I used Area2D and so forth because that's how I think of them. The player isn't 2d node with a kinematic body attached to it, it is the kinematic body. This also saves me the the extra mental overhead of needing to dig an extra layer into everything to get what I need. By that I mean since the player is a KinematicBody2D I can call move_and_slide(..) directly. If I used the "has a" relationship I would need to first "get" the KinematicBody2D that belonged to the player.

Sometimes though there's not a clear idea. Usually this happens for more abstract concepts that are a collection of things, none of which is more the main idea than the others. An example might be an explosion effect consisting of a sound, sprite & animation player. None of those parts better represents the idea of the explosion effect more than others. So in that case I used a Node2D and added the other nodes as children. Another example is the "main" or "world" scene. It's more a collection of stuff than it is the embodiment of one particular thing or idea. For that I would use Node because it doesn't need the position, rotation, etc that comes with Node2D.

Disclaimer: I've only used GDScript with Godot. One of the other bindings might lead me to a more composition heavy design as that's tended to be my preference with other engines / frameworks.

In general, there's not necessarily a right answer for selecting one over the other in Godot. There's probably a measurable typically negligible performance impact to adding an extra layer of indirection in the scenes. But to my in mind, selecting one over the other is largely a matter of selecting the relationships that make it easier to manage the complexity of developing the game.

To me, using the primary "idea" as the parent node type is represents an "is a" type relationship whereas including one node as a child of another represents the "has a" relationship (also known as composition). And keep in mind that TileMap and Sprite2D (and any other node of the 2D category that I bothered to look up) inherit from Node2D. Since they are already specialized versions on Node2D, they have a built in "is a" type relationship with Node2D.

In Godot, I almost always lead with the main idea of a thing to select the "is a" relationship for the parent node. For the player and enemies I used KinematicBody2D, for hit boxes and hurt boxes I used Area2D and so forth because that's how I think of them. The player isn't 2d node with a kinematic body attached to it, it is the kinematic body. This also saves me the the extra mental overhead of needing to dig an extra layer into everything to get what I need. By that I mean since the player is a KinematicBody2D I can call move_and_slide(..) directly. If I used the "has a" relationship I would need to first "get" the KinematicBody2D that belonged to the player.

Sometimes though there's not a clear idea. Usually this happens for more abstract concepts that are a collection of things, none of which is more the main idea than the others. An example might be an explosion effect consisting of a sound, sprite & animation player. None of those parts better represents the idea of the explosion effect more than others. So in that case I used a Node2D and added the other nodes as children. Another example is the "main" or "world" scene. It's more a collection of stuff than it is the embodiment of one particular thing or idea. For that I would use Node because it doesn't need the position, rotation, etc that comes with Node2D.

Disclaimer: I've only used GDScript with Godot. One of the other bindings might lead me to a more composition heavy design as that's tended to be my preference with other engines / frameworks.