Raph "Theory of Fun" Koster has a wonderful, readable, theory-rich article that helps unpick the discussion about when a game is a clone of another game, when it's a skin, when it's a variant, and when it's a new game.
The article ends with an inspiring prescription for understanding how to synthesize new games using the best primitives from existing successful games and game-mechanics. This is Koster at his best, and why he's one of the leading theorists of game development.
So, the recipe for inventing a truly new game:
* Identify a new mathematical model. This is often done by finding a new kind of scenario to model: human relationships (The Sims), gardening (Farm Town), etc.
* Proffer a dimensional change on an existing ruleset, such as Tetris modifying the classic game of pentominoes by adding time and movement vector. Pac-Man and Miner 2049er and Flip & Flop are almost the same game (traverse every node on the graph). But the rule changes are major.
* Explore alternate sorts of graph structures, such as Blokus to Blokus Trigon or Gemblo. Jumpman vs Miner 2049er is a good example here, or indeed any other “gather things” platformer; changing the graph of points that require visiting alters much.
* Offer a replacement goal within an extant rule structure, which can force a major variant. A racing game versus a demolition derby sort of racing game is an example here.
When is a Clone [Raph Koster]