A Chess variant with low draw rate

[hgm]

960 is considered a variant, isn't it? It seems this is along the same lines.

I don't consider Chess960 a variant. Nothing is really different there. It is some sort of generalization, though, as rules have to be clarified in some situations that cannot occur in orthodox Chess.

That of course doesn't mean that it cannot have its own name; there are any sub-variants of 10x8 Chess that only differ in start position, and each go by a different name (Capablanca, Gothic, Bird, Carrera, Schoolbook, Embassy).

[Nordlandia]

I dunno why 960 ain't so popular. I mean classical chess is always on the agenda.

[Ovyron]

I don't consider Chess960 a variant.

Do you have a clear, objective line drawn that separates those that are variants from those that aren't?

It's difficult to know what'd you consider a variant or not.

For instance, I create the following:

Whenever a King returns to its starting position (e1 for the white king, e8 for the black one) and there's a rook it can castle with, or a rook returns to its starting position (a1/h1/a8/h8) and there's a king in its starting square, castling rights become available again, and castling can be played multiple times in the same game.

Is this a variant? If not, what is it?

[hgm]

I guess there will always be boundary cases, no matter what definition you employ. On chessvariats.com they use the concept of 'modest variant', for something that only differs from orthodox Chess in some minor detail. (Like having stalemate as a win, lacking e.p. capture or limiting promotion choice.) I would say what you describe qualifies as such a 'modest variant'.

I percieve this as more different from orthodox Chess than Chess960. Common positions in the respective game trees are limited to those where there are no Rooks at all; it is like the Rook has been replaced by another piece. The Chess and Chess960 game trees differ much less; there it is just that for any position with a King and Rook on the back rank Chess960 also has a version of that position where castling rights still exist. That is only a small minority of the positions, and it is only as large as it is because you consider all 960 start positions; if you would limit yourself to one Chess960 start position the King and Rook would have to be on very specific squares to be able to have castling rights. So considering Chess960 as a single sub-variant, rather than 960 different sub-variants, already skews matters.

A somewhat similar question is whether Janus Chess (featuring two Archbishops and no Chancellor) is a sub-variant of 10x8 Chess like Capablanca Chess, or a separate variant. I always classified it as a separate variants, because Pawns cannot promote to Chancellor. So any position with Pawns would be different, the Pawns are really a different piece type.

[Ovyron]

I guess there will always be boundary cases, no matter what definition you employ.

Challenge accepted, I submit this one:

It's a chess variant if you need to add extra code to chess engines to play it, or need to make modification to a GUI to replay games.

There you go, no boundaries or edge cases, and Chess960 is a variant, because current engines would not accept some castlings as a legal move, and wouldn't be playing the variant if their tree were missing potential castling. Plus, you can't just paste a game into your common GUI to replay it.

The positions in Chess960 that have kings and rooks in their starting positions wouldn't be a variant.

[hgm]

I don't think that is a valid definition, because it makes a prior assumption on how Chess GUIs for orthodox Chess are supposed to work. It is conceivable that a GUI for orthodox Chess happened to implement castling in such a way that it automatically works for Fischer castling. Furthermore, I don't think that notation issues (such as castling rights in FEN) should ever have any influence on whether something is a variant or not. Notation conventions are not part of the game rules.

As a more realistic example: a GUI that doesn't check legality (such as early versions of WinBoard) is able to handle any variant on 8x8 board where each side has 8 + 2 + 2 + 2 + 1 + 1 pieces, without any code modification.

Fairy-Max does not need extra code to play, say, Knightmate. It just requires other data in its configuration file.

A real-life example is the ICC variant TwoKings. This was created by setting up a position with extra Kings in an ICS designed for normal Chess only. This of course gave rise to all kind of unexpected and illogical behavior, which was then declared to be the game rules.

I guess the main point is that you cannot predict how Chess software would behave when you present it with positions that could never occur in the game they were designed to handle. Such as multiple Kings, Pawns on the first rank, or castling rights with Kings and Rooks in unusual positions. Basing a definition on what engines happen to do in the novel positions would not give a unique verdict.