I have long felt that floats should default to NaN, so that any attempt to perform operations with them before they're initialized results in an error.