I made a few refinements to betzaNew.js:

• I added a parameter clipX=N. This can be used to inform the AI that the board really only has N files, and that any additional files defined by the files parameter are just for creating areas to store pieces that can be dropped or gated onto the board. Without this the AI's idea of centralizing pieces became mixed up in variants with extensive board extensions. By default the files=N command would also set clipX to N. But use of an explicit clipX after that can alter that value.
• Bracket notation now also understands a step (W or F) followed by a single non-oblique leap (A, D, G, H).
• I fixed some problems with use of a (for 'all directions') in bracket notation. 'All directions' is the default continuation direction in XBetza multi-leg moves, but it is a relative notion there. When the atom on which it acts has only 4 directions 'all' stays limited to these original 4 (or 3, if the 4 would include strictly backward). When the path included 45-degree bends, like needed for expanding [W-aF], an explicit vs is needed in XBetza: avsW. The old preprocessor failed to add that, so that the result became that for [W-aW] instead.
• As to elimination of always succeeding (i.e irrespective of square occupancy) intermediate legs in a multi-leg move, by merging those with the following leap, I now use the following system: intermediates inserted by expansion of bracket notation will always be eliminated. When no captureMatrix is defined, mp leaps will be eliminated when there is no hole defined amongst the pieces, and mpo leaps when there are. This is a bit course, but erring in the safe direction. More accurate would be to test whether the capture matrix actually specifies any hop bans, or better yet, whether it specifies those for the piece the move belongs to. Also note that it does not test whether there actually are any holes in the board; just whether the hole type is defined. So if you want to use a multi-leg move to test whether there is remaining space on board by a back-and-forth mp step on one of the pieces, you can add a line hole:::: to force the Diagram to leave mp legs alone.
• The value parameter for setting piece values by hand now can accept a value relative to an earlier piece. When you write value=V/N after the definition line of a piece, that piece will get the value of piece number N, plus V. This feature is provided to handle pieces that start off board. Usually two piece types have to be provided for this, where the in-hand type has some special move for introducing it on the board, where it then morphs into the type with its normal move. The value of the in-hand pieces should then not be related to this introduction move, but must be slightly under that. So that introduction of the piece is encouraged, but not at all cost. Due to the stochastic nature of the heuristic process of value determination, it is hard to achieve that by a fixed value.
• I am not sure whether I ever announced the following: there also is a shooter=T parameter that can be used after a definition line for pieces that can make locust captures. The 'shooter type' T (0, 1 or 2) determines how the locust capture should be entered with the mouse. T=0 would indicate a 'trampler', i.e. a piece that first moves on top of its victim, and then with a third click to its destination. T=2 is for 'shooters', whic first move to their destination, and from there take out a victim with a 3rd click. The Diagram attempts to make this distinction automatically, but this doesn't always result in the desired classification. Finally, T=2 (which can be set only by hand) indicates implied locust capture, where you only have to click the final destination, and all side effects are performed automatically. This can only be used on pieces where the destination doesn't leave any choice about what the side effects will be. (Comparable to entering castling or e.p. capture.)