Eigentlich nur Punkt 2.3 könnte ein Problem sein bzw. werden, wenn es das ganze Match über anhält oder sich gar noch mehr häuft. Jetzt habe ich mir die Bewertungsprofile dieser Partien (3,4,7,8,11,12) angesehen: Unverdächtig. Die Bewertungen waren ausgangs der Eröffnungsvorgabe überwiegend ziemlich klein(*), oder wurden mitunter später zwischendurch noch etwas kleiner. Außer in 7+8, das war Königsgambit das beide als Weiße nicht rechtfertigen konnten.

*) Wobei nicht zu vergessen ist, daß Stocki quasi doppelt so große Evals ausspuckt als von anderen Engines gewohnt.

Ich bin sicher daß Nelson Hernandez (u.a. bekannt aus dem Rybkaforum), welcher die Auswahl für das Superfinale traf, die Stellungen nach dem jeweils 6. Zug mit Engines geprüft hat um zu unausgewogene zu vermeiden.

Als Extrembeispiel dafür, warum unausgewogene Vorgaben schlecht sind: Wenn ich nur Varianten wähle an deren Schluß eine Farbe verläßlich auf Gewinn steht, kann ich damit unentschiedene Matches zwischen jedweden halbwegs guten Engines herbeiführen, selbst bei hunderten von Elos Stärkedifferenz: Weil jede Variante beim zweimaligen Ausspielen 1:1 ausginge. Und diesen Nivellierungseffekt will man natürlich überhaupt nicht haben, auch nicht in weniger extremen Ausformungen.

Gentlemen, I am not fluent in German so I used Google Translate to get the essence of your questions/comments.  In response let me first direct you to my comments from November 18 that were posted to the nTCEC Facebook page:  https://www.facebook.com/tcec.chess

I should start by explaining that my being appointed as the source of the Superfinal openings was as much a surprise to me as it was to you.  Apparently some of my comments in Stage 4 concerning how one might go about reducing draw-rates combined with my statement that you needed lots of data to do it properly gained traction (aided by my credible demonstrations that I had such data).  Martin sent me an email about four days before the Superfinal was to begin and solicited 24 opening lines with the intention of bringing the draw-rate down.  In 24 hours I produced a spreadsheet with my proposals, backed up with statistics from my human and engine databases (as explained in the link above).

I had nothing to do with the order these 24 openings were played.  As it happened, some of the most unbalanced openings of the 24 were front-loaded, resulting in an extraordinary run of decisive games in the early going.  Some of these were really wild games (game #1's incredible start had the chat room in a state of great excitement). 

My approach to selecting these openings was, of necessity, quite simple.  My mandate from Martin was to produce 24 six-move openings.  My criteria were 1) opening lines representing a well-diversified set of opening systems, including a couple of wildcards, 2) lines ending in a position that was well below-average in terms of draw-rate in both my human and engine-based databases, and 3) play-balance, so that openings would not be so one-sided as to be easy wins for the same color in two straight games.  My approach could have been much more sophisticated than this, but given the time constraint I was under a search for truly optimal openings could not have been accomplished in the time allotted.

So far the results have been good, I think.  Not 100% successful to be sure: in practice it has turned out that some of my choices were too one-sided and whimsical, but to be fair keep in mind I had no control past the sixth move and the engines sometimes made move-choices immediately afterwards that left all known theory.  I think that by the end of the Superfinal we will all be in a better position to look at the totality of the series and judge the overall success of the effort.  Keep in mind my stated goal was a draw-rate in the 60-65% range.  It seems likely to me draws could have been as high as 75-80% if conventional GM lines had been played and such a performance, while perfectly consistent with chess entities of this playing strength, would not have been consistent with the nTCEC's main objective of "entertainment".

If Martin asks me to participate in the future there can be no doubt the approach will be less hasty and more methodical.  I am already working on developing a comprehensive solution to the matter of openings that would pair up a .pgn with tens of thousands of unique GM-produced opening lines to metrics from my formidable database, nine years in the making.  From that database the best 5% would keep nTCEC stocked in openings for many years.  Anyway, that's what I hope to do.