Now it is the turn for the Pioneer 2. The top level is 'Karpov" and I thought I would see how the Pioneer 2 would fare against my Palm Hiarcs 12.



The Pioneer 2 did very well here to get a draw. Now for game 2.

Another exciting game with the Palm managing a victory this time. But it was not easy!

Keep up the good work Arne!

Yes Alex S, that is definitely true , it is just that the style and personality layer add human based play and plan coherence ,using a dynamic lookup of the plans for a position that come from the library of backward extracted plans of the Personality that is being emulated. This booosts moves that fit deeper plans the actual engine cannot see yet because they are part of causal chains that sometimes reach very deep. If a move that the engine provides as a safe move fits deeper plans , it will be boosted. And this apparently adds consistency and strength for less compute , but at which level and how much exactly is still to be determined. For me the magic happens at "low" depth settings like for 1600 or 1800 elo, because the system will still play using the extracted knowledge of the plan library, the style rules , etc, but the palette of 10 moves offered in Multipv are based on shallower tactical calculations at depth 6 or so. So the system will never insert blunders or unnatural moves, but will have tactical oversights somehow , that you really need to search to beat it as a human. But is it s not random or obvious as with UCI limitstrength. But as said the testing will reveal further data on this system. We will need to find out out how much strength the personality layer adds to raw calculation by various testing vs fixed depth settings, not vs UCI limit strength. So that is the calibration challenge. It did not use to be like this: At the beginning when the personality system was not so refined , yet, the style layer actually weakened the pure engine performance so we had to compensate by adding more depth an d tus more calculation. But as several new elements were introduced to the system like backward plan extraction and classifaction per theme , causal chain detection , plan memory boosting, all of a sudden the system became very strong and even unplayable with the added depth compensation (this was using stockfish as backend move provider) . So I had to take back the depth compensation to find out how much stronger the sytem had become with all these enhancements. And that is what I am trying to find out now with the help of the community here . The game that made it clear for me that something was happening was this one with the Smyslov personality at 2300 elo vs the Boss Engine Bot level 25 at Chess.com (The "unbeatable" engine)

[Event "ChessMind Game"]
[Site "chess.com"]
[Date "2026.01.25"]
[Round "?"]
[White "Chess.com Komodo bot level 25 3200 elo ]
[Black "Chessmind Smyslov 2300 elo"]
[Result "0-1"]

1. e4 e5 2. Nf3 Nc6 3. Bb5 a6 4. Ba4 Nf6 5. O-O Be7 6. Re1 b5 7. Bb3 O-O 8. a4 b4 9. d3 d6 10. Nbd2 h6 11. a5 Re8 12. Nc4 Be6 13. Ba4 Bd7 14. h3 Bf8 15. Bd2 Rb8 16. Ne3 Ne7 17. Bb3 Ng6 18. Qb1 Nf4 19. Qa2 Re7 20. Nf5 Bxf5 21. exf5 c5 22. Be3 Rc7 23. Bd2 Qc8 24. Nh4 N6h5 25. Rad1 d5 26. Rxe5 Be7 27. Nf3 Bd6 28. Rxd5 Nxd5 29. Bxd5 Qxf5 30. Re1 Bf4 31. Bc4 Bxd2 32. Nxd2 Nf4 33. Ne4 Re7 34. Kh2 Qc8 35. Nf6+ Kf8 36. Nh7+ Ke8 37. Bxf7+ Kd8 38. Rxe7 Kxe7 39. Qc4 g5 40. Bd5 Nxd5 41. Qxd5 Qe6 42. Qxc5+ Qd6+ 43. Qxd6+ Kxd6 44. Nf6 Ke6 45. Ng4 b3 46. cxb3 h5 47. Ne3 Rxb3 48. d4 Rxb2 49. Kg3 h4+ 50. Kg4 Rb5 51. g3 Rxa5 52. gxh4 gxh4 53. f4 Rb5 54. Kxh4 a5 55. Kg4 Rb3 56. f5+ Kf6 57. Kf4 a4 58. Ng4+ Kg7 59. d5 a3 60. f6+ Kf8 61. d6 a2 62. d7 Rb8 63. Ne5 a1=Q 64. Ke4 Qa4+ 65. Kf5 Qc2+ 66. Kf4 Qd2+ 67. Kg4 Qd4+ 68. Kf5 Rb5 69. d8=Q+ Qxd8 70. Ke4 Qe8 71. Kd3 Qxe5 72. h4 Rb3+ 73. Kc2 Qe3 74. h5 Qd3+ 75. Kc1 Rb1# 0-1

This felt like the scene in the movie 300 where Leonidas says lets see how immortal those immortals really are. So yes , We are still finding out things.

Some serious testing going on there , great work Clive!! Yes the Ultimate Engine is a relative lightweight (178 kb) and it does not use NNUE or even Tablebases so I am aware that the endgame handling would need to be solved, but tablebases are quite big, and since we have also databases for the middlegame, I have not yet found a way to implement thoseendgame tablebases , of course i need to do this.. that is the next thing to implement, I assumed the baseline of the ultimate C engine to be somewhere around 2400-2500 elo. It is basically like going through a crash course of all chess programming challenges from the last 40 years

I remember a story about the Rebel computer missing the world Championship in the eighties because it had a bug in the endgame or a missing piece of knowledge about a certain endgame promotion situation, almost giving Ossi Weiner A heart attack , but i am not sure i got all the details right.

Hallo Udo, Danke es gab ein par Anpassungen mit den Server heute, leider, ich musste ein paar Bugs beseitigen. Da es bei mir gehostet wird auf mein Dell Workstation, is Server Administrator noch etwas neues was ich lernen musste XD Sorry! Aber dennoch habe ich eine Frage wie hast du es geschafft den Engine Output auf Deutsch zu bekommen?? *Respekt*

Hallo Arne, Ich habe da gar nichts gemacht, gut, das Programm hatte ich auf Deutsch gestellt. Aber da das ganze ja bei mir im Microsoft Edge Browser läuft, und der so eingestellt ist, das er nach Möglichkeit alles Englische ins Deutsche übersetzt, könnte es vielleicht daran liegen?