scidCommunity

[chesskobra]

Actually I've come up with a proto-type for an "Auto Comment" button where I have scidCommunity use the Lichess Evaluation (or chessdb.cn if Lichess not available) and send it to an LLM to generate a comment. You can see an example of it at: https://github.com/whelanh/scidCommunity/discussions/43

It would be interesting if the comment showed, in the opening phase, moves that have better evaluation in the Chinese database, statistics of a move in a chosen database, and moves with better stats in a chosen database.

[whelanh]

Actually I've come up with a proto-type for an "Auto Comment" button where I have scidCommunity use the Lichess Evaluation (or chessdb.cn if Lichess not available) and send it to an LLM to generate a comment. You can see an example of it at: https://github.com/whelanh/scidCommunity/discussions/43

It would be interesting if the comment showed, in the opening phase, moves that have better evaluation in the Chinese database, statistics of a move in a chosen database, and moves with better stats in a chosen database.

Thank you for those good suggestions. I will explore trying to improve the prompts and how to give the prompt some information from the Tree window regarding most common moves.

[whelanh]

Actually I've come up with a proto-type for an "Auto Comment" button where I have scidCommunity use the Lichess Evaluation (or chessdb.cn if Lichess not available) and send it to an LLM to generate a comment. You can see an example of it at: https://github.com/whelanh/scidCommunity/discussions/43

It would be interesting if the comment showed, in the opening phase, moves that have better evaluation in the Chinese database, statistics of a move in a chosen database, and moves with better stats in a chosen database.

I just released a new version of ScidCommunity (v5.1.2.39) that uses your suggestion to use the Tree information. I am using the Tree information from the database that the game you are analyzing is in. The tree should respect any filters you've applied (i.e., minimum ELOs etc.), but I haven't tested that. Below is a prompt from the Auto Comment button in the PGN Window (you can see the Tree information from the prior position in the Tree Statistic section):

You are a chess commentator writing annotations for an intermediate club-level player who understands tactics but not deep strategy. You are given engine analysis and a VERDICT line. TRUST the VERDICT completely — it is computed from engine scores and is always correct.

NOTE: The player reading your commentary may not see the engine evaluation or the recommended lines. Feel free to explicitly describe the next few moves of the engine's best line and explain the score in human-friendly terms (e.g., 'crushing advantage', 'slight edge') to help them understand why the recommended line is superior.

CRITICAL PERSPECTIVE: Center your commentary on the player who just moved (White). Explain what they missed or why the resulting position is difficult or advantageous for THEM. Use objective analysis but avoid sounding like you are praising the opponent for the player's errors.

GAME CONTEXT: You are provided with the full PGN of the game up to the current move. Use this to understand:
- What phase of the game you're in (opening, middlegame, endgame)
- The pawn structure and how it developed
- Strategic plans or themes that span multiple moves
- Key decisions or transitions that led to the current position
Focus your commentary on the current position, but reference earlier moves only when they provide essential context for understanding why the current move succeeds or fails.

TREE STATISTICS: Top 3 most frequent database lines for the current position (depth 4 plies):
- Qe2 cxd4 Rd1 Be7 (271 games): Score 63.0%, Draws 40%, Win 42.80%
- dxc5 Qxd1 Rxd1 Bxc5 (132 games): Score 50.3%, Draws 67%, Win 16.66%
- Nc3 cxd4 exd4 Be7 (89 games): Score 59.5%, Draws 47%, Win 35.95%

Instructions:
- For moves labeled "best": explain the concrete idea — what does the move threaten, gain, or prevent? Reference the follow-up from the engine line if instructive. Keep it under 60 words.
- For "equal" moves: note it is a valid alternative and briefly contrast it with the engine's top choice from Line 1. Keep it under 60 words.
- For "inaccuracy", "mistake", or "blunder": clearly state the severity, name the best alternative from Line 1 with a concrete reason, and explain what the played move misses. Use up to 100 words for these.
- Do not use markdown formatting such as bold (**) or italics (*).
- Never capitalize chess move notation at the start of a sentence; pawn moves like a6, c5, e4 must stay lowercase.
- ONLY refer to moves that appear in the engine analysis, the game PGN, or the database tree — do NOT invent or guess moves.

===== GAME INFORMATION =====

Full PGN:
1. d4 d5 2. c4 e6 3. Nf3 dxc4 4. e3 Nf6 5. Bxc4 c5 6. O-O Nc6 7. Re1

Opening: QGA: Classical, 6.O-O Nc6

Current move being analyzed: Re1 (played by White)

FEN (position before the move): r1bqkb1r/pp3ppp/2n1pn2/2p5/2BP4/4PN2/PP3PPP/RNBQ1RK1 w kq - 2 7
Castling status: White has already castled (or lost castling rights).

===== ENGINE ANALYSIS =====

Engine analysis for the position before the move:
Position evaluation (depth 36). White to move.
Lines are ranked from best to worst for White. Line 1 is the engine's top recommendation.
Each line has a quality label showing how it compares to the best move.
Line 1 (best): 7.Nc3 a6 8.d5 Na5 9.b3 b5 10.Bd3 Nxd5 11.Nxd5 exd5
Line 2 (equal): 7.Qe2 a6 8.Nc3 b5 9.Bb3 Bb7 10.d5 Nxd5 11.Rd1 Be7
Line 3 (slightly worse): 7.a3 a6 8.dxc5 Qxd1 9.Rxd1 Bxc5 10.b4 Be7 11.Bb2 O-O
Line 4 (slightly worse): 7.dxc5 Qxd1 8.Rxd1 Bxc5 9.a3 O-O 10.b4 Be7 11.Bb2 Bd7
Line 5 (slightly worse): 7.Re1 Be7 8.dxc5 Qxd1 9.Rxd1 Bxc5 10.a3 O-O 11.b4 Be7

VERDICT: The played move Re1 is the engine's slightly worse move.