Skip to content

Verbose Trace (Human analysis)

amchess edited this page Mar 5, 2026 · 1 revision

Verbose Trace: The Virtual Coach

In standard chess engines like Stockfish, the eval command is primarily a debugging tool for developers, outputting raw numerical values. It offers numbers, but not understanding.

ShashChess 41 transforms the eval command into a Virtual Coach. By enabling the UCI_ShowWDL option, the engine activates a "Verbose Trace" mode that deconstructs the neural network's assessment into human-understandable concepts: Shashin Zones, Piece Activity rankings, and Candidate Move ordering.

Note: This tool analyzes the Static State of the board. It asks the neural network: "Without calculating ahead, how promising does this position look right now?"


1. Key Components

🧠 The Shashin Zone Classification

The engine converts the raw internal value (centipawns) into a Win Probability, which is then mapped to the Alexander Shashin Framework. This gives the user an immediate "weather report" of the strategic climate.

  • High Tal (>95% Win): Total Attack mode.
  • Capablanca (50% Win): Perfect equilibrium. Requires strategic maneuvering.
  • Petrosian (<20% Win): Defensive emergency.
  • Chaos Zones: Transition phases where initiative is shifting.

📉 The Makogonov Principle (Piece Activity)

One of the most famous principles of the Soviet Chess School is: "Identify your worst-placed piece and improve it." ShashChess automates this analysis.

The engine iterates through every piece and assigns an "Activity Score" based on:

  1. Mobility: Number of legal moves.
  2. Centralization: Bonus for controlling the center (d4/d5/e4/e5).
  3. King Tropism: Bonus for attacking the enemy King's safety zone.
  4. Advancement: Bonus for pieces deep in enemy territory.

The Output: The engine outputs a sorted list of pieces from Worst to Best, explicitly flagging the worst unit to direct your attention to the problem area.

📋 Legal Moves Analysis (Candidate Ordering)

Standard engines only show the best move after a search. ShashChess performs a "what-if" scenario for every legal move available in the position.

It plays each move on an internal board copy and runs a static NNUE evaluation. The results are sorted by Win Probability, showing you which moves look "natural" to the neural network before any calculation begins.


2. Sample Output

When you type eval in the console (with UCI_ShowWDL enabled), you receive a detailed report like this:

=== SHASHIN STATIC ANALYSIS ===
*** Static activity evaluation based on NNUE network ***

--- PIECE ACTIVITY ANALYSIS (Makogonov Principle) ---
White pieces: Na1(4) <-- Worst unit, Bb2(12), Rf1(14), ...
Black pieces: Nh8(2) <-- Worst unit, ...

--- LEGAL MOVES ANALYSIS ---
Legal moves ordered by static activity:
e4(+0.35 cp, 52%), d4(+0.31 cp, 51%), ...

================================
Final Static Activity Evaluation: 0.35 cp | 52% win probability (White)
WDL: 18% Win / 68% Draw / 14% Loss
Shashin Zone: Chaos: Capablanca-Tal (Initiative)

3. Technical Implementation

  • Perspective Consistency: ShashChess standardizes all Verbose Trace output to White's Perspective. This ensures that specific Shashin Zones are always correctly identified regardless of whose turn it is.

  • Memory Safety: The trace function uses heap allocation (std::make_unique) to prevent stack overflow crashes during deep analysis on systems with limited stack sizes.

Clone this wiki locally