Skip to content

Configurations

One of the most powerful features of PuzzleClone is its separation of a puzzle's logic from its instance.

  • The Specification File (yaml) serves as a plan for batch generation of puzzles, defining the puzzle's structure, rules, and how to randomize it.
  • The Configuration (JSON) is instance-specific, including every random choice made for one specific puzzle instance.

When you have both the specification file and the configuration, you can perfectly recreate a unique puzzle. This separation offers two key advantages:

  1. Verifiability: How do you know your complex spec.yaml is correct? You can generate a puzzle from your seed puzzle's known parameters, creating a configuration file. Then, use that configuration to generate a new puzzle. If the new puzzle and its answer match your original seed puzzle, your specification file is likely to be correct.

  2. Flexibility: Imagine you want to translate a puzzle into another language or change its theme (e.g., from students to wizards) without altering the core logic. Configurations make this trivial. You can reuse the exact same configuration with a new specification file that only changes the descriptive text, guaranteeing 100% logical consistency.

Anatomy of a Configuration

Every puzzle generated by PuzzleClone includes a config field that stores its unique configuration. This configuration captures two types of information:

  1. Variable Values: All the final values of the variables defined in your vars block are stored as simple key-value pairs. json "p_num": 10, "select_num": 6, "names": ["Laura", "Corey", "Austin", ...]

  2. The domains and the randomization pools for generating derived symbols, dynamic conditions, and options of multiple-choice questions. They will be stored as a dict in the format of {"domain": <domain>, "pool": <pool>}

    • For Single-Template Items: The pool is a list where each entry corresponds to a generated item (e.g., one condition). Each item is a 3D list following the [dimension][source][selection] hierarchy.
    • For Multi-Template Items: The pool is a list of objects, each pinpointing the exact template used and the selections made for it: {"template_id": <index>, "config": [[...]]}.
    • Index vs. Value (__ Notation): This is a crucial detail. In the pool, a value wrapped in double underscores is an index, not a literal value.
      • "__0__" means: "the 0-th item from the source list" (e.g., the first student's name).
      • "0" means: "the literal integer value 0".

Configuration-Based Puzzle Generation

You can instruct PuzzleClone to skip all randomization and deterministically build a puzzle using a specific configuration file. This is perfect for verifying, translating, or re-theming puzzles.

The command is:

python translator.py -d <spec_path> -g <config_path> -o <output_path>

This will use the specification file from <spec_path> but follow the exact configuration from <config_path>, producing a perfect clone and saving it to <output_path>.

Mixing Determinism with Randomness

Sometimes, you want a mix: use most of an existing configuration but re-randomize a few specific variables.

Use Case: You have 1000 puzzles about students generated in Chinese. You want to create the exact same 1000 puzzles, with the same underlying logic, but in English and with new, randomized English names.

  1. Create a new spec_en.yaml where you translate all the desc fields to English. In the vars section, change the names variable to use an English name generator.
  2. Use the following command, which adds the --use-spec-vars flag:
python translator.py -d spec_en.yaml -g config_zh.json -o puzzle_en.json --use-spec-vars names

This command tells PuzzleClone:

"Use spec_en.yaml as the specification file and config_zh.json as the configuration. However, for the names variable specifically, ignore the value in the config file and generate a new one based on the rules in spec_en.yaml. For everything else, follow the config."

This powerful feature allows you to precisely control which parts of a puzzle are cloned and which are freshly generated.