diff --git a/reasoning_gym/algorithmic/base_conversion.py b/reasoning_gym/algorithmic/base_conversion.py
index f3b9d6a1..15607ed4 100644
--- a/reasoning_gym/algorithmic/base_conversion.py
+++ b/reasoning_gym/algorithmic/base_conversion.py
@@ -125,8 +125,8 @@ class BaseConversionCurriculum(BaseCurriculum):
         self._define_attributes(
             RangeAttributeDefinition(
                 name="base",
-                levels=[9, 18, 27, 36],
-                default_level=0,
+                levels=[2, 9, 18, 27, 36],
+                default_level=1,
                 description="The base of the number system",
                 attr_type=AttributeType.APPEND,
                 min_value=2,
diff --git a/reasoning_gym/coaching/base_curriculum.py b/reasoning_gym/coaching/base_curriculum.py
index 22c031a1..5b7463c4 100644
--- a/reasoning_gym/coaching/base_curriculum.py
+++ b/reasoning_gym/coaching/base_curriculum.py
@@ -106,3 +106,73 @@ class BaseCurriculum:
             self.set_attr_level(attr_name, current_level - 1)
             return True
         return False
+
+    def get_max_level(self) -> int:
+        """
+        Get the maximum level currently set across all attributes.
+
+        Returns:
+            int: The maximum level currently set across all attributes
+        """
+        if not self._attributes:
+            return 0
+
+        return max(self.get_attr_level(attr_name) for attr_name in self._attributes)
+
+    def set_global_level(self, level: int) -> None:
+        """
+        Set all attributes to the specified level.
+        If the level exceeds the number of defined levels for an attribute,
+        use the highest defined level for that attribute.
+
+        Args:
+            level: The level to set for all attributes
+        """
+        for attr_name, attr in self._attributes.items():
+            # Use the highest defined level if the requested level exceeds available levels
+            attr_level = min(level, len(attr.levels) - 1)
+            self.set_attr_level(attr_name, attr_level)
+
+    def increment_global_level(self) -> bool:
+        """
+        Increment the level of all attributes by one from the current maximum level.
+
+        Returns:
+            bool: True if at least one attribute's level was incremented, False otherwise
+        """
+        current_max = self.get_max_level()
+        target_level = current_max + 1
+
+        # Check if any attribute can be incremented
+        can_increment = any(
+            self.get_attr_level(attr_name) < len(self._attributes[attr_name].levels) - 1
+            for attr_name in self._attributes
+        )
+
+        if can_increment:
+            for attr_name, attr in self._attributes.items():
+                # Only increment if the attribute is not already at its maximum level
+                if self.get_attr_level(attr_name) < len(attr.levels) - 1:
+                    # Don't exceed the attribute's maximum level
+                    new_level = min(target_level, len(attr.levels) - 1)
+                    self.set_attr_level(attr_name, new_level)
+            return True
+        return False
+
+    def decrement_global_level(self) -> bool:
+        """
+        Decrement the level of all attributes by one from the current maximum level.
+
+        Returns:
+            bool: True if at least one attribute's level was decremented, False otherwise
+        """
+        current_max = self.get_max_level()
+
+        if current_max > 0:
+            target_level = current_max - 1
+            for attr_name in self._attributes:
+                # Only decrement if the attribute is at the current maximum level
+                if self.get_attr_level(attr_name) == current_max:
+                    self.set_attr_level(attr_name, target_level)
+            return True
+        return False
diff --git a/scripts/debug_curricula.py b/scripts/debug_curricula.py
new file mode 100755
index 00000000..5e2f76cc
--- /dev/null
+++ b/scripts/debug_curricula.py
@@ -0,0 +1,182 @@
+#!/usr/bin/env -S PYTHONHASHSEED=1 python3
+"""Generate a markdown document showing curriculum progression for all datasets"""
+
+import argparse
+import textwrap
+from copy import deepcopy
+from pathlib import Path
+from typing import Optional
+
+from tqdm import tqdm
+
+from reasoning_gym.factory import CURRICULA, DATASETS, create_curriculum, create_dataset
+
+
+def generate_curricula_doc(
+    num_examples: int = 1, show_config: bool = False, dataset_names: Optional[list[str]] = None
+) -> str:
+    """Generate markdown content showing curriculum progression
+
+    Args:
+        num_examples: Number of examples to generate per difficulty level
+        show_config: Whether to show the effective dataset configuration
+        dataset_names: Optional list of specific dataset names to generate documentation for
+    """
+
+    # Start with header
+    content = ["# Reasoning Gym Curriculum Progression\n"]
+    content.append("This document shows how tasks change as curriculum difficulty increases for each dataset.\n\n")
+
+    # Get datasets with curricula
+    all_datasets_with_curricula = sorted([name for name in DATASETS.keys() if name in CURRICULA])
+
+    # Filter to specific datasets if provided
+    if dataset_names:
+        # Validate all requested datasets
+        for name in dataset_names:
+            if name not in CURRICULA:
+                raise ValueError(f"Dataset '{name}' does not have a curriculum")
+        datasets_with_curricula = dataset_names
+    else:
+        datasets_with_curricula = all_datasets_with_curricula
+
+    # Add index
+    content.append("## Available Curricula\n")
+    for name in datasets_with_curricula:
+        # Create anchor link
+        anchor = name.replace(" ", "-").lower()
+        content.append(f"- [{name}](#{anchor})\n")
+    content.append("\n")
+
+    # Add examples for each dataset with curriculum
+    content.append("## Curriculum Progression Examples\n")
+
+    # Create progress bar for datasets
+    for name in tqdm(datasets_with_curricula, desc="Processing datasets"):
+        # Add dataset header with anchor
+        content.append(f"### {name}\n")
+
+        # Get curriculum and dataset class
+        curriculum = create_curriculum(name)
+        dataset_cls, config_cls = DATASETS[name]
+
+        # Get dataset class docstring if available
+        try:
+            dataset = create_dataset(name, seed=42)
+            if dataset.__class__.__doc__:
+                doc = textwrap.dedent(dataset.__class__.__doc__.strip())
+                content.append(f"{doc}\n\n")
+        except Exception as e:
+            content.append(f"*Error loading dataset: {str(e)}*\n\n")
+            continue
+
+        # Show curriculum attributes
+        content.append("#### Curriculum Attributes\n")
+        for attr_name, attr in curriculum.attributes.items():
+            content.append(f"- **{attr_name}**: {attr.description}\n")
+            content.append(f"  - Levels: {attr.levels}\n")
+        content.append("\n")
+
+        # Show progression with all attributes increasing simultaneously
+        content.append(f"#### Overall Difficulty Progression\n")
+
+        # Find the maximum number of levels across all attributes
+        max_levels = max(len(attr.levels) for attr in curriculum.attributes.values())
+
+        # Show examples at each difficulty level
+        for level in tqdm(range(max_levels), desc=f"Dataset: {name}, Overall Difficulty", leave=False):
+            try:
+                # Reset curriculum to defaults
+                curriculum = create_curriculum(name)
+
+                # Set all attributes to this level using the global level function
+                curriculum.set_global_level(level)
+
+                # Generate config with this level
+                config = curriculum.generate_configuration({"seed": 42 + level})
+
+                # Create dataset with this config
+                dataset = dataset_cls(config=config)
+
+                # Show level and example
+                content.append(f"##### Difficulty Level {level}\n")
+
+                # Show the current values for each attribute
+                content.append("Attribute values:\n")
+                for attr_name, attr in curriculum.attributes.items():
+                    attr_level = min(level, len(attr.levels) - 1)
+                    content.append(f"- {attr_name}: {attr.levels[attr_level]}\n")
+
+                # Show the effective configuration if requested
+                if show_config:
+                    content.append("\nEffective configuration:\n")
+                    for key, value in vars(config).items():
+                        if key != "seed" and key != "size":
+                            content.append(f"- {key}: {value}\n")
+
+                # Generate multiple examples
+                for ex_idx in range(num_examples):
+                    # Get example
+                    example = dataset[ex_idx]
+
+                    content.append(f"\n```\n")
+                    if num_examples > 1:
+                        content.append(f"Example {ex_idx + 1}:\n")
+                    content.append(f"Question: {example['question']}\n")
+                    content.append(f"Answer: {example['answer']}\n")
+                    if example.get("metadata"):
+                        content.append(f"Metadata: {example['metadata']}\n")
+                    content.append("```\n")
+            except Exception as e:
+                content.append(f"##### Difficulty Level {level}\n")
+                content.append(f"*Error generating example: {str(e)}*\n\n")
+
+        content.append("\n")
+
+    return "".join(content)
+
+
+def main():
+    """Generate curricula markdown file"""
+    # Parse command line arguments
+    parser = argparse.ArgumentParser(description="Generate curriculum documentation")
+    parser.add_argument("--examples", type=int, default=3, help="Number of examples to generate per difficulty level")
+    parser.add_argument("--show-config", action="store_true", help="Show the effective dataset configuration")
+    parser.add_argument(
+        "--output", type=str, default="CURRICULA.md", help="Output file path (relative to project root)"
+    )
+    parser.add_argument(
+        "--dataset", type=str, help="Generate documentation for specific datasets (comma-separated list)"
+    )
+    args = parser.parse_args()
+
+    # Ensure scripts directory exists
+    script_dir = Path(__file__).parent
+    if not script_dir.exists():
+        script_dir.mkdir(parents=True)
+
+    print(f"Generating curricula documentation...")
+    print(f"Number of examples per level: {args.examples}")
+    print(f"Show configuration: {args.show_config}")
+
+    # Parse dataset names if provided
+    dataset_names = None
+    if args.dataset:
+        dataset_names = [name.strip() for name in args.dataset.split(",")]
+        print(f"Generating documentation for datasets: {', '.join(dataset_names)}")
+
+    curricula_path = script_dir.parent / args.output
+
+    curricula_content = generate_curricula_doc(
+        num_examples=args.examples, show_config=args.show_config, dataset_names=dataset_names
+    )
+
+    with open(curricula_path, "w") as f:
+        f.write(curricula_content)
+        f.write("\n")
+
+    print(f"Generated curricula documentation at {curricula_path}")
+
+
+if __name__ == "__main__":
+    main()
diff --git a/tests/test_base_conversion.py b/tests/test_base_conversion.py
index 1e0ebe55..054257d0 100644
--- a/tests/test_base_conversion.py
+++ b/tests/test_base_conversion.py
@@ -177,18 +177,18 @@ def test_base_conversion__curriculum():
     base_cfg: BaseConversionConfig = curriculum.generate_configuration(base_value)
     assert base_cfg.seed == 1
     assert base_cfg.size == 150
-    assert base_cfg.min_base == 9 and base_cfg.max_base == 9
+    assert base_cfg.min_base == 2 and base_cfg.max_base == 9
     assert base_cfg.min_value == 1000 and base_cfg.max_value == 1000
 
     # test incrementing attribute levels
     curriculum.increment_attr_level("base")
     curriculum.increment_attr_level("value")
     increased_cfg = curriculum.generate_configuration(base_value)
-    assert increased_cfg.min_base == 9 and increased_cfg.max_base == 18
+    assert increased_cfg.min_base == 2 and increased_cfg.max_base == 18
     assert increased_cfg.min_value == 1000 and increased_cfg.max_value == 10000
 
     # test decrementing attribute level for base again
     curriculum.decrement_attr_level("base")
     partially_decreased_cfg = curriculum.generate_configuration(base_value)
-    assert partially_decreased_cfg.min_base == 9 and partially_decreased_cfg.max_base == 9
+    assert partially_decreased_cfg.min_base == 2 and partially_decreased_cfg.max_base == 9
     assert partially_decreased_cfg.min_value == 1000 and partially_decreased_cfg.max_value == 10000