{"data":{"kind":"file","path":"README.md","version_id":"zm1dp3gju12pfik07aq70jh3","entry":{"name":"README.md","path":"README.md","is_directory":false,"size":3941,"modified_at":"2026-08-30T13:40:26.607000","content_hash":"d72b6c72220aa87957b1811356bf7a44d65d87a55dcf8914722ef37690a548a2"},"entries":[],"content":"# Genetic Algorithm Evolution Simulator\n\n**Version:** 1.1.0\n\nA verifiers environment that evaluates LLM reasoning about genetic algorithms (GAs) and evolutionary computation. Tests understanding of core GA concepts from basic operator design to advanced topics like coevolution and deceptive problems.\n\n## What It Tests\n\nThis environment assesses an LLM's ability to:\n\n- **Design** genetic algorithms for specific optimization problems (TSP, OneMax, process optimization)\n- **Analyze** fitness landscapes, convergence behavior, and schema dynamics\n- **Compare** selection strategies, crossover operators, and diversity maintenance mechanisms\n- **Diagnose** premature convergence, loss of diversity, and deceptive fitness landscapes\n- **Reason** about multi-objective optimization, constraint handling, and coevolutionary systems\n- **Critically evaluate** research claims about GA performance\n\n## Task Categories\n\n| Category | Difficulty | Description |\n|----------|-----------|-------------|\n| Basic Design | 1 | Configure GAs for simple problems |\n| Landscape Analysis | 2 | Analyze local optima and fitness landscapes |\n| Fitness Design | 2 | Design fitness functions for real problems |\n| Operator Analysis | 2 | Compare crossover and mutation operators |\n| Convergence Diagnosis | 3 | Diagnose and fix premature convergence |\n| Multi-Objective | 3 | Pareto optimization and NSGA-II |\n| Prediction | 3 | Predict GA behavior on given problems |\n| Diagnosis | 3 | Analyze run statistics and recommend fixes |\n| Implementation | 3 | Design steady-state vs generational implementations |\n| Advanced Techniques | 4 | Deceptive problems, linkage learning |\n| Theoretical Analysis | 4 | Schema theorem and building block hypothesis |\n| Constraint Handling | 4 | Penalty, repair, and decoder approaches |\n| Real-World | 4 | Manufacturing and engineering applications |\n| Critical Analysis | 4 | Evaluate research claims and experimental methodology |\n| Coevolution | 5 | Multi-population coevolutionary systems |\n\n## Scoring Rubric\n\nFive weighted scoring functions evaluate responses:\n\n| Metric | Weight | What It Measures |\n|--------|--------|-----------------|\n| Completeness | 0.25 | Coverage of key concepts from the reference answer |\n| Technical Accuracy | 0.25 | Correct use of GA terminology and mathematical concepts |\n| Reasoning Depth | 0.20 | Causal reasoning, comparisons, and structured analysis |\n| Practical Applicability | 0.15 | Specific parameters, implementation details, trade-offs |\n| Critical Thinking | 0.15 | Limitations, evaluations, nuanced recommendations |\n\n## Usage\n\n```python\nimport verifiers as vf\n\n# Load with defaults (all training tasks)\nenv = vf.load_environment(\"genetic-algorithm-evolution-simulator\")\n\n# Load with difficulty filter\nenv = vf.load_environment(\"genetic-algorithm-evolution-simulator\", difficulty=3)\n\n# Load evaluation split\nenv = vf.load_environment(\"genetic-algorithm-evolution-simulator\", split=\"eval\")\n\n# Sample subset\nenv = vf.load_environment(\"genetic-algorithm-evolution-simulator\", num_samples=5)\n```\n\n## Parameters\n\n| Parameter | Default | Description |\n|-----------|---------|-------------|\n| `difficulty` | None (all) | Filter tasks by difficulty level (1-5) |\n| `num_samples` | None (all) | Number of tasks to sample |\n| `split` | \"train\" | Dataset split: \"train\" or \"eval\" |\n\n## Dataset\n\n- **Training:** 10 tasks spanning difficulty 1-5\n- **Evaluation:** 5 tasks spanning difficulty 3-4\n- **Total:** 15 unique tasks across 10 categories\n\n## Design Principles\n\n1. **Anti-gaming**: Multi-signal scoring with sigmoid smoothing prevents keyword stuffing from scoring high\n2. **Depth over breadth**: Rewards genuine understanding of GA theory, not just terminology\n3. **Practical grounding**: Expects specific parameters and implementation details, not vague descriptions\n4. **Critical evaluation**: Tests ability to assess claims and identify weaknesses in approaches\n\n## License\n\nMIT\n","encoding":"utf-8","truncated":false,"total_bytes":3941},"status":null}