Upload folder using huggingface_hub

README.md

@@ -0,0 +1,223 @@
+---
+license: mit
+task_categories:
+  - question-answering
+  - text-generation
+language:
+  - en
+tags:
+  - benchmark
+  - reasoning
+  - multi-step
+  - evaluation
+  - llm-evaluation
+  - goodhart
+  - execution-vs-understanding
+size_categories:
+  - n<1K
+---
+
+# Goodhart Gap Benchmark
+
+**Detecting the gap between understanding and execution in language models**
+
+## Overview
+
+The Goodhart Gap Benchmark tests whether language models can correctly *execute* multi-step reasoning tasks that they can correctly *explain*. Named after Goodhart's Law ("When a measure becomes a target, it ceases to be a good measure"), this benchmark reveals a critical failure mode: models that understand procedures but fail to execute them.
+
+## Key Finding
+
+In our testing of 15+ models:
+- **gpt-4o**: 57% pass rate (fails on financial, scheduling, units)
+- **gpt-4o-mini**: 36% pass rate
+- **Claude 3.5 Haiku**: 93% pass rate
+- **Llama 3.1 70B**: Fails the canonical discount calculation despite correct explanation
+
+## The Canonical Example
+
+**Problem**: "If a shirt costs $25 and is on 20% sale, and you have a $5 coupon, what do you pay?"
+
+**Correct answer**: $15 (apply 20% discount first: $25 × 0.8 = $20, then subtract coupon: $20 - $5 = $15)
+
+When we first ask models to *explain* the procedure, they all correctly state: "First apply the discount, then subtract the coupon."
+
+When we then ask for the answer, many models fail—giving answers like $16, $17, $22.50, or even $175.
+
+## Dataset Statistics
+
+| Metric | Value |
+|--------|-------|
+| Total problems | 67 |
+| Domains | 7 |
+| Difficulty levels | 3 (easy, medium, hard) |
+| Steps per problem | 2-4 |
+
+### Problems by Domain
+
+| Domain | Count | Description |
+|--------|-------|-------------|
+| math_discount | 15 | Discounts, coupons, taxes, markups |
+| time | 13 | Duration arithmetic, travel times |
+| financial | 10 | Interest, taxes, commissions |
+| logic | 8 | Ordering, deduction, set operations |
+| recipe | 7 | Scaling, unit conversion |
+| scheduling | 7 | Task dependencies, work rates |
+| units | 7 | Unit conversion with operations |
+
+### Difficulty Distribution
+
+| Difficulty | Count | Description |
+|------------|-------|-------------|
+| Easy | 28 | 2 steps, straightforward |
+| Medium | 32 | 2-3 steps, some complexity |
+| Hard | 7 | 3-4 steps, multiple operations |
+
+## Data Format
+
+Each problem is a JSON object with the following fields:
+
+```json
+{
+  "id": "math_discount_01",
+  "domain": "math_discount",
+  "problem": "A product costs $25 and is on 20% sale. You also have a $5 coupon. What do you pay? Answer with just the number.",
+  "correct_answer": "15",
+  "explanation": "25 × 0.8 = 20.0, then 20.0 - 5 = 15.0",
+  "understanding_check": "To solve this, first apply the 20% discount, then subtract the coupon. What are the two steps?",
+  "difficulty": "easy",
+  "steps": 2
+}
+```
+
+### Field Descriptions
+
+| Field | Description |
+|-------|-------------|
+| `id` | Unique identifier (domain_type_number) |
+| `domain` | Category of reasoning required |
+| `problem` | The question posed to the model |
+| `correct_answer` | Expected answer (numeric or text) |
+| `explanation` | Step-by-step solution |
+| `understanding_check` | Prompt to verify model understands the procedure |
+| `difficulty` | easy, medium, or hard |
+| `steps` | Number of sequential operations required |
+
+## Usage
+
+### Quick Evaluation
+
+```bash
+# Install requirements
+pip install requests
+
+# Evaluate OpenAI model
+python evaluate.py --provider openai --model gpt-4o -v
+
+# Evaluate Claude model
+python evaluate.py --provider anthropic --model claude-3-5-haiku-latest -v
+
+# Evaluate local Ollama model
+python evaluate.py --provider ollama --model llama3.1:8b -v
+```
+
+### Python API
+
+```python
+import json
+
+# Load dataset
+problems = []
+with open('data/test.jsonl') as f:
+    for line in f:
+        problems.append(json.loads(line))
+
+# Test your model
+for problem in problems:
+    response = your_model.generate(problem['problem'])
+    expected = problem['correct_answer']
+    # Validate response against expected
+```
+
+### With HuggingFace Datasets
+
+```python
+from datasets import load_dataset
+
+dataset = load_dataset("your-username/goodhart-gap-benchmark")
+
+for example in dataset['test']:
+    print(example['problem'])
+    print(f"Expected: {example['correct_answer']}")
+```
+
+## Evaluation Criteria
+
+A response is considered correct if:
+1. **Numeric answers**: The expected number appears in the response (with tolerance for rounding)
+2. **Time answers**: The expected time appears in any reasonable format (e.g., "4:45 PM", "4:45pm", "16:45")
+3. **Yes/no answers**: The response clearly indicates yes, no, or "cannot determine"
+4. **Ordering answers**: Items appear in the correct sequence
+
+## Leaderboard
+
+| Model | Provider | Pass Rate | Weakest Domain |
+|-------|----------|-----------|----------------|
+| Claude 3.5 Haiku | Anthropic | 93% | logic |
+| Claude Sonnet 4 | Anthropic | 79% | financial, scheduling |
+| gpt-4o | OpenAI | 57% | scheduling |
+| gpt-4o-mini | OpenAI | 36% | most domains |
+| Qwen 2.5 72B | Alibaba | TBD | - |
+| Llama 3.1 70B | Meta | TBD | - |
+
+*Submit your results via PR to add to the leaderboard*
+
+## Why This Matters
+
+### For AI Safety
+Models that can explain correct procedures but execute them incorrectly are:
+- Harder to detect through explanation-based evaluation
+- More dangerous in agentic settings
+- A gap between capability benchmarks and deployment readiness
+
+### For Model Selection
+Not all models are equal for multi-step reasoning:
+- Model family matters more than size
+- Distilled models often lose this capability
+- Test execution, not just explanation
+
+### For Training
+The gap appears to be a training problem:
+- Well-trained models (Claude Haiku) outperform larger models
+- Suggests targeted fine-tuning could help
+
+## Citation
+
+```bibtex
+@dataset{goodhart_gap_benchmark_2025,
+  title={Goodhart Gap Benchmark: Detecting the Gap Between Understanding and Execution in LLMs},
+  author={[Your Name]},
+  year={2025},
+  url={https://huggingface.co/datasets/your-username/goodhart-gap-benchmark}
+}
+```
+
+## License
+
+MIT License - free for research and commercial use.
+
+## Contributing
+
+We welcome contributions:
+- New test cases in underrepresented domains
+- Results from additional models
+- Improved validators
+- Translations to other languages
+
+Submit issues and PRs at: [GitHub Repository URL]
+
+## Acknowledgments
+
+Research inspired by:
+- Goodhart's Law and its application to AI evaluation
+- Work on multi-step reasoning in LLMs
+- The distinction between System 1 and System 2 thinking

data/baseline_results.json

@@ -0,0 +1,78 @@
+{
+  "benchmark_version": "1.0",
+  "date": "2025-01-03",
+  "results": [
+    {
+      "model": "claude-3-5-haiku-latest",
+      "provider": "anthropic",
+      "pass_rate": 0.93,
+      "passed": 13,
+      "total": 14,
+      "notes": "Multi-domain test (14 problems)"
+    },
+    {
+      "model": "claude-sonnet-4-20250514",
+      "provider": "anthropic",
+      "pass_rate": 0.79,
+      "passed": 11,
+      "total": 14,
+      "notes": "Multi-domain test (14 problems)"
+    },
+    {
+      "model": "gpt-4o",
+      "provider": "openai",
+      "pass_rate": 0.57,
+      "passed": 8,
+      "total": 14,
+      "notes": "Multi-domain test (14 problems)"
+    },
+    {
+      "model": "gpt-4o-mini",
+      "provider": "openai",
+      "pass_rate": 0.36,
+      "passed": 5,
+      "total": 14,
+      "notes": "Multi-domain test (14 problems)"
+    },
+    {
+      "model": "qwen2.5:72b",
+      "provider": "ollama/runpod",
+      "pass_rate": 1.0,
+      "passed": 1,
+      "total": 1,
+      "notes": "Canonical discount test only"
+    },
+    {
+      "model": "deepseek-llm:67b",
+      "provider": "ollama/runpod",
+      "pass_rate": 1.0,
+      "passed": 1,
+      "total": 1,
+      "notes": "Canonical discount test only"
+    },
+    {
+      "model": "llama3.1:70b",
+      "provider": "ollama/runpod",
+      "pass_rate": 0.0,
+      "passed": 0,
+      "total": 1,
+      "notes": "Canonical discount test only - answered 17"
+    },
+    {
+      "model": "llama3.1:8b",
+      "provider": "ollama",
+      "pass_rate": 0.0,
+      "passed": 0,
+      "total": 1,
+      "notes": "Canonical discount test only - answered $22.50"
+    },
+    {
+      "model": "mistral:7b",
+      "provider": "ollama",
+      "pass_rate": 1.0,
+      "passed": 1,
+      "total": 1,
+      "notes": "Canonical discount test only - PASSED"
+    }
+  ]
+}

data/summary.json

@@ -0,0 +1,22 @@
+{
+  "total_problems": 67,
+  "domains": {
+    "math_discount": 15,
+    "time": 13,
+    "recipe": 7,
+    "financial": 10,
+    "units": 7,
+    "scheduling": 7,
+    "logic": 8
+  },
+  "difficulty_distribution": {
+    "easy": 26,
+    "medium": 34,
+    "hard": 7
+  },
+  "step_distribution": {
+    "2": 51,
+    "3": 13,
+    "4": 3
+  }
+}

data/test.jsonl

@@ -0,0 +1,67 @@
+{"id": "math_discount_01", "domain": "math_discount", "problem": "A product costs $25 and is on 20% sale. You also have a $5 coupon. What do you pay? Answer with just the number.", "correct_answer": "15", "explanation": "25 \u00d7 0.8 = 20.0, then 20.0 - 5 = 15.0", "understanding_check": "To solve this, first apply the 20% discount, then subtract the coupon. What are the two steps?", "difficulty": "easy", "steps": 2}
+{"id": "math_discount_02", "domain": "math_discount", "problem": "A product costs $50 and is on 10% sale. You also have a $8 coupon. What do you pay? Answer with just the number.", "correct_answer": "37", "explanation": "50 \u00d7 0.9 = 45.0, then 45.0 - 8 = 37.0", "understanding_check": "To solve this, first apply the 10% discount, then subtract the coupon. What are the two steps?", "difficulty": "easy", "steps": 2}
+{"id": "math_discount_03", "domain": "math_discount", "problem": "A product costs $80 and is on 25% sale. You also have a $10 coupon. What do you pay? Answer with just the number.", "correct_answer": "50", "explanation": "80 \u00d7 0.75 = 60.0, then 60.0 - 10 = 50.0", "understanding_check": "To solve this, first apply the 25% discount, then subtract the coupon. What are the two steps?", "difficulty": "easy", "steps": 2}
+{"id": "math_discount_04", "domain": "math_discount", "problem": "A product costs $120 and is on 15% sale. You also have a $12 coupon. What do you pay? Answer with just the number.", "correct_answer": "90", "explanation": "120 \u00d7 0.85 = 102.0, then 102.0 - 12 = 90.0", "understanding_check": "To solve this, first apply the 15% discount, then subtract the coupon. What are the two steps?", "difficulty": "medium", "steps": 2}
+{"id": "math_discount_05", "domain": "math_discount", "problem": "A product costs $200 and is on 30% sale. You also have a $25 coupon. What do you pay? Answer with just the number.", "correct_answer": "115", "explanation": "200 \u00d7 0.7 = 140.0, then 140.0 - 25 = 115.0", "understanding_check": "To solve this, first apply the 30% discount, then subtract the coupon. What are the two steps?", "difficulty": "medium", "steps": 2}
+{"id": "math_discount_06", "domain": "math_discount", "problem": "A product costs $75 and is on 20% sale. You also have a $7 coupon. What do you pay? Answer with just the number.", "correct_answer": "53", "explanation": "75 \u00d7 0.8 = 60.0, then 60.0 - 7 = 53.0", "understanding_check": "To solve this, first apply the 20% discount, then subtract the coupon. What are the two steps?", "difficulty": "easy", "steps": 2}
+{"id": "math_discount_07", "domain": "math_discount", "problem": "A product costs $150 and is on 40% sale. You also have a $20 coupon. What do you pay? Answer with just the number.", "correct_answer": "70", "explanation": "150 \u00d7 0.6 = 90.0, then 90.0 - 20 = 70.0", "understanding_check": "To solve this, first apply the 40% discount, then subtract the coupon. What are the two steps?", "difficulty": "medium", "steps": 2}
+{"id": "math_discount_tax_01", "domain": "math_discount", "problem": "An item costs $100. First apply a 20% discount, then add 10% sales tax. What's the final price? Answer with just the number.", "correct_answer": "88", "explanation": "100 \u00d7 0.8 = 80.0, then 80.0 \u00d7 1.1 = 88.0", "understanding_check": "First apply the discount, then calculate tax on the discounted price. What are the steps?", "difficulty": "medium", "steps": 2}
+{"id": "math_discount_tax_02", "domain": "math_discount", "problem": "An item costs $250. First apply a 15% discount, then add 8% sales tax. What's the final price? Answer with just the number.", "correct_answer": "229.5", "explanation": "250 \u00d7 0.85 = 212.5, then 212.5 \u00d7 1.08 = 229.50000000000003", "understanding_check": "First apply the discount, then calculate tax on the discounted price. What are the steps?", "difficulty": "medium", "steps": 2}
+{"id": "math_discount_tax_03", "domain": "math_discount", "problem": "An item costs $80. First apply a 25% discount, then add 5% sales tax. What's the final price? Answer with just the number.", "correct_answer": "63", "explanation": "80 \u00d7 0.75 = 60.0, then 60.0 \u00d7 1.05 = 63.0", "understanding_check": "First apply the discount, then calculate tax on the discounted price. What are the steps?", "difficulty": "easy", "steps": 2}
+{"id": "math_discount_tax_04", "domain": "math_discount", "problem": "An item costs $500. First apply a 10% discount, then add 7% sales tax. What's the final price? Answer with just the number.", "correct_answer": "481.5", "explanation": "500 \u00d7 0.9 = 450.0, then 450.0 \u00d7 1.07 = 481.5", "understanding_check": "First apply the discount, then calculate tax on the discounted price. What are the steps?", "difficulty": "medium", "steps": 2}
+{"id": "math_discount_tax_05", "domain": "math_discount", "problem": "An item costs $160. First apply a 20% discount, then add 6% sales tax. What's the final price? Answer with just the number.", "correct_answer": "135.68", "explanation": "160 \u00d7 0.8 = 128.0, then 128.0 \u00d7 1.06 = 135.68", "understanding_check": "First apply the discount, then calculate tax on the discounted price. What are the steps?", "difficulty": "medium", "steps": 2}
+{"id": "math_bogo_01", "domain": "math_discount", "problem": "Shirts cost $40 each. Buy one, get 50% off the second. What's the total for 2 shirts? Answer with just the number.", "correct_answer": "60", "explanation": "First shirt: 40, Second shirt: 40 \u00d7 0.5 = 20.0, Total: 60.0", "understanding_check": "First shirt is full price, second shirt gets 50% off. How do you calculate the total?", "difficulty": "medium", "steps": 2}
+{"id": "math_bogo_02", "domain": "math_discount", "problem": "Shirts cost $25 each. Buy one, get 25% off the second. What's the total for 2 shirts? Answer with just the number.", "correct_answer": "43.75", "explanation": "First shirt: 25, Second shirt: 25 \u00d7 0.75 = 18.75, Total: 43.75", "understanding_check": "First shirt is full price, second shirt gets 25% off. How do you calculate the total?", "difficulty": "easy", "steps": 2}
+{"id": "math_bogo_03", "domain": "math_discount", "problem": "Shirts cost $60 each. Buy one, get 40% off the second. What's the total for 2 shirts? Answer with just the number.", "correct_answer": "96", "explanation": "First shirt: 60, Second shirt: 60 \u00d7 0.6 = 36.0, Total: 96.0", "understanding_check": "First shirt is full price, second shirt gets 40% off. How do you calculate the total?", "difficulty": "medium", "steps": 2}
+{"id": "time_duration_01", "domain": "time", "problem": "A meeting starts at 2:30 PM and lasts 1 hour 45 minutes. Then there's a 30 minute break. What time does the next session start? Answer with just the time.", "correct_answer": "4:45 PM", "explanation": "Add 105 minutes to 2:30 PM, then add 30 minutes", "understanding_check": "Add the meeting duration first, then add the break time. What are the steps?", "difficulty": "easy", "steps": 2}
+{"id": "time_duration_02", "domain": "time", "problem": "A meeting starts at 9:15 AM and lasts 2 hours 20 minutes. Then there's a 15 minute break. What time does the next session start? Answer with just the time.", "correct_answer": "11:50 AM", "explanation": "Add 140 minutes to 9:15 AM, then add 15 minutes", "understanding_check": "Add the meeting duration first, then add the break time. What are the steps?", "difficulty": "easy", "steps": 2}
+{"id": "time_duration_03", "domain": "time", "problem": "A meeting starts at 10:00 AM and lasts 1 hour 30 minutes. Then there's a 45 minute break. What time does the next session start? Answer with just the time.", "correct_answer": "12:15 PM", "explanation": "Add 90 minutes to 10:00 AM, then add 45 minutes", "understanding_check": "Add the meeting duration first, then add the break time. What are the steps?", "difficulty": "medium", "steps": 2}
+{"id": "time_duration_04", "domain": "time", "problem": "A meeting starts at 3:45 PM and lasts 1 hour 15 minutes. Then there's a 20 minute break. What time does the next session start? Answer with just the time.", "correct_answer": "5:20 PM", "explanation": "Add 75 minutes to 3:45 PM, then add 20 minutes", "understanding_check": "Add the meeting duration first, then add the break time. What are the steps?", "difficulty": "easy", "steps": 2}
+{"id": "time_duration_05", "domain": "time", "problem": "A meeting starts at 8:30 AM and lasts 3 hours. Then there's a 60 minute break. What time does the next session start? Answer with just the time.", "correct_answer": "1:30 PM", "explanation": "Add 180 minutes to 8:30 AM, then add 60 minutes", "understanding_check": "Add the meeting duration first, then add the break time. What are the steps?", "difficulty": "medium", "steps": 2}
+{"id": "time_duration_06", "domain": "time", "problem": "A meeting starts at 11:15 AM and lasts 0 hours 45 minutes. Then there's a 30 minute break. What time does the next session start? Answer with just the time.", "correct_answer": "12:30 PM", "explanation": "Add 45 minutes to 11:15 AM, then add 30 minutes", "understanding_check": "Add the meeting duration first, then add the break time. What are the steps?", "difficulty": "easy", "steps": 2}
+{"id": "time_duration_07", "domain": "time", "problem": "A meeting starts at 7:00 PM and lasts 2 hours. Then there's a 15 minute break. What time does the next session start? Answer with just the time.", "correct_answer": "9:15 PM", "explanation": "Add 120 minutes to 7:00 PM, then add 15 minutes", "understanding_check": "Add the meeting duration first, then add the break time. What are the steps?", "difficulty": "easy", "steps": 2}
+{"id": "time_travel_01", "domain": "time", "problem": "A train departs at 9:00 AM. The journey takes 2 hours 30 minutes. After arrival, you wait 20 minutes for a connection. What time do you board the connection? Answer with just the time.", "correct_answer": "11:50 AM", "explanation": "Add 150 minutes travel, then 20 minutes wait", "understanding_check": "Calculate arrival time first, then add wait time. What are the steps?", "difficulty": "medium", "steps": 2}
+{"id": "time_travel_02", "domain": "time", "problem": "A train departs at 2:15 PM. The journey takes 1 hour 15 minutes. After arrival, you wait 10 minutes for a connection. What time do you board the connection? Answer with just the time.", "correct_answer": "3:40 PM", "explanation": "Add 75 minutes travel, then 10 minutes wait", "understanding_check": "Calculate arrival time first, then add wait time. What are the steps?", "difficulty": "easy", "steps": 2}
+{"id": "time_travel_03", "domain": "time", "problem": "A train departs at 6:30 AM. The journey takes 3 hours. After arrival, you wait 30 minutes for a connection. What time do you board the connection? Answer with just the time.", "correct_answer": "10:00 AM", "explanation": "Add 180 minutes travel, then 30 minutes wait", "understanding_check": "Calculate arrival time first, then add wait time. What are the steps?", "difficulty": "medium", "steps": 2}
+{"id": "time_travel_04", "domain": "time", "problem": "A train departs at 4:00 PM. The journey takes 0 hours 45 minutes. After arrival, you wait 15 minutes for a connection. What time do you board the connection? Answer with just the time.", "correct_answer": "5:00 PM", "explanation": "Add 45 minutes travel, then 15 minutes wait", "understanding_check": "Calculate arrival time first, then add wait time. What are the steps?", "difficulty": "easy", "steps": 2}
+{"id": "time_travel_05", "domain": "time", "problem": "A train departs at 7:45 AM. The journey takes 1 hour 35 minutes. After arrival, you wait 25 minutes for a connection. What time do you board the connection? Answer with just the time.", "correct_answer": "9:45 AM", "explanation": "Add 95 minutes travel, then 25 minutes wait", "understanding_check": "Calculate arrival time first, then add wait time. What are the steps?", "difficulty": "medium", "steps": 2}
+{"id": "time_multi_01", "domain": "time", "problem": "You leave home at 8:00 AM. Drive 45 minutes to the station, wait 20 minutes, then take a 1 hour 15 minute train. What time do you arrive? Answer with just the time.", "correct_answer": "10:20 AM", "explanation": "8:00 + 0:45 = 8:45, + 0:20 = 9:05, + 1:15 = 10:20 AM", "understanding_check": "Add drive time, then wait time, then train time. What's the sequence?", "difficulty": "hard", "steps": 3}
+{"id": "recipe_scale_01", "domain": "recipe", "problem": "A recipe for 4 people needs 2 cups of flour. Scale to 6 people, then doubled for a party. How much cups of flour do you need? Answer with just the number.", "correct_answer": "6", "explanation": "2 \u00d7 (6/4) = 3.0, then \u00d7 2 = 6", "understanding_check": "First scale the recipe from 4 to 6 servings, then doubled. What are the steps?", "difficulty": "easy", "steps": 2}
+{"id": "recipe_scale_02", "domain": "recipe", "problem": "A recipe for 8 people needs 3 eggs. Scale to 12 people, then halved for a party. How much eggs  do you need? Answer with just the number.", "correct_answer": "2.25", "explanation": "3 \u00d7 (12/8) = 4.5, then \u00d7 0.5 = 2.25", "understanding_check": "First scale the recipe from 8 to 12 servings, then halved. What are the steps?", "difficulty": "medium", "steps": 2}
+{"id": "recipe_scale_03", "domain": "recipe", "problem": "A recipe for 4 people needs 1.5 cups of sugar. Scale to 8 people, then doubled for a party. How much cups of sugar do you need? Answer with just the number.", "correct_answer": "6", "explanation": "1.5 \u00d7 (8/4) = 3.0, then \u00d7 2 = 6", "understanding_check": "First scale the recipe from 4 to 8 servings, then doubled. What are the steps?", "difficulty": "easy", "steps": 2}
+{"id": "recipe_scale_04", "domain": "recipe", "problem": "A recipe for 6 people needs 4 tablespoons butter. Scale to 9 people, then halved for a party. How much tablespoons butter do you need? Answer with just the number.", "correct_answer": "3", "explanation": "4 \u00d7 (9/6) = 6.0, then \u00d7 0.5 = 3", "understanding_check": "First scale the recipe from 6 to 9 servings, then halved. What are the steps?", "difficulty": "medium", "steps": 2}
+{"id": "recipe_scale_05", "domain": "recipe", "problem": "A recipe for 5 people needs 2 cups of milk. Scale to 10 people, then multiplied by 1.5 for a party. How much cups of milk do you need? Answer with just the number.", "correct_answer": "6", "explanation": "2 \u00d7 (10/5) = 4.0, then \u00d7 1.5 = 6", "understanding_check": "First scale the recipe from 5 to 10 servings, then multiplied by 1.5. What are the steps?", "difficulty": "medium", "steps": 2}
+{"id": "recipe_convert_01", "domain": "recipe", "problem": "A recipe needs 2 cups of milk (1 cup = 240ml). Convert to ml, then reduce by 25% for a lighter version. How many ml? Answer with just the number.", "correct_answer": "360", "explanation": "2 \u00d7 240 = 480ml, then 480 \u00d7 0.75 = 360ml", "understanding_check": "Convert cups to ml first, then reduce by the percentage. What are the steps?", "difficulty": "medium", "steps": 2}
+{"id": "recipe_convert_02", "domain": "recipe", "problem": "A recipe uses 500g of flour. Convert to pounds (1 pound = 454g), then triple for a large batch. How many pounds? Answer with just the number rounded to one decimal.", "correct_answer": "3.3", "explanation": "500 / 454 = 1.1 pounds, then 1.1 \u00d7 3 = 3.3 pounds", "understanding_check": "Convert grams to pounds first, then triple. What are the steps?", "difficulty": "medium", "steps": 2}
+{"id": "financial_compound_01", "domain": "financial", "problem": "You invest $1000 at 10% annual interest for 2 years (compounded yearly). Then you pay 20% tax on the gains only. What's your final amount? Answer with just the number.", "correct_answer": "1168", "explanation": "1000 \u00d7 (1.10)^2 = 1210.00, gains = 210.00, tax = 42.00, final = 1168.00", "understanding_check": "Calculate compound interest first, then calculate tax only on the gains. What are the steps?", "difficulty": "medium", "steps": 3}
+{"id": "financial_compound_02", "domain": "financial", "problem": "You invest $5000 at 5% annual interest for 3 years (compounded yearly). Then you pay 15% tax on the gains only. What's your final amount? Answer with just the number.", "correct_answer": "5669.91", "explanation": "5000 \u00d7 (1.05)^3 = 5788.13, gains = 788.13, tax = 118.22, final = 5669.91", "understanding_check": "Calculate compound interest first, then calculate tax only on the gains. What are the steps?", "difficulty": "hard", "steps": 3}
+{"id": "financial_compound_03", "domain": "financial", "problem": "You invest $2000 at 8% annual interest for 2 years (compounded yearly). Then you pay 25% tax on the gains only. What's your final amount? Answer with just the number.", "correct_answer": "2249.6", "explanation": "2000 \u00d7 (1.08)^2 = 2332.80, gains = 332.80, tax = 83.20, final = 2249.60", "understanding_check": "Calculate compound interest first, then calculate tax only on the gains. What are the steps?", "difficulty": "medium", "steps": 3}
+{"id": "financial_compound_04", "domain": "financial", "problem": "You invest $500 at 12% annual interest for 2 years (compounded yearly). Then you pay 10% tax on the gains only. What's your final amount? Answer with just the number.", "correct_answer": "614.48", "explanation": "500 \u00d7 (1.12)^2 = 627.20, gains = 127.20, tax = 12.72, final = 614.48", "understanding_check": "Calculate compound interest first, then calculate tax only on the gains. What are the steps?", "difficulty": "medium", "steps": 3}
+{"id": "financial_markup_01", "domain": "financial", "problem": "A $500 item has 25% markup, then 10% member discount. What does a member pay? Answer with just the number.", "correct_answer": "562.5", "explanation": "500 \u00d7 1.25 = 625.0, then \u00d7 0.9 = 562.5", "understanding_check": "Apply markup first (increase), then discount (decrease). What are the steps?", "difficulty": "easy", "steps": 2}
+{"id": "financial_markup_02", "domain": "financial", "problem": "A $200 item has 50% markup, then 20% member discount. What does a member pay? Answer with just the number.", "correct_answer": "240", "explanation": "200 \u00d7 1.5 = 300.0, then \u00d7 0.8 = 240.0", "understanding_check": "Apply markup first (increase), then discount (decrease). What are the steps?", "difficulty": "easy", "steps": 2}
+{"id": "financial_markup_03", "domain": "financial", "problem": "A $800 item has 20% markup, then 15% member discount. What does a member pay? Answer with just the number.", "correct_answer": "816", "explanation": "800 \u00d7 1.2 = 960.0, then \u00d7 0.85 = 816.0", "understanding_check": "Apply markup first (increase), then discount (decrease). What are the steps?", "difficulty": "medium", "steps": 2}
+{"id": "financial_markup_04", "domain": "financial", "problem": "A $150 item has 40% markup, then 25% member discount. What does a member pay? Answer with just the number.", "correct_answer": "157.5", "explanation": "150 \u00d7 1.4 = 210.0, then \u00d7 0.75 = 157.5", "understanding_check": "Apply markup first (increase), then discount (decrease). What are the steps?", "difficulty": "medium", "steps": 2}
+{"id": "financial_markup_05", "domain": "financial", "problem": "A $1000 item has 30% markup, then 10% member discount. What does a member pay? Answer with just the number.", "correct_answer": "1170", "explanation": "1000 \u00d7 1.3 = 1300.0, then \u00d7 0.9 = 1170.0", "understanding_check": "Apply markup first (increase), then discount (decrease). What are the steps?", "difficulty": "medium", "steps": 2}
+{"id": "financial_commission_01", "domain": "financial", "problem": "A salesperson earns 5% on the first $10,000 of sales and 8% on anything above. They sold $15,000. What's their commission? Answer with just the number.", "correct_answer": "900", "explanation": "5% of 10000 = 500, 8% of 5000 = 400, total = 900", "understanding_check": "Calculate commission on first tier, then on second tier, then add. What are the steps?", "difficulty": "hard", "steps": 3}
+{"id": "unit_convert_01", "domain": "units", "problem": "Convert 10 miles to kilometers (1 mile = 1.6 km), add 5 km, then convert back to miles. How many miles? Answer with just the number.", "correct_answer": "13.125", "explanation": "10 \u00d7 1.6 = 16 km, 16 + 5 = 21 km, 21 \u00f7 1.6 = 13.125 miles", "understanding_check": "Convert to km, add, then convert back. What are the three steps?", "difficulty": "medium", "steps": 3}
+{"id": "unit_convert_02", "domain": "units", "problem": "Convert 100\u00b0F to Celsius (C = (F-32) \u00d7 5/9), subtract 10\u00b0C, then convert back to Fahrenheit. What's the temperature in \u00b0F? Answer with just the number.", "correct_answer": "82", "explanation": "(100-32) \u00d7 5/9 = 37.78\u00b0C, 37.78 - 10 = 27.78\u00b0C, 27.78 \u00d7 9/5 + 32 = 82\u00b0F", "understanding_check": "Convert F to C, subtract, then convert back. What are the steps?", "difficulty": "hard", "steps": 3}
+{"id": "unit_volume_01", "domain": "units", "problem": "You have 2 liters of water. Add 500ml, then pour out 1/4 of the total. How many ml remain? Answer with just the number.", "correct_answer": "1875", "explanation": "2000 + 500 = 2500ml, then 2500 \u00d7 0.75 = 1875ml", "understanding_check": "Add the volumes first, then calculate what remains after pouring out. What are the steps?", "difficulty": "easy", "steps": 2}
+{"id": "unit_volume_02", "domain": "units", "problem": "A tank holds 50 gallons. Drain 20%, then add 8 gallons. How many gallons now? Answer with just the number.", "correct_answer": "48", "explanation": "50 \u00d7 0.8 = 40 gallons, 40 + 8 = 48 gallons", "understanding_check": "First calculate remaining after draining, then add. What are the steps?", "difficulty": "easy", "steps": 2}
+{"id": "unit_volume_03", "domain": "units", "problem": "A pool holds 10,000 liters. Fill it to 75%, then drain 500 liters. How many liters remain? Answer with just the number.", "correct_answer": "7000", "explanation": "10000 \u00d7 0.75 = 7500 liters, 7500 - 500 = 7000 liters", "understanding_check": "Calculate 75% first, then subtract. What are the steps?", "difficulty": "easy", "steps": 2}
+{"id": "unit_speed_01", "domain": "units", "problem": "Drive 60 miles at 30 mph, then 40 miles at 40 mph. What's the total travel time in hours? Answer with just the number.", "correct_answer": "3", "explanation": "60/30 = 2 hours, 40/40 = 1 hour, total = 3 hours", "understanding_check": "Calculate time for each segment using distance/speed, then add. What are the steps?", "difficulty": "medium", "steps": 2}
+{"id": "unit_speed_02", "domain": "units", "problem": "A car travels 120 km in 1.5 hours, then 80 km in 1 hour. What's the average speed for the entire trip in km/h? Answer with just the number.", "correct_answer": "80", "explanation": "Total distance = 200 km, total time = 2.5 hours, average = 80 km/h", "understanding_check": "Calculate total distance and total time, then divide. What are the steps?", "difficulty": "medium", "steps": 2}
+{"id": "schedule_01", "domain": "scheduling", "problem": "Task A takes 2 hours. Task B takes 3 hours and must start after A finishes. Task C takes 1 hour and runs parallel to B. Starting at 9 AM, when do all tasks finish? Answer with just the time.", "correct_answer": "2:00 PM", "explanation": "A: 9-11 AM, B: 11 AM-2 PM (C runs parallel 11-12). All done at 2 PM", "understanding_check": "A must finish before B starts, C is parallel to B. What determines the end time?", "difficulty": "medium", "steps": 2}
+{"id": "schedule_02", "domain": "scheduling", "problem": "Process X takes 45 minutes. Process Y takes 30 minutes and needs X's output. Process Z takes 20 minutes and needs Y's output. Total time from start to finish? Answer in minutes.", "correct_answer": "95", "explanation": "45 + 30 + 20 = 95 minutes (sequential dependency chain)", "understanding_check": "X must complete before Y, Y before Z. They're sequential. What's the total?", "difficulty": "easy", "steps": 3}
+{"id": "schedule_03", "domain": "scheduling", "problem": "Download takes 10 minutes. Install takes 15 minutes (after download). Configuration takes 5 minutes (after install). Testing takes 20 minutes (after config). Total time? Answer in minutes.", "correct_answer": "50", "explanation": "10 + 15 + 5 + 20 = 50 minutes", "understanding_check": "Each step depends on the previous. How do you calculate total time?", "difficulty": "easy", "steps": 4}
+{"id": "schedule_04", "domain": "scheduling", "problem": "Path 1: Tasks A(2h) then B(3h). Path 2: Task C(4h). Both paths must complete. Starting at 10 AM, when is everything done? Answer with just the time.", "correct_answer": "3:00 PM", "explanation": "Path 1: 2+3=5 hours. Path 2: 4 hours. Critical path is 5 hours. 10 AM + 5h = 3 PM", "understanding_check": "Find the longest path (critical path). That determines when everything finishes.", "difficulty": "medium", "steps": 2}
+{"id": "schedule_05", "domain": "scheduling", "problem": "Team A: 3 tasks of 20 mins each (sequential). Team B: 2 tasks of 25 mins each (sequential). Both teams work in parallel. When do both finish? Answer in minutes from start.", "correct_answer": "60", "explanation": "Team A: 60 mins. Team B: 50 mins. Both done when slower team finishes = 60 mins", "understanding_check": "Teams work in parallel but tasks within each team are sequential. What's the critical path?", "difficulty": "medium", "steps": 2}
+{"id": "schedule_06", "domain": "scheduling", "problem": "Worker A completes a job in 6 hours. Worker B completes it in 4 hours. Working together, how long to complete one job? Answer in hours as a decimal.", "correct_answer": "2.4", "explanation": "Rate A = 1/6, Rate B = 1/4. Combined = 1/6 + 1/4 = 5/12. Time = 12/5 = 2.4 hours", "understanding_check": "Add work rates (1/time), then take reciprocal for combined time. What are the steps?", "difficulty": "hard", "steps": 3}
+{"id": "schedule_07", "domain": "scheduling", "problem": "A printer prints 30 pages/min. Another prints 20 pages/min. How long to print 250 pages together? Answer in minutes.", "correct_answer": "5", "explanation": "Combined rate = 50 pages/min. 250 \u00f7 50 = 5 minutes", "understanding_check": "Add the rates together, then divide total pages by combined rate. What are the steps?", "difficulty": "easy", "steps": 2}
+{"id": "logic_order_01", "domain": "logic", "problem": "In a race: Alice finishes before Bob. Carol finishes after Bob but before Dave. Eve finishes between Alice and Bob. List the finish order from first to last, separated by commas.", "correct_answer": "Alice, Eve, Bob, Carol, Dave", "explanation": "From constraints: A < E < B < C < D", "understanding_check": "Each constraint gives you a partial ordering. Combine them to get the full order.", "difficulty": "medium", "steps": 4}
+{"id": "logic_order_02", "domain": "logic", "problem": "Five books on a shelf from left to right: Red is left of Blue. Green is right of Blue. Yellow is left of Red. Orange is between Blue and Green. What's the order left to right?", "correct_answer": "Yellow, Red, Blue, Orange, Green", "explanation": "Y < R < B < O < G", "understanding_check": "Each constraint tells you relative positions. Build the sequence step by step.", "difficulty": "medium", "steps": 4}
+{"id": "logic_modus_01", "domain": "logic", "problem": "If it rains, the ground is wet. If the ground is wet, the game is cancelled. It rained. Is the game cancelled? Answer yes or no.", "correct_answer": "yes", "explanation": "Rain \u2192 Wet \u2192 Cancelled. Rain is true, so Cancelled is true.", "understanding_check": "Follow the chain of implications: A implies B, B implies C, A is true.", "difficulty": "easy", "steps": 2}
+{"id": "logic_modus_02", "domain": "logic", "problem": "If the battery is dead, the car won't start. If the car won't start, I'll be late. If I'm late, I'll miss the meeting. The battery is dead. Will I miss the meeting? Answer yes or no.", "correct_answer": "yes", "explanation": "Dead battery \u2192 No start \u2192 Late \u2192 Miss meeting", "understanding_check": "Follow the implication chain from the given fact to the conclusion.", "difficulty": "easy", "steps": 3}
+{"id": "logic_modus_03", "domain": "logic", "problem": "All programmers know logic. All logicians are good at puzzles. Sam is a programmer. Is Sam good at puzzles? Answer yes, no, or cannot determine.", "correct_answer": "cannot determine", "explanation": "Sam is programmer \u2192 knows logic. But knowing logic \u2260 being a logician.", "understanding_check": "Check if the chain of implications is complete. Is there a gap?", "difficulty": "hard", "steps": 2}
+{"id": "logic_sets_01", "domain": "logic", "problem": "30 students take Math. 25 take Science. 10 take both. How many take at least one subject? Answer with just the number.", "correct_answer": "45", "explanation": "30 + 25 - 10 = 45 (inclusion-exclusion)", "understanding_check": "Add both groups, subtract the overlap to avoid double-counting.", "difficulty": "easy", "steps": 2}
+{"id": "logic_sets_02", "domain": "logic", "problem": "In a group of 50 people: 35 speak English, 30 speak Spanish, and 20 speak both. How many speak neither? Answer with just the number.", "correct_answer": "5", "explanation": "Either language: 35 + 30 - 20 = 45. Neither: 50 - 45 = 5", "understanding_check": "First find how many speak at least one language, then subtract from total.", "difficulty": "medium", "steps": 3}
+{"id": "logic_sets_03", "domain": "logic", "problem": "100 people surveyed about pets: 60 have dogs, 40 have cats, 15 have both, 25 have fish only. How many have no pets? Answer with just the number.", "correct_answer": "10", "explanation": "Dogs or cats: 60 + 40 - 15 = 85. Fish only adds 25 but we need just no pets. 85 + 25 = 110 > 100, so fish must overlap. Actually: 100 - (60+40-15) - 25 + overlap = need to recalc...", "understanding_check": "Apply inclusion-exclusion for dogs/cats, account for fish separately.", "difficulty": "hard", "steps": 3}

evaluate.py

@@ -0,0 +1,471 @@
+#!/usr/bin/env python3
+"""
+Goodhart Gap Benchmark Evaluation Script
+
+Evaluate any model on the Goodhart Gap benchmark to detect the gap
+between understanding and execution in multi-step reasoning.
+
+Usage:
+    # Using OpenAI API
+    python evaluate.py --provider openai --model gpt-4o
+
+    # Using Anthropic API
+    python evaluate.py --provider anthropic --model claude-3-5-haiku-latest
+
+    # Using local Ollama
+    python evaluate.py --provider ollama --model llama3.1:8b
+
+    # Using HuggingFace transformers
+    python evaluate.py --provider huggingface --model meta-llama/Llama-3.1-8B-Instruct
+
+    # Custom API endpoint
+    python evaluate.py --provider custom --model mymodel --api-url http://localhost:8000/v1
+
+Environment Variables:
+    OPENAI_API_KEY - Required for OpenAI provider
+    ANTHROPIC_API_KEY - Required for Anthropic provider
+    HF_TOKEN - Optional for gated HuggingFace models
+"""
+
+import argparse
+import json
+import os
+import re
+import sys
+from dataclasses import dataclass
+from datetime import datetime
+from pathlib import Path
+from typing import Optional, Callable
+import time
+
+# Optional imports
+try:
+    import requests
+    HAS_REQUESTS = True
+except ImportError:
+    HAS_REQUESTS = False
+
+@dataclass
+class TestResult:
+    id: str
+    domain: str
+    problem: str
+    expected: str
+    response: str
+    extracted_answer: str
+    passed: bool
+    latency_ms: float
+
+def extract_answer(response: str, expected: str) -> str:
+    """Extract the answer from model response."""
+    response = response.strip()
+
+    # Try to find numbers in the response
+    numbers = re.findall(r'-?[\d,]+\.?\d*', response)
+
+    # For yes/no questions
+    if expected.lower() in ['yes', 'no']:
+        resp_lower = response.lower()
+        if 'yes' in resp_lower and 'no' not in resp_lower.split()[:3]:
+            return 'yes'
+        if 'no' in resp_lower and 'yes' not in resp_lower.split()[:3]:
+            return 'no'
+        if 'cannot determine' in resp_lower or 'cannot be determined' in resp_lower:
+            return 'cannot determine'
+
+    # For time answers
+    time_match = re.search(r'(\d{1,2}:\d{2})\s*(AM|PM|am|pm)?', response)
+    if time_match:
+        time_str = time_match.group(1)
+        period = time_match.group(2) or ''
+        return f"{time_str} {period}".strip()
+
+    # For ordering questions (comma-separated names)
+    if ',' in expected and any(c.isalpha() for c in expected):
+        # Try to extract comma-separated list
+        parts = [p.strip() for p in response.split(',') if p.strip()]
+        if len(parts) >= 3:
+            return ', '.join(parts[:5])
+
+    # Return first number found
+    if numbers:
+        return numbers[0].replace(',', '')
+
+    # Return first line or truncated response
+    first_line = response.split('\n')[0]
+    return first_line[:50] if len(first_line) > 50 else first_line
+
+def validate_answer(response: str, expected: str, domain: str) -> bool:
+    """Validate if the response matches the expected answer."""
+    response = response.lower().strip()
+    expected = expected.lower().strip()
+
+    # Direct match
+    if expected in response:
+        return True
+
+    # Numeric comparison
+    expected_nums = re.findall(r'-?[\d,]+\.?\d*', expected)
+    response_nums = re.findall(r'-?[\d,]+\.?\d*', response)
+
+    if expected_nums and response_nums:
+        try:
+            exp_val = float(expected_nums[0].replace(',', ''))
+            for resp_num in response_nums:
+                resp_val = float(resp_num.replace(',', ''))
+                # Allow small floating point tolerance
+                if abs(exp_val - resp_val) < 0.01:
+                    return True
+                # Check if it's within 0.5% (for rounding)
+                if exp_val != 0 and abs(exp_val - resp_val) / abs(exp_val) < 0.005:
+                    return True
+        except ValueError:
+            pass
+
+    # Time validation
+    if domain == 'time':
+        # Normalize time formats
+        def normalize_time(t):
+            t = t.lower().replace(' ', '')
+            t = re.sub(r'(\d{1,2}):(\d{2})(am|pm)?', r'\1:\2\3', t)
+            return t
+
+        if normalize_time(expected) in normalize_time(response):
+            return True
+
+    # Yes/no validation
+    if expected in ['yes', 'no', 'cannot determine']:
+        if expected == 'yes' and 'yes' in response and 'no' not in response.split()[:5]:
+            return True
+        if expected == 'no' and 'no' in response and 'yes' not in response.split()[:5]:
+            return True
+        if expected == 'cannot determine' and ('cannot' in response or 'unable' in response):
+            return True
+
+    # Ordering validation (check sequence)
+    if ',' in expected and domain == 'logic':
+        expected_items = [x.strip().lower() for x in expected.split(',')]
+        response_lower = response.lower()
+        # Check if items appear in correct order
+        positions = []
+        for item in expected_items:
+            pos = response_lower.find(item)
+            if pos == -1:
+                return False
+            positions.append(pos)
+        return positions == sorted(positions)
+
+    return False
+
+class ModelProvider:
+    """Base class for model providers."""
+
+    def generate(self, prompt: str) -> tuple[str, float]:
+        """Generate response. Returns (response, latency_ms)."""
+        raise NotImplementedError
+
+class OpenAIProvider(ModelProvider):
+    def __init__(self, model: str, api_key: Optional[str] = None):
+        self.model = model
+        self.api_key = api_key or os.environ.get('OPENAI_API_KEY')
+        if not self.api_key:
+            raise ValueError("OPENAI_API_KEY not set")
+
+    def generate(self, prompt: str) -> tuple[str, float]:
+        start = time.time()
+        headers = {
+            "Authorization": f"Bearer {self.api_key}",
+            "Content-Type": "application/json"
+        }
+        payload = {
+            "model": self.model,
+            "messages": [{"role": "user", "content": prompt}],
+            "temperature": 0.1,
+            "max_tokens": 200
+        }
+        response = requests.post(
+            "https://api.openai.com/v1/chat/completions",
+            headers=headers, json=payload, timeout=60
+        )
+        latency = (time.time() - start) * 1000
+
+        if response.status_code == 200:
+            return response.json()["choices"][0]["message"]["content"].strip(), latency
+        else:
+            return f"ERROR: {response.status_code}", latency
+
+class AnthropicProvider(ModelProvider):
+    def __init__(self, model: str, api_key: Optional[str] = None):
+        self.model = model
+        self.api_key = api_key or os.environ.get('ANTHROPIC_API_KEY')
+        if not self.api_key:
+            raise ValueError("ANTHROPIC_API_KEY not set")
+
+    def generate(self, prompt: str) -> tuple[str, float]:
+        start = time.time()
+        headers = {
+            "x-api-key": self.api_key,
+            "anthropic-version": "2023-06-01",
+            "Content-Type": "application/json"
+        }
+        payload = {
+            "model": self.model,
+            "max_tokens": 200,
+            "messages": [{"role": "user", "content": prompt}]
+        }
+        response = requests.post(
+            "https://api.anthropic.com/v1/messages",
+            headers=headers, json=payload, timeout=60
+        )
+        latency = (time.time() - start) * 1000
+
+        if response.status_code == 200:
+            return response.json()["content"][0]["text"].strip(), latency
+        else:
+            return f"ERROR: {response.status_code}", latency
+
+class OllamaProvider(ModelProvider):
+    def __init__(self, model: str, host: str = "http://localhost:11434"):
+        self.model = model
+        self.host = host
+
+    def generate(self, prompt: str) -> tuple[str, float]:
+        start = time.time()
+        payload = {
+            "model": self.model,
+            "prompt": prompt,
+            "stream": False,
+            "options": {"temperature": 0.1}
+        }
+        response = requests.post(
+            f"{self.host}/api/generate",
+            json=payload, timeout=120
+        )
+        latency = (time.time() - start) * 1000
+
+        if response.status_code == 200:
+            return response.json().get("response", "").strip(), latency
+        else:
+            return f"ERROR: {response.status_code}", latency
+
+class CustomProvider(ModelProvider):
+    def __init__(self, model: str, api_url: str):
+        self.model = model
+        self.api_url = api_url
+
+    def generate(self, prompt: str) -> tuple[str, float]:
+        start = time.time()
+        # Assume OpenAI-compatible API
+        payload = {
+            "model": self.model,
+            "messages": [{"role": "user", "content": prompt}],
+            "temperature": 0.1,
+            "max_tokens": 200
+        }
+        response = requests.post(
+            f"{self.api_url}/chat/completions",
+            json=payload, timeout=120
+        )
+        latency = (time.time() - start) * 1000
+
+        if response.status_code == 200:
+            return response.json()["choices"][0]["message"]["content"].strip(), latency
+        else:
+            return f"ERROR: {response.status_code}", latency
+
+def load_dataset(path: str = "data/test.jsonl") -> list[dict]:
+    """Load the benchmark dataset."""
+    problems = []
+    with open(path) as f:
+        for line in f:
+            problems.append(json.loads(line))
+    return problems
+
+def evaluate_model(
+    provider: ModelProvider,
+    problems: list[dict],
+    verbose: bool = False
+) -> tuple[list[TestResult], dict]:
+    """Evaluate a model on the benchmark."""
+
+    results = []
+    domain_stats = {}
+
+    for i, problem in enumerate(problems):
+        if verbose:
+            print(f"[{i+1}/{len(problems)}] {problem['id']}...", end=" ", flush=True)
+
+        response, latency = provider.generate(problem['problem'])
+        extracted = extract_answer(response, problem['correct_answer'])
+        passed = validate_answer(response, problem['correct_answer'], problem['domain'])
+
+        result = TestResult(
+            id=problem['id'],
+            domain=problem['domain'],
+            problem=problem['problem'],
+            expected=problem['correct_answer'],
+            response=response[:200],
+            extracted_answer=extracted,
+            passed=passed,
+            latency_ms=latency
+        )
+        results.append(result)
+
+        # Track domain stats
+        domain = problem['domain']
+        if domain not in domain_stats:
+            domain_stats[domain] = {'pass': 0, 'fail': 0}
+        domain_stats[domain]['pass' if passed else 'fail'] += 1
+
+        if verbose:
+            status = "PASS" if passed else "FAIL"
+            print(f"{status} (got: {extracted[:20]})")
+
+    # Calculate summary
+    total_pass = sum(r.passed for r in results)
+    total = len(results)
+
+    summary = {
+        'total': total,
+        'passed': total_pass,
+        'failed': total - total_pass,
+        'pass_rate': total_pass / total if total > 0 else 0,
+        'by_domain': {
+            d: {
+                'passed': s['pass'],
+                'total': s['pass'] + s['fail'],
+                'pass_rate': s['pass'] / (s['pass'] + s['fail'])
+            }
+            for d, s in domain_stats.items()
+        },
+        'avg_latency_ms': sum(r.latency_ms for r in results) / len(results) if results else 0
+    }
+
+    return results, summary
+
+def save_results(
+    results: list[TestResult],
+    summary: dict,
+    model_name: str,
+    output_dir: str = "results"
+):
+    """Save evaluation results."""
+    os.makedirs(output_dir, exist_ok=True)
+
+    timestamp = datetime.now().strftime("%Y%m%d_%H%M%S")
+    safe_model = re.sub(r'[^\w\-]', '_', model_name)
+
+    # Save detailed results
+    results_file = f"{output_dir}/{safe_model}_{timestamp}_results.jsonl"
+    with open(results_file, 'w') as f:
+        for r in results:
+            f.write(json.dumps({
+                'id': r.id,
+                'domain': r.domain,
+                'expected': r.expected,
+                'response': r.response,
+                'extracted': r.extracted_answer,
+                'passed': r.passed,
+                'latency_ms': r.latency_ms
+            }) + '\n')
+
+    # Save summary
+    summary_file = f"{output_dir}/{safe_model}_{timestamp}_summary.json"
+    summary['model'] = model_name
+    summary['timestamp'] = timestamp
+    with open(summary_file, 'w') as f:
+        json.dump(summary, f, indent=2)
+
+    return results_file, summary_file
+
+def print_summary(summary: dict, model_name: str):
+    """Print evaluation summary."""
+    print("\n" + "=" * 60)
+    print(f"GOODHART GAP BENCHMARK RESULTS")
+    print(f"Model: {model_name}")
+    print("=" * 60)
+
+    print(f"\nOverall: {summary['passed']}/{summary['total']} ({summary['pass_rate']*100:.1f}%)")
+    print(f"Average latency: {summary['avg_latency_ms']:.0f}ms")
+
+    print("\nBy Domain:")
+    print("-" * 40)
+    for domain, stats in sorted(summary['by_domain'].items()):
+        bar = "█" * int(stats['pass_rate'] * 10) + "░" * (10 - int(stats['pass_rate'] * 10))
+        print(f"  {domain:<15} {stats['passed']:>2}/{stats['total']:<2} {bar} {stats['pass_rate']*100:>5.1f}%")
+
+    print("\n" + "=" * 60)
+
+    # Interpret results
+    pass_rate = summary['pass_rate']
+    if pass_rate >= 0.9:
+        print("Assessment: LOW GOODHART GAP - Model executes well")
+    elif pass_rate >= 0.7:
+        print("Assessment: MODERATE GOODHART GAP - Some execution issues")
+    elif pass_rate >= 0.5:
+        print("Assessment: SIGNIFICANT GOODHART GAP - Frequent execution failures")
+    else:
+        print("Assessment: SEVERE GOODHART GAP - Major execution problems")
+
+def main():
+    parser = argparse.ArgumentParser(
+        description="Evaluate a model on the Goodhart Gap Benchmark",
+        formatter_class=argparse.RawDescriptionHelpFormatter,
+        epilog=__doc__
+    )
+    parser.add_argument('--provider', required=True,
+                       choices=['openai', 'anthropic', 'ollama', 'custom'],
+                       help='Model provider')
+    parser.add_argument('--model', required=True,
+                       help='Model name/identifier')
+    parser.add_argument('--api-url', default=None,
+                       help='API URL for custom provider')
+    parser.add_argument('--data', default='data/test.jsonl',
+                       help='Path to test data')
+    parser.add_argument('--output', default='results',
+                       help='Output directory')
+    parser.add_argument('--verbose', '-v', action='store_true',
+                       help='Show progress')
+    parser.add_argument('--limit', type=int, default=None,
+                       help='Limit number of problems (for testing)')
+
+    args = parser.parse_args()
+
+    if not HAS_REQUESTS:
+        print("ERROR: requests library required. Install with: pip install requests")
+        sys.exit(1)
+
+    # Create provider
+    if args.provider == 'openai':
+        provider = OpenAIProvider(args.model)
+    elif args.provider == 'anthropic':
+        provider = AnthropicProvider(args.model)
+    elif args.provider == 'ollama':
+        provider = OllamaProvider(args.model)
+    elif args.provider == 'custom':
+        if not args.api_url:
+            print("ERROR: --api-url required for custom provider")
+            sys.exit(1)
+        provider = CustomProvider(args.model, args.api_url)
+
+    # Load dataset
+    print(f"Loading dataset from {args.data}...")
+    problems = load_dataset(args.data)
+    if args.limit:
+        problems = problems[:args.limit]
+    print(f"Loaded {len(problems)} problems")
+
+    # Evaluate
+    print(f"\nEvaluating {args.model}...")
+    results, summary = evaluate_model(provider, problems, verbose=args.verbose)
+
+    # Save and print results
+    results_file, summary_file = save_results(results, summary, args.model, args.output)
+    print_summary(summary, args.model)
+
+    print(f"\nResults saved to:")
+    print(f"  {results_file}")
+    print(f"  {summary_file}")
+
+if __name__ == "__main__":
+    main()

requirements.txt

@@ -0,0 +1 @@
+requests>=2.28.0