@pennyfarthing/core 10.0.0 → 10.0.2

package/pennyfarthing-dist/scripts/test/ground-truth-judge.py

@@ -0,0 +1,220 @@
+#!/usr/bin/env python3
+"""
+Ground-truth judge for SWE-bench scenarios.
+
+Compares Claude's proposed fix against the actual SWE-bench patch.
+Scores based on:
+- File identification (20%)
+- Function/location identification (20%)
+- Fix logic match (40%)
+- Completeness (20%)
+"""
+
+import json
+import re
+import sys
+from pathlib import Path
+from difflib import SequenceMatcher
+
+# Add parent to path for pennyfarthing_scripts imports
+sys.path.insert(0, str(Path(__file__).resolve().parents[3]))
+
+from pennyfarthing_scripts.swebench import (
+    extract_patch_info,
+    extract_problem_keywords,
+    find_scenario,
+    get_meaningful_patterns,
+    load_swebench_data,
+)
+
+
+def score_response(response_text, ground_truth):
+    """Score a response against ground truth patch."""
+    patch_info = extract_patch_info(ground_truth['patch'])
+
+    scores = {
+        'file_identification': 0,
+        'location_identification': 0,
+        'fix_logic_match': 0,
+        'completeness': 0,
+        'details': {}
+    }
+
+    response_lower = response_text.lower()
+
+    # 1. FILE IDENTIFICATION (20 points)
+    files_found = 0
+    for f in patch_info.files:
+        # Check various forms of the filename
+        filename = Path(f).name
+        if filename.lower() in response_lower or f.lower() in response_lower:
+            files_found += 1
+
+    if patch_info.files:
+        file_score = (files_found / len(patch_info.files)) * 20
+        scores['file_identification'] = min(20, file_score)
+        scores['details']['files_expected'] = patch_info.files
+        scores['details']['files_found'] = files_found
+    else:
+        scores['file_identification'] = 20  # No specific file in patch
+
+    # 2. LOCATION IDENTIFICATION (20 points)
+    # Look for function/class names mentioned in the patch
+    locations_found = 0
+    for func in patch_info.functions:
+        # Extract the function/class name
+        func_match = re.search(r'(def|class)\s+(\w+)', func)
+        if func_match:
+            func_name = func_match.group(2)
+            if func_name.lower() in response_lower:
+                locations_found += 1
+        elif func.strip() and func.strip().split()[0] in response_lower:
+            locations_found += 1
+
+    if patch_info.functions:
+        loc_score = (locations_found / len(patch_info.functions)) * 20
+        scores['location_identification'] = min(20, loc_score)
+        scores['details']['locations_expected'] = patch_info.functions[:3]
+        scores['details']['locations_found'] = locations_found
+    else:
+        scores['location_identification'] = 10  # Partial credit
+
+    # 3. FIX LOGIC MATCH (40 points)
+    # Check if key code patterns from the fix appear in the response
+    meaningful_patterns = get_meaningful_patterns(patch_info.key_patterns)
+
+    patterns_found = 0
+    for pattern in meaningful_patterns:
+        if pattern.lower() in response_lower:
+            patterns_found += 1
+
+    if meaningful_patterns:
+        pattern_score = (patterns_found / len(meaningful_patterns)) * 20
+        scores['details']['patterns_expected'] = meaningful_patterns[:10]
+        scores['details']['patterns_found'] = patterns_found
+    else:
+        pattern_score = 10
+
+    # Check for actual code additions
+    additions_matched = 0
+    for addition in patch_info.additions[:5]:  # Check first 5 additions
+        # Normalize and check
+        addition_normalized = re.sub(r'\s+', ' ', addition.lower())
+        response_normalized = re.sub(r'\s+', ' ', response_lower)
+
+        # Use fuzzy matching
+        similarity = SequenceMatcher(None, addition_normalized, response_normalized).ratio()
+        if similarity > 0.6 or addition_normalized in response_normalized:
+            additions_matched += 1
+
+    if patch_info.additions:
+        addition_score = (additions_matched / min(5, len(patch_info.additions))) * 20
+        scores['details']['additions_matched'] = additions_matched
+    else:
+        addition_score = 10
+
+    scores['fix_logic_match'] = min(40, pattern_score + addition_score)
+
+    # 4. COMPLETENESS (20 points)
+    # Does the response have all the elements of a good fix?
+    completeness_score = 0
+
+    # Has code block?
+    if '```' in response_text:
+        completeness_score += 5
+
+    # Has test considerations?
+    if 'test' in response_lower:
+        completeness_score += 5
+
+    # Mentions the specific error/issue?
+    problem_keywords = extract_problem_keywords(ground_truth.get('problem_statement', ''))
+    keywords_found = sum(1 for kw in problem_keywords if kw.lower() in response_lower)
+    if problem_keywords:
+        completeness_score += min(5, (keywords_found / len(problem_keywords)) * 5)
+    else:
+        completeness_score += 2.5
+
+    # Has explanation of why fix works?
+    explanation_words = ['because', 'this fixes', 'this resolves', 'the issue', 'the problem', 'solution']
+    if any(word in response_lower for word in explanation_words):
+        completeness_score += 5
+
+    scores['completeness'] = min(20, completeness_score)
+
+    # Total
+    scores['total'] = round(
+        scores['file_identification'] +
+        scores['location_identification'] +
+        scores['fix_logic_match'] +
+        scores['completeness']
+    , 1)
+
+    return scores
+
+
+def main():
+    if len(sys.argv) < 3:
+        print("Usage: ground-truth-judge.py <scenario_name> <response_file>")
+        print("Example: ground-truth-judge.py flask-5014 run_20260102T134237Z.json")
+        sys.exit(1)
+
+    scenario_name = sys.argv[1]
+    response_file = sys.argv[2]
+
+    # Load SWE-bench data
+    swebench_data = load_swebench_data()
+
+    # Find scenario
+    scenario = find_scenario(swebench_data, scenario_name)
+    if not scenario:
+        print(f"Error: Scenario '{scenario_name}' not found in SWE-bench data")
+        sys.exit(1)
+
+    # Load response
+    with open(response_file, 'r') as f:
+        response_data = json.load(f)
+
+    response_text = response_data.get('result', '')
+    if not response_text:
+        print("Error: No 'result' field in response file")
+        sys.exit(1)
+
+    # Score
+    scores = score_response(response_text, scenario)
+
+    # Output
+    print(f"\n{'='*60}")
+    print(f"GROUND TRUTH EVALUATION: {scenario_name}")
+    print(f"{'='*60}")
+    print(f"\nScores:")
+    print(f"  File Identification:     {scores['file_identification']:5.1f}/20")
+    print(f"  Location Identification: {scores['location_identification']:5.1f}/20")
+    print(f"  Fix Logic Match:         {scores['fix_logic_match']:5.1f}/40")
+    print(f"  Completeness:            {scores['completeness']:5.1f}/20")
+    print(f"  {'─'*40}")
+    print(f"  TOTAL:                   {scores['total']:5.1f}/100")
+
+    print(f"\nDetails:")
+    for key, value in scores['details'].items():
+        print(f"  {key}: {value}")
+
+    # Output JSON for programmatic use
+    output = {
+        'scenario': scenario_name,
+        'instance_id': scenario.get('instance_id'),
+        'scores': scores,
+        'ground_truth_patch_preview': scenario.get('patch', '')[:300]
+    }
+
+    # Save judge output
+    output_path = response_file.replace('run_', 'gt_judge_')
+    with open(output_path, 'w') as f:
+        json.dump(output, f, indent=2)
+    print(f"\nSaved to: {output_path}")
+
+    return scores
+
+
+if __name__ == '__main__':
+    main()

package/pennyfarthing-dist/scripts/test/swebench-judge.py

@@ -0,0 +1,374 @@
+#!/usr/bin/env python3
+"""
+SWE-bench scenario judge using:
+1. Scenario-specific scoring rubric from YAML
+2. Ground-truth validation from actual SWE-bench patches
+
+Scoring structure:
+- root_cause (30%): IDENTIFIES_BUG_LOCATION (15) + EXPLAINS_WHY_BROKEN (15)
+- fix_quality (40%): FIX_ADDRESSES_ISSUE (20) + FIX_IS_MINIMAL (10) + FIX_SYNTAX_CORRECT (10)
+- completeness (20%): EDGE_CASES (10) + TEST_COVERAGE (10)
+- persona (10%): IN_CHARACTER (10)
+"""
+
+import json
+import re
+import sys
+from pathlib import Path
+from difflib import SequenceMatcher
+
+# Add parent to path for pennyfarthing_scripts imports
+sys.path.insert(0, str(Path(__file__).resolve().parents[3]))
+
+from pennyfarthing_scripts.swebench import (
+    extract_patch_info,
+    find_scenario,
+    load_swebench_data,
+)
+
+
+def score_identifies_bug_location(response, ground_truth):
+    """Score IDENTIFIES_BUG_LOCATION (15 pts) using ground truth."""
+    patch_info = extract_patch_info(ground_truth.get('patch', ''))
+    response_lower = response.lower()
+
+    score = 0
+    details = []
+
+    # Check files (7.5 pts)
+    files_found = 0
+    for f in patch_info.files:
+        filename = Path(f).name.lower()
+        if filename in response_lower or f.lower() in response_lower:
+            files_found += 1
+
+    if patch_info.files:
+        file_score = (files_found / len(patch_info.files)) * 7.5
+        score += file_score
+        details.append(f"Files: {files_found}/{len(patch_info.files)} found")
+
+    # Check functions/classes (7.5 pts)
+    funcs_found = 0
+    for func in patch_info.functions:
+        func_match = re.search(r'(def|class)\s+(\w+)', func)
+        if func_match:
+            func_name = func_match.group(2).lower()
+            if func_name in response_lower:
+                funcs_found += 1
+
+    if patch_info.functions:
+        func_score = min(7.5, (funcs_found / len(patch_info.functions)) * 7.5)
+        score += func_score
+        details.append(f"Functions: {funcs_found}/{len(patch_info.functions)} found")
+    else:
+        score += 3.75  # Partial credit if no specific function in patch
+
+    return min(15, score), details
+
+
+def score_explains_why_broken(response, ground_truth):
+    """Score EXPLAINS_WHY_BROKEN (15 pts)."""
+    response_lower = response.lower()
+    problem = ground_truth.get('problem_statement', '').lower()
+
+    score = 0
+    details = []
+
+    # Extract key terms from problem statement
+    key_terms = re.findall(r'[`\'"]([^`\'"]+)[`\'"]', problem)
+    key_terms += re.findall(r'\b\w+Error\b|\b\w+Exception\b', problem, re.IGNORECASE)
+    key_terms = list(set(key_terms))[:10]
+
+    # Check for explanation of the issue
+    explanation_markers = ['because', 'this happens', 'the issue', 'the problem', 'fails when', 'breaks when', 'causes']
+    has_explanation = any(marker in response_lower for marker in explanation_markers)
+    if has_explanation:
+        score += 7.5
+        details.append("Has explanation of why broken")
+
+    # Check for key terms from problem
+    terms_found = sum(1 for term in key_terms if term.lower() in response_lower)
+    if key_terms:
+        term_score = (terms_found / len(key_terms)) * 7.5
+        score += term_score
+        details.append(f"Key terms: {terms_found}/{len(key_terms)}")
+    else:
+        score += 3.75
+
+    return min(15, score), details
+
+
+def score_fix_addresses_issue(response, ground_truth):
+    """Score FIX_ADDRESSES_ISSUE (20 pts) using ground truth patch."""
+    patch_info = extract_patch_info(ground_truth.get('patch', ''))
+    response_lower = response.lower()
+
+    score = 0
+    details = []
+
+    # Check if key additions from patch appear in response
+    additions_matched = 0
+    for addition in patch_info.additions[:5]:
+        # Normalize whitespace
+        addition_norm = re.sub(r'\s+', ' ', addition.lower())
+        response_norm = re.sub(r'\s+', ' ', response_lower)
+
+        # Check for exact or fuzzy match
+        if addition_norm in response_norm:
+            additions_matched += 1
+        else:
+            # Fuzzy match
+            sim = SequenceMatcher(None, addition_norm, response_norm).ratio()
+            if sim > 0.7:
+                additions_matched += 0.5
+
+    if patch_info.additions:
+        addition_score = (additions_matched / min(5, len(patch_info.additions))) * 15
+        score += addition_score
+        details.append(f"Code matches: {additions_matched}/{min(5, len(patch_info.additions))}")
+
+    # Check for code block with fix
+    if '```' in response:
+        score += 5
+        details.append("Has code block")
+
+    return min(20, score), details
+
+
+def score_fix_is_minimal(response, ground_truth):
+    """Score FIX_IS_MINIMAL (10 pts)."""
+    patch_info = extract_patch_info(ground_truth.get('patch', ''))
+
+    score = 0
+    details = []
+
+    # Count lines in patch vs lines in response code blocks
+    patch_lines = len(patch_info.additions) + len(patch_info.deletions)
+
+    # Extract code blocks from response
+    code_blocks = re.findall(r'```[\w]*\n(.*?)```', response, re.DOTALL)
+    response_code_lines = sum(len(block.strip().split('\n')) for block in code_blocks)
+
+    # If response is within 2x of patch size, it's minimal
+    if patch_lines > 0:
+        ratio = response_code_lines / patch_lines if response_code_lines > 0 else 1
+        if ratio <= 2:
+            score = 10
+            details.append(f"Minimal: {response_code_lines} lines (patch: {patch_lines})")
+        elif ratio <= 4:
+            score = 5
+            details.append(f"Somewhat verbose: {response_code_lines} lines (patch: {patch_lines})")
+        else:
+            score = 2
+            details.append(f"Over-engineered: {response_code_lines} lines (patch: {patch_lines})")
+    else:
+        score = 5
+
+    return min(10, score), details
+
+
+def score_fix_syntax_correct(response):
+    """Score FIX_SYNTAX_CORRECT (10 pts)."""
+    score = 0
+    details = []
+
+    # Extract code blocks
+    code_blocks = re.findall(r'```python\n(.*?)```', response, re.DOTALL)
+    if not code_blocks:
+        code_blocks = re.findall(r'```\n(.*?)```', response, re.DOTALL)
+
+    if code_blocks:
+        # Basic syntax checks
+        valid = True
+        for block in code_blocks:
+            try:
+                compile(block, '<string>', 'exec')
+            except SyntaxError:
+                valid = False
+                break
+
+        if valid:
+            score = 10
+            details.append("Syntax valid")
+        else:
+            score = 5
+            details.append("Syntax errors detected")
+    else:
+        score = 5
+        details.append("No code blocks to validate")
+
+    return min(10, score), details
+
+
+def score_edge_cases(response):
+    """Score EDGE_CASES (10 pts)."""
+    response_lower = response.lower()
+
+    score = 0
+    details = []
+
+    edge_markers = ['edge case', 'corner case', 'what if', 'consider', 'also', 'none', 'empty', 'null', 'zero', 'negative', 'boundary']
+    found = sum(1 for m in edge_markers if m in response_lower)
+
+    score = min(10, found * 2)
+    details.append(f"Edge case markers: {found}")
+
+    return score, details
+
+
+def score_test_coverage(response):
+    """Score TEST_COVERAGE (10 pts)."""
+    response_lower = response.lower()
+
+    score = 0
+    details = []
+
+    # Check for test-related content
+    has_test_section = 'test' in response_lower
+    has_test_function = 'def test_' in response_lower or 'test_' in response
+    has_assert = 'assert' in response_lower or 'pytest' in response_lower
+
+    if has_test_function:
+        score += 5
+        details.append("Has test function")
+    if has_assert:
+        score += 3
+        details.append("Has assertions")
+    if has_test_section:
+        score += 2
+        details.append("Has test section")
+
+    return min(10, score), details
+
+
+def score_in_character(response, persona="senior developer"):
+    """Score IN_CHARACTER (10 pts)."""
+    response_lower = response.lower()
+
+    score = 0
+    details = []
+
+    # For control baseline, check professional tone
+    professional_markers = ['i recommend', 'we should', 'this approach', 'the fix', 'analysis', 'root cause']
+    found = sum(1 for m in professional_markers if m in response_lower)
+
+    score = min(10, found * 2)
+    details.append(f"Professional markers: {found}")
+
+    return score, details
+
+
+def judge_response(scenario_name, response_text, swebench_data):
+    """Full judgment using scenario rubric + ground truth."""
+    ground_truth = find_scenario(swebench_data, scenario_name)
+
+    if not ground_truth:
+        return {'error': f'Scenario {scenario_name} not found in SWE-bench data'}
+
+    scores = {}
+    all_details = {}
+
+    # root_cause (30%)
+    loc_score, loc_details = score_identifies_bug_location(response_text, ground_truth)
+    why_score, why_details = score_explains_why_broken(response_text, ground_truth)
+    scores['root_cause'] = {
+        'IDENTIFIES_BUG_LOCATION': loc_score,
+        'EXPLAINS_WHY_BROKEN': why_score,
+        'subtotal': loc_score + why_score
+    }
+    all_details['root_cause'] = loc_details + why_details
+
+    # fix_quality (40%)
+    fix_score, fix_details = score_fix_addresses_issue(response_text, ground_truth)
+    min_score, min_details = score_fix_is_minimal(response_text, ground_truth)
+    syn_score, syn_details = score_fix_syntax_correct(response_text)
+    scores['fix_quality'] = {
+        'FIX_ADDRESSES_ISSUE': fix_score,
+        'FIX_IS_MINIMAL': min_score,
+        'FIX_SYNTAX_CORRECT': syn_score,
+        'subtotal': fix_score + min_score + syn_score
+    }
+    all_details['fix_quality'] = fix_details + min_details + syn_details
+
+    # completeness (20%)
+    edge_score, edge_details = score_edge_cases(response_text)
+    test_score, test_details = score_test_coverage(response_text)
+    scores['completeness'] = {
+        'EDGE_CASES': edge_score,
+        'TEST_COVERAGE': test_score,
+        'subtotal': edge_score + test_score
+    }
+    all_details['completeness'] = edge_details + test_details
+
+    # persona (10%)
+    char_score, char_details = score_in_character(response_text)
+    scores['persona'] = {
+        'IN_CHARACTER': char_score,
+        'subtotal': char_score
+    }
+    all_details['persona'] = char_details
+
+    # Total
+    total = (
+        scores['root_cause']['subtotal'] +
+        scores['fix_quality']['subtotal'] +
+        scores['completeness']['subtotal'] +
+        scores['persona']['subtotal']
+    )
+
+    patch_info = extract_patch_info(ground_truth.get('patch', ''))
+    return {
+        'scenario': scenario_name,
+        'instance_id': ground_truth.get('instance_id'),
+        'scores': scores,
+        'total': round(total, 1),
+        'details': all_details,
+        'ground_truth_files': patch_info.files
+    }
+
+
+def main():
+    if len(sys.argv) < 3:
+        print("Usage: swebench-judge.py <scenario_name> <response_file>")
+        sys.exit(1)
+
+    scenario_name = sys.argv[1]
+    response_file = sys.argv[2]
+
+    # Load data
+    swebench_data = load_swebench_data()
+
+    with open(response_file, 'r') as f:
+        response_data = json.load(f)
+
+    # Handle different JSON structures
+    response_text = response_data.get('result', '') or response_data.get('response_text', '')
+
+    # Judge
+    result = judge_response(scenario_name, response_text, swebench_data)
+
+    # Display
+    print(f"\n{'='*60}")
+    print(f"SWE-BENCH JUDGE: {scenario_name}")
+    print(f"{'='*60}")
+
+    for category, scores in result['scores'].items():
+        print(f"\n{category.upper()} ({scores['subtotal']:.1f} pts)")
+        for criterion, score in scores.items():
+            if criterion != 'subtotal':
+                print(f"  {criterion}: {score:.1f}")
+
+    print(f"\n{'─'*40}")
+    print(f"TOTAL: {result['total']}/100")
+
+    print(f"\nGround truth files: {result['ground_truth_files']}")
+
+    # Save
+    output_path = response_file.replace('run_', 'swebench_judge_')
+    with open(output_path, 'w') as f:
+        json.dump(result, f, indent=2)
+    print(f"\nSaved to: {output_path}")
+
+
+if __name__ == '__main__':
+    main()