diagnose_scoring.py

#!/usr/bin/env python3
"""
Diagnostic Script - Understanding Why Empty/Minimal Resumes Get High Scores
"""

from app import ATSCompatibilityAnalyzer

analyzer = ATSCompatibilityAnalyzer()

# Test cases to diagnose
test_cases = [
    ("Empty Resume", "", "Software Engineer with Python experience"),
    ("Just Name", "John Doe", "Looking for a Data Analyst with SQL and Python"),
    ("Just 'Hi'", "Hi", "Senior Software Engineer"),
    ("Random Gibberish", "asdfghjkl qwertyuiop zxcvbnm", "Machine Learning Engineer"),
    ("Chef for ML Role", "Chef John - 10 years cooking - French cuisine, pastry", "Machine Learning Engineer with PhD and PyTorch"),
]

print("=" * 80)
print("DIAGNOSIS: WHY ARE EMPTY/MINIMAL RESUMES GETTING HIGH SCORES?")
print("=" * 80)

for name, resume, jd in test_cases:
    print(f"\n{'='*60}")
    print(f"TEST: {name}")
    print(f"Resume: '{resume[:50]}...' ({len(resume)} chars)" if len(resume) > 50 else f"Resume: '{resume}' ({len(resume)} chars)")
    print(f"JD: '{jd[:50]}...'" if len(jd) > 50 else f"JD: '{jd}'")
    print("-" * 60)
    
    result = analyzer.analyze(resume, jd)
    
    print(f"\n📊 TOTAL SCORE: {result['total_score']}%  <-- THIS IS THE PROBLEM!")
    print("\n🔍 BREAKDOWN (with weights):")
    
    weights = analyzer.weights
    for metric, score in result['breakdown'].items():
        weight = weights.get(metric, 0)
        weighted = score * weight
        print(f"   {metric:20} = {score:5.1f}% × {weight:.2f} = {weighted:5.1f}")
    
    print(f"\n   {'='*40}")
    print(f"   WEIGHTED TOTAL: {result['total_score']}%")

print("\n\n" + "=" * 80)
print("🐛 ROOT CAUSE ANALYSIS")
print("=" * 80)
print("""
The scoring functions have ARTIFICIALLY HIGH BASELINES:

1. _format_score:       baseline = 80  (even empty resume gets 80)
2. _section_score:      baseline = 80  (even no sections gets 80)
3. _action_verb_score:  baseline = 75  (0 verbs = 75%)
4. _quantification:     baseline = 68  (0 numbers = 68%)
5. _tfidf_score:        60 + (raw * 0.45)  (0% match = 60%)
6. _skills_match:       baseline = 75  (0 matches = 75%)
7. _semantic_match:     returns 75-85 default with no match

This design was meant to prevent "harsh" scoring but it's BROKEN:
- Empty resumes should score 0-10%, not 70%+
- Completely irrelevant resumes should score <30%, not 80%+

RECOMMENDED FIX:
- Remove artificial baselines
- Score from 0, not from 60-80
- Apply minimum thresholds for valid input
""")

print("\n" + "=" * 80)
print("WEIGHTS BEING USED:")
print("=" * 80)
for metric, weight in analyzer.weights.items():
    print(f"   {metric:20} = {weight:.2f}")
print(f"\nTotal weights sum: {sum(analyzer.weights.values()):.2f}")