test_generation.py

"""
Test the game generation pipeline with retrieval and generation.

Run this script to:
1. Load and normalize the dataset
2. Retrieve similar games for a sample config
3. Generate a game using the generator
4. Validate the generated game against the schema
5. Display the results
"""

import json
from app.services.retrieval import load_games_dataset, normalize_game_record, retrieve_examples
from app.services.generator import generate_game, build_generation_prompt
from app.services.schema_validator import validate_game_schema


def main():
    print("\n" + "=" * 80)
    print("PHASE 2, TASK 6: GAME GENERATION TEST")
    print("=" * 80)
    
    # Load dataset
    print("\n1. Loading dataset...")
    raw_records = load_games_dataset("app/data/games_dataset.json")
    normalized_records = [normalize_game_record(r) for r in raw_records]
    print(f"✓ Loaded {len(normalized_records)} records")
    
    # Test cases with different configs
    test_configs = [
        {
            "name": "Scavenger Hunt - Adults - Medium - Le Marais",
            "config": {
                "game_type": "scavenger_hunt",
                "city": "Paris",
                "area": "Le Marais",
                "location_type": "mixed",
                "duration_minutes": 60,
                "num_players": 4,
                "difficulty": "medium",
                "age_group": "adults",
                "photo_enabled": True
            }
        },
        {
            "name": "Hide & Seek - Kids - Easy - Park",
            "config": {
                "game_type": "hide_and_seek",
                "city": "Paris",
                "area": "Parc des Buttes-Chaumont",
                "location_type": "park",
                "duration_minutes": 45,
                "num_players": 5,
                "difficulty": "easy",
                "age_group": "kids",
                "photo_enabled": False
            }
        },
        {
            "name": "Tag - Teens - Hard - Trocadéro",
            "config": {
                "game_type": "tag",
                "city": "Paris",
                "area": "Jardins du Trocadéro",
                "location_type": "park",
                "duration_minutes": 30,
                "num_players": 8,
                "difficulty": "hard",
                "age_group": "teens",
                "photo_enabled": False
            }
        }
    ]
    
    # Test each config
    results = []
    for test in test_configs:
        print("\n" + "=" * 80)
        print(f"TEST: {test['name']}")
        print("=" * 80)
        
        config = test['config']
        
        # Step 1: Retrieve similar games
        print("\n2. Retrieving similar games...")
        retrieved = retrieve_examples(config, normalized_records, k=3)
        print(f"✓ Retrieved {len(retrieved)} similar games:")
        for i, ex in enumerate(retrieved, 1):
            print(f"  {i}. {ex['id']} (score: {ex['retrieval_score']:.1f})")
        
        # Step 2: Show prompt snippet
        print("\n3. Building generation prompt...")
        prompt = build_generation_prompt(config, retrieved)
        print(f"✓ Prompt built ({len(prompt)} chars)")
        print(f"\nPrompt preview:")
        print(prompt[:300] + "...\n")
        
        # Step 3: Generate game
        print("4. Generating game...")
        try:
            game = generate_game(config, retrieved)
            print(f"✓ Game generated: {game['game_id']}")
            
            # Step 4: Validate against schema
            print("\n5. Validating against schema...")
            is_valid, errors = validate_game_schema(game)
            
            if is_valid:
                print("✓ Game VALID against schema")
            else:
                print(f"✗ Game INVALID - {len(errors)} errors:")
                for error in errors[:3]:
                    print(f"  - {error}")
            
            # Step 5: Display game details
            print("\n6. Generated Game Details:")
            print(f"  ID: {game['game_id']}")
            print(f"  Title: {game['title']}")
            print(f"  Theme: {game['theme']}")
            print(f"  Area: {game['setup']['area']}")
            print(f"  Duration: {game['setup']['duration_minutes']} min | Players: {game['setup']['num_players']}")
            print(f"  Rules: {len(game['rules'])} rules")
            print(f"  Tasks: {len(game['tasks'])} tasks")
            print(f"  Safety: Adult supervision = {game['safety']['adult_supervision']}")
            print(f"  Story tone: {game['story_seed']['tone']}")
            
            # Show first 2 tasks
            print(f"\n  First 2 Tasks:")
            for task in game['tasks'][:2]:
                print(f"    • {task['task_id']}: {task['title']}")
                print(f"      Points: {task['points']} | Proof: {task['proof_type']} | Time: {task['time_limit_minutes']} min")
            
            results.append({
                'config_name': test['name'],
                'valid': is_valid,
                'game_id': game['game_id'],
                'error_count': len(errors)
            })
            
        except Exception as e:
            print(f"✗ Generation failed: {e}")
            results.append({
                'config_name': test['name'],
                'valid': False,
                'game_id': None,
                'error_count': 1
            })
    
    # Summary
    print("\n" + "=" * 80)
    print("TEST SUMMARY")
    print("=" * 80)
    
    for result in results:
        status = "✓ PASS" if result['valid'] else "✗ FAIL"
        print(f"{status}: {result['config_name']}")
        if result['game_id']:
            print(f"    Generated: {result['game_id']}")
        if result['error_count'] > 0:
            print(f"    Errors: {result['error_count']}")
    
    total_pass = sum(1 for r in results if r['valid'])
    print(f"\nTotal: {total_pass}/{len(results)} tests passed")
    
    print("\n" + "=" * 80)
    print("GENERATION READY FOR PHASE 2 TASK 7: VALIDATION")
    print("=" * 80)


if __name__ == "__main__":
    main()