Update app.py

app.py

@@ -436,7 +436,7 @@ class ContentAnalyzer:
                 time.sleep(self.retry_delay * (2 ** attempt))
 
     def _create_analysis_prompt(self, transcript: str) -> str:
-        """Create the analysis prompt with more balanced evaluation criteria"""
+        """Create the analysis prompt with balanced evaluation criteria"""
         # First try to extract existing timestamps
         timestamps = re.findall(r'\[(\d{2}:\d{2})\]', transcript)
         
@@ -462,7 +462,9 @@ Example: If a quote starts at word 300, timestamp would be [02:00] (300 words /
                 marked_transcript += word + " "
             transcript = marked_transcript
 
-        prompt_template = """Analyze this teaching content with balanced evaluation criteria. Score 1 if most key requirements are met.
+        prompt_template = """Analyze this teaching content objectively. Each criterion should be evaluated independently with reasonable standards - neither too strict nor too lenient.
+
+Score 1 if the criterion meets reasonable standards, 0 if it does not.
 
 Transcript:
 {transcript}
@@ -474,110 +476,92 @@ Required JSON structure:
 {{
     "Concept Assessment": {{
         "Subject Matter Accuracy": {{
-            "Score": 1,  # Score 1 if technical information is generally accurate with minor errors acceptable
+            "Score": 0 or 1,  # Score 1 if content is generally accurate with only minor errors
             "Citations": ["[MM:SS] Quote demonstrating accuracy or error"]
         }},
         "First Principles Approach": {{
-            "Score": 1,  # Score 1 if most concepts are built from fundamentals with clear progression
-            "Citations": ["[MM:SS] Quote showing fundamental concept explanation"]
+            "Score": 0 or 1,  # Score 1 if key concepts are built from fundamentals
+            "Citations": ["[MM:SS] Quote showing concept explanation"]
         }},
         "Examples and Business Context": {{
-            "Score": 1,  # Score 1 if at least 2 relevant real-world examples are provided
+            "Score": 0 or 1,  # Score 1 if at least 2 relevant examples are provided
             "Citations": ["[MM:SS] Quote containing practical example"]
         }},
         "Cohesive Storytelling": {{
-            "Score": 1,  # Score 1 if most transitions are smooth with clear connections between topics
-            "Citations": ["[MM:SS] Quote showing topic transition or connection"]
+            "Score": 0 or 1,  # Score 1 if content flows logically most of the time
+            "Citations": ["[MM:SS] Quote showing topic transition"]
         }},
         "Engagement and Interaction": {{
-            "Score": 1,  # Score 1 if at least 1 engagement technique is used effectively
+            "Score": 0 or 1,  # Score 1 if there's meaningful audience engagement
             "Citations": ["[MM:SS] Quote showing audience engagement"]
         }},
         "Professional Tone": {{
-            "Score": 1,  # Score 1 if language is generally professional with occasional casual expressions acceptable
+            "Score": 0 or 1,  # Score 1 if tone is generally professional
             "Citations": ["[MM:SS] Quote demonstrating tone"]
         }}
     }},
     "Code Assessment": {{
         "Depth of Explanation": {{
-            "Score": 1,  # Score 1 if most code concepts include what and why explanations
+            "Score": 0 or 1,  # Score 1 if code concepts are explained clearly
             "Citations": ["[MM:SS] Quote showing code explanation"]
         }},
         "Output Interpretation": {{
-            "Score": 1,  # Score 1 if most code outputs are explained with their significance
+            "Score": 0 or 1,  # Score 1 if important outputs are explained
             "Citations": ["[MM:SS] Quote demonstrating output explanation"]
         }},
         "Breaking down Complexity": {{
-            "Score": 1,  # Score 1 if complex concepts are generally broken into understandable components
+            "Score": 0 or 1,  # Score 1 if complex concepts are made understandable
             "Citations": ["[MM:SS] Quote showing concept breakdown"]
         }}
     }}
 }}
 
-Balanced Evaluation Criteria:
-
-Concept Assessment:
-1. Subject Matter Accuracy:
-   - Generally accurate technical information
-   - Minor errors acceptable if core concepts are correct
-   - Clear explanation of key technical terms
-   - Comprehensive coverage of main concepts
-
-2. First Principles Approach:
-   - Most concepts built from fundamental principles
-   - Clear progression in concept difficulty
-   - Basic concepts explained before advanced ones
-   - Logical sequence of topics
-
-3. Examples and Business Context:
-   - At least 2 relevant real-world examples
-   - Examples clearly related to concepts
-   - Business value explained
-   - Practical applications discussed
-
-4. Cohesive Storytelling:
-   - Most topics connected logically
-   - Generally clear narrative flow
-   - Smooth transitions between main topics
-   - Consistent overall structure
-
-5. Engagement and Interaction:
-   - At least 1 engagement technique used
-   - Some audience involvement
-   - Occasional checks for understanding
-   - Varied teaching approach
-
-6. Professional Tone:
-   - Generally professional language
-   - Occasional casual expressions acceptable
-   - Clear delivery
-   - Appropriate level of formality
-
-Code Assessment:
-1. Depth of Explanation:
-   - Most code concepts explain what and why
-   - Key implementation details covered
-   - Important design decisions explained
-   - Best practices mentioned
-
-2. Output Interpretation:
-   - Most code outputs explained
-   - Key error cases covered
-   - Expected results discussed
-   - Basic validation covered
-
-3. Breaking down Complexity:
-   - Complex concepts divided into components
-   - Step-by-step explanation of key concepts
-   - Logical organization
-   - Dependencies explained
-
-Important:
-- Score 1 if most criteria in a category are met
-- Minor gaps or imperfections acceptable
-- Each citation must include timestamp and relevant quote
-- Citations should demonstrate how criteria are met
-- Be balanced in scoring - perfection not required"""
+Balanced Scoring Guidelines:
+
+Subject Matter Accuracy:
+✓ Pass: Content is technically sound with occasional minor errors
+✗ Fail: Multiple significant errors or fundamental misunderstandings
+
+First Principles Approach:
+✓ Pass: Core concepts are explained from basic principles
+✗ Fail: Advanced concepts introduced without basic foundation
+
+Examples and Business Context:
+✓ Pass: At least 2 relevant examples that illustrate concepts
+✗ Fail: Examples missing or irrelevant to the topic
+
+Cohesive Storytelling:
+✓ Pass: Topics generally flow well with clear connections
+✗ Fail: Frequent jumps between unrelated topics
+
+Engagement and Interaction:
+✓ Pass: Some effective engagement with audience
+✗ Fail: One-way lecture with no audience consideration
+
+Professional Tone:
+✓ Pass: Generally professional with occasional casual moments
+✗ Fail: Consistently unprofessional or inappropriate
+
+Depth of Explanation:
+✓ Pass: Key code concepts explained with reasonable detail
+✗ Fail: Code presented without meaningful explanation
+
+Output Interpretation:
+✓ Pass: Important outputs and their significance explained
+✗ Fail: Outputs shown without context or explanation
+
+Breaking down Complexity:
+✓ Pass: Complex topics broken into understandable parts
+✗ Fail: Complex concepts left unexplained
+
+Important Notes:
+- Evaluate each criterion independently
+- Perfect delivery is not required for a passing score
+- Look for evidence of competent teaching rather than flawless execution
+- Consider the overall effectiveness for the target audience
+- One or two minor issues should not result in failure
+- Citations must support the scoring decision
+- Different criteria can and should receive different scores based on their individual merits"""
 
         return prompt_template.format(
             transcript=transcript,