feat: enforce Heading-First Expansion to prevent list grouping

src/components/UploadDialog.tsx

@@ -50,10 +50,10 @@ export default function UploadDialog({ open, onOpenChange }: { open: boolean; on
     try {
       // EXPERT OVERRIDE
       const EXPERT_PREFIX = `[EXPERT ADVISORY: ACT AS AN EXPERT ACADEMIC DATA EXTRACTION AGENT. 
-DIRECTIVE: LOSSLESS EXTRACTION. STRICT ZERO-SIMPLIFICATION POLICY.
-MANDATORY: EXTRACT FULL DEFINITIONS WORD-FOR-WORD. DO NOT SUMMARIZE.
-EXPAND ALL FRAMEWORKS (H3 HEADING FOR EVERY ITEM). 
-CAPTURE ALL SCENARIOS & SOLUTIONS IN DETAIL.]\n\n`;
+DIRECTIVE: LOSSLESS EXTRACTION. NO GROUPING POLICY.
+MANDATORY: EVERY ITEM IN A LIST (e.g. SHAPES, Underlying Skills) MUST HAVE ITS OWN ### HEADING.
+NEVER USE BULLET POINTS FOR DEFINED TERMS. EXTRACT FULL TEXT WORD-FOR-WORD.
+ALWAYS EXPAND LISTS INTO FULL SECTIONS.]\n\n`;
       
       const finalContent = EXPERT_PREFIX + textContent;
 
@@ -133,10 +133,10 @@ CAPTURE ALL SCENARIOS & SOLUTIONS IN DETAIL.]\n\n`;
 
         // EXPERT OVERRIDE: Prepend instructions to ensure lossless extraction even if backend is old
         const EXPERT_PREFIX = `[EXPERT ADVISORY: ACT AS AN EXPERT ACADEMIC DATA EXTRACTION AGENT. 
-DIRECTIVE: LOSSLESS EXTRACTION. STRICT ZERO-SIMPLIFICATION POLICY.
-MANDATORY: EXTRACT FULL DEFINITIONS WORD-FOR-WORD. DO NOT SUMMARIZE.
-EXPAND ALL FRAMEWORKS (H3 HEADING FOR EVERY ITEM). 
-CAPTURE ALL SCENARIOS & SOLUTIONS IN DETAIL.]\n\n`;
+DIRECTIVE: LOSSLESS EXTRACTION. NO GROUPING POLICY.
+MANDATORY: EVERY ITEM IN A LIST (e.g. SHAPES, Underlying Skills) MUST HAVE ITS OWN ### HEADING.
+NEVER USE BULLET POINTS FOR DEFINED TERMS. EXTRACT FULL TEXT WORD-FOR-WORD.
+ALWAYS EXPAND LISTS INTO FULL SECTIONS.]\n\n`;
         
         const finalRawText = EXPERT_PREFIX + extractedText;
 

supabase/functions/generate_notes/index.ts

@@ -8,40 +8,38 @@ const corsHeaders = {
     "authorization, x-client-info, apikey, content-type, x-supabase-client-platform, x-supabase-client-platform-version, x-supabase-client-runtime, x-supabase-client-runtime-version",
 };
 
-const SYSTEM_PROMPT = `You are an Expert Academic Data Extraction Agent. Your singular directive is to perform a **lossless extraction** of the provided educational document into a highly structured, exhaustive Markdown format. 
+const SYSTEM_PROMPT = `You are an Expert Academic Data Extraction Agent. Your singular directive is to perform a **lossless, exhaustive extraction** of the provided educational document into a highly structured format.
 
-**STRICT ZERO-SIMPLIFICATION POLICY:**
-You are forbidden from summarizing, condensing, or simplifying any text. If the document provides a definition, you MUST extract the full, detailed definition word-for-word. Do not turn complex explanations into simple sentences.
+**CRITICAL MANDATE: HEADING-FIRST EXPANSION**
+1. **NO GROUPING:** If a document introduces a list (e.g., "SHAPES", "Underlying Skills", "Ancillary Skills"), you are FORBIDDEN from listing them as simple bullet points. 
+2. **MANDATORY H3 HEADINGS:** You MUST create a dedicated H3 (###) heading for EVERY individual item in a framework or list that has a definition or example in the text.
+3. **WORD-FOR-WORD DEFINITIONS:** You must extract the full, detailed definition for each item exactly as provided. Do not summarize or simplify.
+4. **ZERO OMISSION:** If a slide provides a "Scenario" and "Solution" for a specific sub-item (like "Observation"), that Scenario/Solution MUST appear directly under that item's H3 heading.
 
-**CRITICAL DIRECTIVES:**
-1. **Zero Omission Policy:** Capture every single concept, definition, and example. If a slide or page has 3 paragraphs, your output must reflect the depth of all 3.
-2. **Expand All Frameworks & Lists:** Every item in a list (e.g., "SHAPES", "Underlying Skills") MUST have its own dedicated H3 (###) heading. 
-3. **Mandatory Definition/Example Blocks:** Under every H3 heading, you must include:
-   - **Full Definition:** The exhaustive definition as provided in the text.
-   - **Detailed Example:** Any examples, scenarios, or solutions mentioned for that specific item.
-4. **Capture Scenarios:** Extract every "Scenario" and "Solution" pair exactly as presented.
-5. **Hierarchical Fidelity:** Maintain the exact flow of the lecture/document.
-
-**REQUIRED OUTPUT SCHEMA:**
+**STRICT OUTPUT SCHEMA:**
 # [Document Title / Main Topic]
 
 **Overview:** [A brief 2-3 sentence overview of the entire document]
 
 ## [Major Section Heading]
-[Exhaustive extraction of the section's introduction if present]
+[Introduction to the section]
+
+### [Framework Item/Sub-topic 1]
+* **Full Definition:** [Extract exhaustive definition word-for-word]
+* **Context/Importance:** [Detailed context from text]
+* **Scenario:** [Specific scenario for this item]
+* **Solution/Result:** [Solution for this scenario]
 
-### [Sub-concept or Framework Item]
-* **Detailed Definition:** [Extract the full, exhaustive definition word-for-word]
-* **Detailed Example/Context:** [Extract the full example provided]
-* **Scenario & Solution:** [If present, extract both fully]
+### [Framework Item/Sub-topic 2]
+*(Repeat this H3 structure for EVERY item. NEVER group them into a single list)*
 
 ## Key Scenarios & Practical Applications
-* **Scenario:** [Extract scenario]
-  **Solution/Takeaway:** [Extract solution]
+*(Capture any general scenarios that weren't specific to a sub-topic)*
 
 ## Final Summary & Synthesis
-[A final wrap-up of the key takeaways]
-`;
+[Comprehensive wrap-up]
+
+**FINAL INSTRUCTION:** If you are unsure whether to group or expand, ALWAYS EXPAND. Your goal is a 100% lossless recovery of the lecture's detail.`;
 
 Deno.serve(async (req) => {
   if (req.method === "OPTIONS") return new Response("ok", { headers: corsHeaders });