Prompt 1: DiscoveryI have two sets of business descriptions from microenterprises. Group A and Group B represent businesses that differ on some characteristics. GROUP A: {group_a_text} GROUP B: {group_b_text} Analyze these descriptions and identify systematic differences between the two groups. What patterns distinguish Group A from Group B? Return your response as a JSON array where each element represents a distinct hypothesis about how the groups differ: [ { "hypothesis": "Clear comparative statement describing how Group A differs from Group B on a specific dimension (e.g., 'Group A... while Group B...')", "dimension": "Name of the dimension being compared" }, ... ] Each hypothesis must explicitly compare both groups. Focus on concrete, observable characteristics that could be reliably coded by researchers. Return only valid JSON with no additional text.

Prompt 2: ConsolidationI have generated multiple hypotheses about what distinguishes Group A from Group B businesses. Please consolidate these into a clear, non-redundant list of measurable features. HYPOTHESES TO CONSOLIDATE: {hypotheses_text} Please: 1. Group similar hypotheses together 2. Create clear, operational definitions for each distinct feature 3. Eliminate redundancies and overlaps 4. Produce multiple distinct, measurable dimensions Return your response as a JSON array where each element represents a consolidated feature: [ { "feature_name": "Clear, concise name for the feature", "definition": "Operational definition of what this feature measures", "group_difference": "How Group A typically differs from Group B on this dimension", "source_hypotheses": ["List of hypothesis dimensions that contributed to this feature"] }, ... ] Return only valid JSON with no additional text.

Prompt 4: Final ConsolidationI have collected features from multiple discovery iterations that may express similar concepts using different definitions. Please group semantically similar features together and create standardized definitions. ALL FEATURES ACROSS ITERATIONS: {features_text} Please: 1. Group features based on their definitions and group differences, not on any similarity in wording 2. Features should only be grouped if they measure the same underlying concept 3. Count how many original features belong to each group 4. Preserve the Group A vs Group B pattern information from the original features 5. Only group features that are genuinely similar - don't force unrelated features together 6. Ensure each feature represents one distinct dimension only - do not combine multiple conceptually separate dimensions into a single feature 7. If a feature seems to combine multiple dimensions, split it into separate features for each distinct concept 8. After grouping based on conceptual similarity, assign an appropriate standardized name to each group Each feature should measure only one clear, distinct concept. Avoid combining unrelated dimensions into compound features. Return your response as a JSON array where each element represents a grouped feature: [ { "feature_name": "Standardized name assigned after conceptual grouping", "definition": "Clear operational definition of this single dimension", "group_difference": "How Group A typically differs from Group B on this dimension", "count": number_of_original_features_grouped }, ... ] Return only valid JSON with no additional text.