# semantic_library/narrative_strategies.yaml # Version: 2.0 # Description: Defines strategies for analyzing data and generating different types of narratives. strategies: - strategy_id: STRAT_COMPANY_FINANCIAL_IMPACT_OVERVIEW # Renamed and enhanced description: "Provides an overview of a company, identifies key active drivers, and assesses their probabilistic first-order financial impacts." target_audience: "Investor / Analyst needing quantitative insights" information_gathering_plan: # Steps for ReasoningEngine - step: "GET_COMPANY_DETAILS" source: "knowledge_graph" details: "Fetch company node properties (name, industry, financials, etc.)." - step: "IDENTIFY_ACTIVE_DRIVERS" source: "knowledge_graph_and_driver_detection_patterns" # Uses drivers_knowledge_base.yaml details: "Determine active drivers for the company based on detection_patterns and KG links. Prioritize top 3-5 by potential significance." - step: "CALCULATE_PROBABILISTIC_IMPACTS" source: "drivers_knowledge_base.impact_model" details: "For each active driver, calculate its first_order_impacts on relevant target_variables, considering probabilities and effect_functions." analytical_steps: # Further analysis on gathered/calculated data - step: "AGGREGATE_IMPACTS" details: "Summarize key positive and negative potential impacts across drivers. Identify potential conflicting or compounding effects (conceptual for now)." - step: "ASSESS_OVERALL_SENTIMENT_AND_RISK_BALANCE" details: "Qualitatively assess if the balance of impacts points to a net positive, negative, or neutral outlook based on probabilities and magnitudes." narrative_flow: # High-level structure for the LLM output - section: "EXECUTIVE_SUMMARY" llm_instructions: "Start with a 2-3 sentence executive summary covering the company, overall sentiment from analysis, and mention of 1-2 most critical drivers." - section: "COMPANY_CONTEXT" llm_instructions: "Briefly describe the company, its industry, and current key financial metrics (e.g., Revenue, P/E) as context." - section: "KEY_ACTIVE_DRIVERS_IDENTIFIED" llm_instructions: "List the key active drivers identified (e.g., top 3-5). For each, briefly state its nature using 'narrative_logic.key_insights_to_extract' from drivers_knowledge_base.yaml. This section should *not* detail the calculated impacts yet, just introduce the drivers." - section: "QUANTITATIVE_IMPACT_ANALYSIS" # New or refined section name llm_instructions: - "This section presents the calculated probabilistic first-order impacts of the key active drivers." - "For each key active driver (or for each significant calculated impact):" - " State the driver name (and ID)." - " Describe the specific impact: 'Impacts {target_variable} by an estimated {calculated_impact_value} (e.g., a change of X units, or new value of Y).'" - " State the likelihood: 'This impact has an assessed probability of {probability_of_occurrence*100}%.'" - " Specify the timeframe: 'Expected over {time_horizon}.'" - " Mention any key conditions: 'This assessment assumes {conditions_evaluated}.'" - " Use probabilistic_language_rules to phrase statements about uncertainty naturally within the description." - " Incorporate relevant 'explanation_patterns_llm' from the driver's 'narrative_logic' to provide context for the numbers." - " Present this information clearly, perhaps using bullet points for multiple impacts from a single driver, or sub-sections per driver." - " Focus on the *calculated* impacts provided by the ReasoningEngine." - section: "SYNTHESIS_AND_OUTLOOK" llm_instructions: - "Synthesize the combined potential effects of these drivers." - "Discuss any significant second or third-order effects if they were identified and are material (from driver's impact_model)." - "Provide a balanced overall outlook, explicitly mentioning key uncertainties or dependencies." llm_persona_and_tone: persona: "Objective Financial Analyst" tone: "Neutral, data-driven, cautious but insightful." complexity: "Moderate (assume familiarity with basic financial terms)" probabilistic_language_rules: # Guidance for LLM on expressing uncertainty - condition: "probability >= 0.75" phrase_template: "is highly likely to result in {impact_description}" - condition: "probability >= 0.50 && probability < 0.75" phrase_template: "is likely to cause {impact_description}" - condition: "probability >= 0.25 && probability < 0.50" phrase_template: "could potentially lead to {impact_description}" - condition: "probability < 0.25" phrase_template: "there is a slight chance of {impact_description}" overall_llm_instructions: - "The final narrative should be coherent, well-structured according to the narrative_flow, and provide actionable insights." - "Clearly differentiate between observed data, model-driven probabilistic impacts, and qualitative assessments." - "Focus on the 'why' behind the numbers and the 'so what' for the company." - "Total length should be appropriate for an analytical summary (e.g., 4-6 paragraphs or equivalent bulleted sections)." - strategy_id: STRAT_RISK_DEEP_DIVE # Formerly TPL_RISK_FOCUS_SUMMARY description: "A detailed analysis focusing on key risks, their probabilistic impacts, and potential mitigating factors for a company." target_audience: "Risk Manager / Due Diligence Analyst" information_gathering_plan: - step: "GET_COMPANY_DETAILS" - step: "IDENTIFY_ACTIVE_RISK_DRIVERS" details: "Identify drivers with 'risk' tag or predominantly negative 'impact_model' outcomes. Prioritize by (impact_magnitude * probability)." - step: "CALCULATE_RISK_IMPACTS" details: "Quantify probabilistic first-order impacts of these risk drivers." - step: "IDENTIFY_MITIGATING_FACTORS_OR_OPPORTUNITIES" details: "Look for drivers that counteract identified risks, or company strengths/strategies that mitigate them." analytical_steps: - step: "ASSESS_NET_RISK_EXPOSURE" details: "Evaluate the company's overall vulnerability considering both risks and mitigants." narrative_flow: - section: "RISK_SUMMARY_STATEMENT" - section: "DETAILED_RISK_DRIVER_ANALYSIS" # For each key risk driver llm_instructions: - "Describe the risk driver." - "Detail its probabilistic negative impacts (magnitude, likelihood, timeframe)." - "Mention data points supporting this risk assessment." - section: "MITIGATING_FACTORS_AND_CONTROLS" llm_instructions: "Discuss any company actions, strengths, or positive counter-drivers that may mitigate these risks. Assess effectiveness if possible." - section: "OVERALL_RISK_PROFILE_AND_RECOMMENDATIONS" # Optional recommendations llm_persona_and_tone: persona: "Prudent Risk Analyst" tone: "Objective, critical, focused on downside." probabilistic_language_rules: # (Can inherit from a global set or be strategy-specific) - condition: "probability >= 0.70" phrase_template: "poses a significant threat of {impact_description}" - condition: "probability >= 0.40 && probability < 0.70" phrase_template: "presents a moderate risk of {impact_description}" - condition: "probability < 0.40" phrase_template: "there is a potential, though less likely, risk of {impact_description}" overall_llm_instructions: - "Clearly articulate each significant risk and its potential consequences." - "Be specific about the sources and nature of uncertainty." - strategy_id: STRAT_COMPARATIVE_ANALYSIS_BASIC description: "A basic comparative analysis of two companies on key metrics and drivers." target_audience: "Analyst / Investor" # This strategy would require the ReasoningEngine to handle context for two companies. # The `information_gathering_plan` and `analytical_steps` would need to be adapted. # The LLM prompt would need to be structured to compare and contrast. information_gathering_plan: - step: "GET_COMPANY_DETAILS_COMPANY_A" # parameters: { company_id: "{company_A_id}" } # Passed in at runtime details: "Fetch details for Company A." - step: "GET_COMPANY_DETAILS_COMPANY_B" # parameters: { company_id: "{company_B_id}" } details: "Fetch details for Company B." - step: "IDENTIFY_ACTIVE_DRIVERS_COMPANY_A" - step: "IDENTIFY_ACTIVE_DRIVERS_COMPANY_B" - step: "CALCULATE_PROBABILISTIC_IMPACTS_COMPANY_A" - step: "CALCULATE_PROBABILISTIC_IMPACTS_COMPANY_B" analytical_steps: - step: "COMPARE_KEY_FINANCIALS" details: "Compare P/E, revenue growth, margins etc. for Company A vs Company B." - step: "COMPARE_COMMON_DRIVERS_IMPACT" details: "Identify common active drivers and compare their calculated impacts on both companies." - step: "IDENTIFY_UNIQUE_DRIVERS_IMPACT" details: "Highlight drivers uniquely impacting one company but not the other and their significance." - step: "ASSESS_RELATIVE_STRENGTHS_WEAKNESSES" details: "Based on financials and driver impacts, provide a qualitative assessment." narrative_flow: - section: "INTRODUCTION_COMPARISON_PAIR" llm_instructions: "Introduce Company A ({company_A_name}) and Company B ({company_B_name}) and the purpose of this comparison." - section: "FINANCIAL_SNAPSHOT_COMPARISON" llm_instructions: - "Present a table or comparative statements of key financials (e.g., P/E, Revenue, Growth Rate, Debt/Equity)." - "Highlight key differences." - section: "SHARED_DRIVERS_AND_DIFFERENTIAL_IMPACT" llm_instructions: - "Discuss key drivers affecting both companies (e.g., {shared_driver_1.name})." - "Explain how the *calculated probabilistic impact* of these shared drivers might differ for each company and why (e.g., due to different sensitivities, market positions defined in their respective contexts)." - "Example: 'While both companies are affected by DRV001, its impact on Company A's P/E is calculated at X% (Prob: Y%), whereas for Company B it's Z% (Prob: W%). This difference is likely due to...'" - section: "COMPANY_SPECIFIC_DRIVERS_AND_IMPACTS" llm_instructions: - "Highlight significant drivers and their calculated impacts that are unique or more pronounced for Company A." - "Do the same for Company B." - section: "COMPARATIVE_OUTLOOK_AND_CONCLUSION" llm_instructions: "Based on the comparative analysis of financials and driver impacts, provide a summary outlook for each company relative to the other. Which company appears better positioned given current drivers? What are key differentiating factors?" llm_persona_and_tone: persona: "Comparative Analyst" tone: "Objective, analytical, highlighting contrasts." probabilistic_language_rules: # Can inherit or be specific - condition: "probability >= 0.70" phrase_template: "is very likely to experience {impact_description}" - condition: "probability >= 0.50 && probability < 0.70" phrase_template: "will probably see {impact_description}" - condition: "probability < 0.50" phrase_template: "may face {impact_description}" overall_llm_instructions: - "Focus on clear comparisons and contrasts." - "Use data and calculated impacts to support assertions." - "Conclude with a relative positioning statement." ...