diff --git "a/data/07_cf/train.jsonl" "b/data/07_cf/train.jsonl" new file mode 100644--- /dev/null +++ "b/data/07_cf/train.jsonl" @@ -0,0 +1,30 @@ +{"task_id": "CF-CN_A-P01a", "category": "CF", "variant": "paired_counterfactual_v2", "time_band": "T1", "status": "ready", "cutoff_date": "2024-03-29", "seed": {"stock_code": "600519", "stock_name": "贵州茅台", "historical_price_series": [1695.4301, 1670.0, 1717.66, 1718.38, 1714.09, 1694.98, 1689.5, 1681.55, 1695.0, 1685.0601, 1676.3, 1695.1, 1680.55, 1683.63, 1671.4301, 1694.49, 1745.0, 1727.45, 1716.63, 1717.49, 1708.26, 1708.02, 1709.6, 1709.0, 1701.35, 1689.01, 1709.29, 1701.0, 1701.64, 1702.9], "historical_trend_direction": "flat", "counterfactual_news": "公司披露本季度净利润同比下降15%，而公告前市场一致预期为下降30%；收入同比增长4%，管理层维持全年指引不变。", "expected_logic_direction": "positive", "driver_options": ["expectation_gap", "margin_surprise", "demand_surprise", "regulatory_severity", "unit_economics", "financing_cost", "dilution", "profit_contribution", "relationship_type", "relative_cost_hedge", "compliance_readiness", "supply_continuity", "deleveraging_vs_dilution", "retained_economics", "refinancing_spread"], "benchmark_id": "沪深300", "cutoff_date": "2024-03-29", "market": "CN_A"}, "prompt": "以下是 贵州茅台 过去 30 个交易日的收盘价序列（从旧到新）：\n[1695.4301, 1670.0, 1717.66, 1718.38, 1714.09, 1694.98, 1689.5, 1681.55, 1695.0, 1685.0601, 1676.3, 1695.1, 1680.55, 1683.63, 1671.4301, 1694.49, 1745.0, 1727.45, 1716.63, 1717.49, 1708.26, 1708.02, 1709.6, 1709.0, 1701.35, 1689.01, 1709.29, 1701.0, 1701.64, 1702.9]\n\n相对收益基准为：沪深300。\n\n以下是一则完全虚构的公告（仅用于逻辑推理测试，公告中的数字和关系优先于公司真实历史）：\n「公司披露本季度净利润同比下降15%，而公告前市场一致预期为下降30%；收入同比增长4%，管理层维持全年指引不变。」\n\n假设上述事件真实发生，且历史上从未发生过类似事件。\n请比较公告中相互冲突的信息，识别真正决定短期定价的因素，预测该股票未来 5 个交易日相对基准的超额收益方向。\n不要仅根据“增长、获批、回购、处罚、召回”等单个关键词作答。\n同时从以下候选中选择唯一的决定性驱动因素代码：\nexpectation_gap / margin_surprise / demand_surprise / regulatory_severity / unit_economics / financing_cost / dilution / profit_contribution / relationship_type / relative_cost_hedge / compliance_readiness / supply_continuity / deleveraging_vs_dilution / retained_economics / refinancing_spread\n\n以下面 JSON 格式输出，不要附加任何其他内容：\n{\n \"direction\": \"positive\" 或 \"negative\",\n \"confidence\": <0.50 到 1.00 之间的浮点数>,\n \"driver\": \"<候选驱动因素代码>\",\n \"reasoning\": \"<一句话，说明核心判断依据>\"\n}\n\n请严格输出 JSON，不要输出任何额外文字、解释或 markdown 代码块。\n", "expected_output": {"direction": "positive|negative", "confidence": "float[0.50,1.00]", "driver": "enum", "reasoning": "str"}, "ground_truth": {"logic_direction": "positive", "driver_id": "expectation_gap", "decisive_factor": "实际业绩相对一致预期的偏差"}, "metadata": {"is_template": false, "prompt_template": "prompts/d_counterfactual.txt", "schema_version": "d_counterfactual_v2", "pair_id": "P01", "pair_side": "a", "family": "expectation_gap", "driver_id": "expectation_gap", "trend_conflict": false, "generation_source": "configs/d_event_archetypes_v2.json + configs/d_counterfactual_specs_v2.csv", "market": "CN_A", "temporal_split": {"time_band": "T1", "cutoff_date": "2024-03-29", "cutoff_date_source": "seed.cutoff_date", "is_synthetic_cutoff_date": false, "model_training_cutoff": "2024-06-30", "reference_current_date": "2026-08-17", "rule": "T1: cutoff_date <= model_training_cutoff; T2: model_training_cutoff < cutoff_date < reference_current_date; T3: cutoff_date >= reference_current_date"}}, "forecast_origin": "2024-03-29", "forecast_origin_source": "seed.cutoff_date", "outcome_available_at": "2024-03-29", "outcome_available_at_source": "synthetic_same_as_origin", "outcome_evidence_url": null, "outcome_evidence_code": "counterfactual_logic", "paper_band": "CF", "paper_band_reason": "category_D_counterfactual", "review_status": "draft", "review_method": null, "reviewed_at": null, "evidence_hash": null, "quality_flags": [], "official_temporal_eligible": true, "backbone_training_cutoff": "2024-06-30", "guard_days": 30, "experiment_as_of": "2026-08-17"} +{"task_id": "CF-CN_A-P01b", "category": "CF", "variant": "paired_counterfactual_v2", "time_band": "T1", "status": "ready", "cutoff_date": "2024-03-29", "seed": {"stock_code": "600519", "stock_name": "贵州茅台", "historical_price_series": [1695.4301, 1670.0, 1717.66, 1718.38, 1714.09, 1694.98, 1689.5, 1681.55, 1695.0, 1685.0601, 1676.3, 1695.1, 1680.55, 1683.63, 1671.4301, 1694.49, 1745.0, 1727.45, 1716.63, 1717.49, 1708.26, 1708.02, 1709.6, 1709.0, 1701.35, 1689.01, 1709.29, 1701.0, 1701.64, 1702.9], "historical_trend_direction": "flat", "counterfactual_news": "公司披露本季度净利润同比下降15%，而公告前市场一致预期为下降5%；收入同比增长4%，管理层维持全年指引不变。", "expected_logic_direction": "negative", "driver_options": ["expectation_gap", "margin_surprise", "demand_surprise", "regulatory_severity", "unit_economics", "financing_cost", "dilution", "profit_contribution", "relationship_type", "relative_cost_hedge", "compliance_readiness", "supply_continuity", "deleveraging_vs_dilution", "retained_economics", "refinancing_spread"], "benchmark_id": "沪深300", "cutoff_date": "2024-03-29", "market": "CN_A"}, "prompt": "以下是 贵州茅台 过去 30 个交易日的收盘价序列（从旧到新）：\n[1695.4301, 1670.0, 1717.66, 1718.38, 1714.09, 1694.98, 1689.5, 1681.55, 1695.0, 1685.0601, 1676.3, 1695.1, 1680.55, 1683.63, 1671.4301, 1694.49, 1745.0, 1727.45, 1716.63, 1717.49, 1708.26, 1708.02, 1709.6, 1709.0, 1701.35, 1689.01, 1709.29, 1701.0, 1701.64, 1702.9]\n\n相对收益基准为：沪深300。\n\n以下是一则完全虚构的公告（仅用于逻辑推理测试，公告中的数字和关系优先于公司真实历史）：\n「公司披露本季度净利润同比下降15%，而公告前市场一致预期为下降5%；收入同比增长4%，管理层维持全年指引不变。」\n\n假设上述事件真实发生，且历史上从未发生过类似事件。\n请比较公告中相互冲突的信息，识别真正决定短期定价的因素，预测该股票未来 5 个交易日相对基准的超额收益方向。\n不要仅根据“增长、获批、回购、处罚、召回”等单个关键词作答。\n同时从以下候选中选择唯一的决定性驱动因素代码：\nexpectation_gap / margin_surprise / demand_surprise / regulatory_severity / unit_economics / financing_cost / dilution / profit_contribution / relationship_type / relative_cost_hedge / compliance_readiness / supply_continuity / deleveraging_vs_dilution / retained_economics / refinancing_spread\n\n以下面 JSON 格式输出，不要附加任何其他内容：\n{\n \"direction\": \"positive\" 或 \"negative\",\n \"confidence\": <0.50 到 1.00 之间的浮点数>,\n \"driver\": \"<候选驱动因素代码>\",\n \"reasoning\": \"<一句话，说明核心判断依据>\"\n}\n\n请严格输出 JSON，不要输出任何额外文字、解释或 markdown 代码块。\n", "expected_output": {"direction": "positive|negative", "confidence": "float[0.50,1.00]", "driver": "enum", "reasoning": "str"}, "ground_truth": {"logic_direction": "negative", "driver_id": "expectation_gap", "decisive_factor": "实际业绩相对一致预期的偏差"}, "metadata": {"is_template": false, "prompt_template": "prompts/d_counterfactual.txt", "schema_version": "d_counterfactual_v2", "pair_id": "P01", "pair_side": "b", "family": "expectation_gap", "driver_id": "expectation_gap", "trend_conflict": false, "generation_source": "configs/d_event_archetypes_v2.json + configs/d_counterfactual_specs_v2.csv", "market": "CN_A", "temporal_split": {"time_band": "T1", "cutoff_date": "2024-03-29", "cutoff_date_source": "seed.cutoff_date", "is_synthetic_cutoff_date": false, "model_training_cutoff": "2024-06-30", "reference_current_date": "2026-08-17", "rule": "T1: cutoff_date <= model_training_cutoff; T2: model_training_cutoff < cutoff_date < reference_current_date; T3: cutoff_date >= reference_current_date"}}, "forecast_origin": "2024-03-29", "forecast_origin_source": "seed.cutoff_date", "outcome_available_at": "2024-03-29", "outcome_available_at_source": "synthetic_same_as_origin", "outcome_evidence_url": null, "outcome_evidence_code": "counterfactual_logic", "paper_band": "CF", "paper_band_reason": "category_D_counterfactual", "review_status": "draft", "review_method": null, "reviewed_at": null, "evidence_hash": null, "quality_flags": [], "official_temporal_eligible": true, "backbone_training_cutoff": "2024-06-30", "guard_days": 30, "experiment_as_of": "2026-08-17"} +{"task_id": "CF-CN_A-P02a", "category": "CF", "variant": "paired_counterfactual_v2", "time_band": "T1", "status": "ready", "cutoff_date": "2024-03-29", "seed": {"stock_code": "601318", "stock_name": "中国平安", "historical_price_series": [42.73, 42.35, 43.59, 43.7, 44.07, 42.95, 43.12, 42.65, 42.9, 42.99, 42.34, 42.81, 42.5, 43.06, 42.88, 43.16, 43.81, 42.63, 42.45, 42.48, 42.56, 41.83, 41.92, 42.03, 40.56, 40.44, 40.9, 40.65, 40.69, 40.81], "historical_trend_direction": "negative", "counterfactual_news": "公司季度收入同比增长8%，略低于此前10%的指引；毛利率为34%，市场此前预期29%。公司去年收入增长7%、毛利率31%。", "expected_logic_direction": "positive", "driver_options": ["expectation_gap", "margin_surprise", "demand_surprise", "regulatory_severity", "unit_economics", "financing_cost", "dilution", "profit_contribution", "relationship_type", "relative_cost_hedge", "compliance_readiness", "supply_continuity", "deleveraging_vs_dilution", "retained_economics", "refinancing_spread"], "benchmark_id": "沪深300", "cutoff_date": "2024-03-29", "market": "CN_A"}, "prompt": "以下是 ��国平安 过去 30 个交易日的收盘价序列（从旧到新）：\n[42.73, 42.35, 43.59, 43.7, 44.07, 42.95, 43.12, 42.65, 42.9, 42.99, 42.34, 42.81, 42.5, 43.06, 42.88, 43.16, 43.81, 42.63, 42.45, 42.48, 42.56, 41.83, 41.92, 42.03, 40.56, 40.44, 40.9, 40.65, 40.69, 40.81]\n\n相对收益基准为：沪深300。\n\n以下是一则完全虚构的公告（仅用于逻辑推理测试，公告中的数字和关系优先于公司真实历史）：\n「公司季度收入同比增长8%，略低于此前10%的指引；毛利率为34%，市场此前预期29%。公司去年收入增长7%、毛利率31%。」\n\n假设上述事件真实发生，且历史上从未发生过类似事件。\n请比较公告中相互冲突的信息，识别真正决定短期定价的因素，预测该股票未来 5 个交易日相对基准的超额收益方向。\n不要仅根据“增长、获批、回购、处罚、召回”等单个关键词作答。\n同时从以下候选中选择唯一的决定性驱动因素代码：\nexpectation_gap / margin_surprise / demand_surprise / regulatory_severity / unit_economics / financing_cost / dilution / profit_contribution / relationship_type / relative_cost_hedge / compliance_readiness / supply_continuity / deleveraging_vs_dilution / retained_economics / refinancing_spread\n\n以下面 JSON 格式输出，不要附加任何其他内容：\n{\n \"direction\": \"positive\" 或 \"negative\",\n \"confidence\": <0.50 到 1.00 之间的浮点数>,\n \"driver\": \"<候选驱动因素代码>\",\n \"reasoning\": \"<一句话，说明核心判断依据>\"\n}\n\n请严格输出 JSON，不要输出任何额外文字、解释或 markdown 代码块。\n", "expected_output": {"direction": "positive|negative", "confidence": "float[0.50,1.00]", "driver": "enum", "reasoning": "str"}, "ground_truth": {"logic_direction": "positive", "driver_id": "margin_surprise", "decisive_factor": "毛利率相对预期及其利润贡献"}, "metadata": {"is_template": false, "prompt_template": "prompts/d_counterfactual.txt", "schema_version": "d_counterfactual_v2", "pair_id": "P02", "pair_side": "a", "family": "expectation_gap", "driver_id": "margin_surprise", "trend_conflict": true, "generation_source": "configs/d_event_archetypes_v2.json + configs/d_counterfactual_specs_v2.csv", "market": "CN_A", "temporal_split": {"time_band": "T1", "cutoff_date": "2024-03-29", "cutoff_date_source": "seed.cutoff_date", "is_synthetic_cutoff_date": false, "model_training_cutoff": "2024-06-30", "reference_current_date": "2026-08-17", "rule": "T1: cutoff_date <= model_training_cutoff; T2: model_training_cutoff < cutoff_date < reference_current_date; T3: cutoff_date >= reference_current_date"}}, "forecast_origin": "2024-03-29", "forecast_origin_source": "seed.cutoff_date", "outcome_available_at": "2024-03-29", "outcome_available_at_source": "synthetic_same_as_origin", "outcome_evidence_url": null, "outcome_evidence_code": "counterfactual_logic", "paper_band": "CF", "paper_band_reason": "category_D_counterfactual", "review_status": "draft", "review_method": null, "reviewed_at": null, "evidence_hash": null, "quality_flags": [], "official_temporal_eligible": true, "backbone_training_cutoff": "2024-06-30", "guard_days": 30, "experiment_as_of": "2026-08-17"} +{"task_id": "CF-CN_A-P02b", "category": "CF", "variant": "paired_counterfactual_v2", "time_band": "T1", "status": "ready", "cutoff_date": "2024-03-29", "seed": {"stock_code": "601318", "stock_name": "中国平安", "historical_price_series": [42.73, 42.35, 43.59, 43.7, 44.07, 42.95, 43.12, 42.65, 42.9, 42.99, 42.34, 42.81, 42.5, 43.06, 42.88, 43.16, 43.81, 42.63, 42.45, 42.48, 42.56, 41.83, 41.92, 42.03, 40.56, 40.44, 40.9, 40.65, 40.69, 40.81], "historical_trend_direction": "negative", "counterfactual_news": "公司季度收入同比增长8%，略低于此前10%的指引；毛利率为30%，市场此前预期34%。公司去年收入增长7%、毛利率31%。", "expected_logic_direction": "negative", "driver_options": ["expectation_gap", "margin_surprise", "demand_surprise", "regulatory_severity", "unit_economics", "financing_cost", "dilution", "profit_contribution", "relationship_type", "relative_cost_hedge", "compliance_readiness", "supply_continuity", "deleveraging_vs_dilution", "retained_economics", "refinancing_spread"], "benchmark_id": "沪深300", "cutoff_date": "2024-03-29", "market": "CN_A"}, "prompt": "以下是 中国平安 过去 30 个交易日的收盘价序列（从旧到新）：\n[42.73, 42.35, 43.59, 43.7, 44.07, 42.95, 43.12, 42.65, 42.9, 42.99, 42.34, 42.81, 42.5, 43.06, 42.88, 43.16, 43.81, 42.63, 42.45, 42.48, 42.56, 41.83, 41.92, 42.03, 40.56, 40.44, 40.9, 40.65, 40.69, 40.81]\n\n相对收益基准为：沪深300。\n\n以下是一则完全虚构的公告（仅用于逻辑推理测试，公告中的数字和关系优先于公司真实历史）：\n「公司季度收入同比增长8%，略低于此前10%的指引；毛利率为30%，市场此前预期34%。公司去年收入增长7%、毛利率31%。」\n\n假设上述事件真实发生，且历史上从未发生过类似事件。\n请��较公告中相互冲突的信息，识别真正决定短期定价的因素，预测该股票未来 5 个交易日相对基准的超额收益方向。\n不要仅根据“增长、获批、回购、处罚、召回”等单个关键词作答。\n同时从以下候选中选择唯一的决定性驱动因素代码：\nexpectation_gap / margin_surprise / demand_surprise / regulatory_severity / unit_economics / financing_cost / dilution / profit_contribution / relationship_type / relative_cost_hedge / compliance_readiness / supply_continuity / deleveraging_vs_dilution / retained_economics / refinancing_spread\n\n以下面 JSON 格式输出，不要附加任何其他内容：\n{\n \"direction\": \"positive\" 或 \"negative\",\n \"confidence\": <0.50 到 1.00 之间的浮点数>,\n \"driver\": \"<候选驱动因素代码>\",\n \"reasoning\": \"<一句话，说明核心判断依据>\"\n}\n\n请严格输出 JSON，不要输出任何额外文字、解释或 markdown 代码块。\n", "expected_output": {"direction": "positive|negative", "confidence": "float[0.50,1.00]", "driver": "enum", "reasoning": "str"}, "ground_truth": {"logic_direction": "negative", "driver_id": "margin_surprise", "decisive_factor": "毛利率相对预期及其利润贡献"}, "metadata": {"is_template": false, "prompt_template": "prompts/d_counterfactual.txt", "schema_version": "d_counterfactual_v2", "pair_id": "P02", "pair_side": "b", "family": "expectation_gap", "driver_id": "margin_surprise", "trend_conflict": false, "generation_source": "configs/d_event_archetypes_v2.json + configs/d_counterfactual_specs_v2.csv", "market": "CN_A", "temporal_split": {"time_band": "T1", "cutoff_date": "2024-03-29", "cutoff_date_source": "seed.cutoff_date", "is_synthetic_cutoff_date": false, "model_training_cutoff": "2024-06-30", "reference_current_date": "2026-08-17", "rule": "T1: cutoff_date <= model_training_cutoff; T2: model_training_cutoff < cutoff_date < reference_current_date; T3: cutoff_date >= reference_current_date"}}, "forecast_origin": "2024-03-29", "forecast_origin_source": "seed.cutoff_date", "outcome_available_at": "2024-03-29", "outcome_available_at_source": "synthetic_same_as_origin", "outcome_evidence_url": null, "outcome_evidence_code": "counterfactual_logic", "paper_band": "CF", "paper_band_reason": "category_D_counterfactual", "review_status": "draft", "review_method": null, "reviewed_at": null, "evidence_hash": null, "quality_flags": [], "official_temporal_eligible": true, "backbone_training_cutoff": "2024-06-30", "guard_days": 30, "experiment_as_of": "2026-08-17"} +{"task_id": "CF-CN_A-P03a", "category": "CF", "variant": "paired_counterfactual_v2", "time_band": "T1", "status": "ready", "cutoff_date": "2024-03-29", "seed": {"stock_code": "300750", "stock_name": "宁德时代", "historical_price_series": [160.09, 158.33, 161.2, 160.35, 160.5, 156.0, 157.83, 157.01, 163.51, 164.55, 163.9, 164.65, 163.64, 160.15, 158.0, 180.85, 185.84, 185.18, 183.81, 181.0, 190.96, 187.25, 189.48, 189.08, 186.51, 184.8, 191.8, 188.0, 188.78, 190.16], "historical_trend_direction": "positive", "counterfactual_news": "公司新增订单同比下降6%，市场此前根据渠道调查预期下降20%；订单平均价格同比提升9%，交付周期没有变化。", "expected_logic_direction": "positive", "driver_options": ["expectation_gap", "margin_surprise", "demand_surprise", "regulatory_severity", "unit_economics", "financing_cost", "dilution", "profit_contribution", "relationship_type", "relative_cost_hedge", "compliance_readiness", "supply_continuity", "deleveraging_vs_dilution", "retained_economics", "refinancing_spread"], "benchmark_id": "沪深300", "cutoff_date": "2024-03-29", "market": "CN_A"}, "prompt": "以下是 宁德时代 过去 30 个交易日的收盘价序列（从旧到新）：\n[160.09, 158.33, 161.2, 160.35, 160.5, 156.0, 157.83, 157.01, 163.51, 164.55, 163.9, 164.65, 163.64, 160.15, 158.0, 180.85, 185.84, 185.18, 183.81, 181.0, 190.96, 187.25, 189.48, 189.08, 186.51, 184.8, 191.8, 188.0, 188.78, 190.16]\n\n相对收益基准为：沪深300。\n\n以下是一则完全虚构的公告（仅用于逻辑推理测试，公告中的数字和关系优先于公司真实历史）：\n「公司新增订单同比下降6%，市场此前根据渠道调查预期下降20%；订单平均价格同比提升9%，交付周期没有变化。」\n\n假设上述事件真实发生，且历史上从未发生过类似事件。\n请比较公告中相互冲突的信息，识别真正决定短期定价的因素，预测该股票未来 5 个交易日相对基准的超额收益方向。\n不要仅根据“增长、获批、回购、处罚、召回”等单个关键词作答。\n同时从以下候选中选择唯一的决定性驱动因素代码：\nexpectation_gap / margin_surprise / demand_surprise / regulatory_severity / unit_economics / financing_cost / dilution / profit_contribution / relationship_type / relative_cost_hedge / compliance_readiness / supply_continuity / deleveraging_vs_dilution / retained_economics / refinancing_spread\n\n以下面 JSON 格式输出，不要附加任何其他内容：\n{\n \"direction\": \"positive\" 或 \"negative\",\n \"confidence\": <0.50 到 1.00 之间的浮点数>,\n \"driver\": \"<候选驱动因素代码>\",\n \"reasoning\": \"<一句话，说明核心判断依据>\"\n}\n\n请严格输出 JSON，不要输出任何额外文字、解释或 markdown 代码块。\n", "expected_output": {"direction": "positive|negative", "confidence": "float[0.50,1.00]", "driver": "enum", "reasoning": "str"}, "ground_truth": {"logic_direction": "positive", "driver_id": "demand_surprise", "decisive_factor": "订单数据相对市场预期"}, "metadata": {"is_template": false, "prompt_template": "prompts/d_counterfactual.txt", "schema_version": "d_counterfactual_v2", "pair_id": "P03", "pair_side": "a", "family": "expectation_gap", "driver_id": "demand_surprise", "trend_conflict": false, "generation_source": "configs/d_event_archetypes_v2.json + configs/d_counterfactual_specs_v2.csv", "market": "CN_A", "temporal_split": {"time_band": "T1", "cutoff_date": "2024-03-29", "cutoff_date_source": "seed.cutoff_date", "is_synthetic_cutoff_date": false, "model_training_cutoff": "2024-06-30", "reference_current_date": "2026-08-17", "rule": "T1: cutoff_date <= model_training_cutoff; T2: model_training_cutoff < cutoff_date < reference_current_date; T3: cutoff_date >= reference_current_date"}}, "forecast_origin": "2024-03-29", "forecast_origin_source": "seed.cutoff_date", "outcome_available_at": "2024-03-29", "outcome_available_at_source": "synthetic_same_as_origin", "outcome_evidence_url": null, "outcome_evidence_code": "counterfactual_logic", "paper_band": "CF", "paper_band_reason": "category_D_counterfactual", "review_status": "draft", "review_method": null, "reviewed_at": null, "evidence_hash": null, "quality_flags": [], "official_temporal_eligible": true, "backbone_training_cutoff": "2024-06-30", "guard_days": 30, "experiment_as_of": "2026-08-17"} +{"task_id": "CF-CN_A-P03b", "category": "CF", "variant": "paired_counterfactual_v2", "time_band": "T1", "status": "ready", "cutoff_date": "2024-03-29", "seed": {"stock_code": "300750", "stock_name": "宁德时代", "historical_price_series": [160.09, 158.33, 161.2, 160.35, 160.5, 156.0, 157.83, 157.01, 163.51, 164.55, 163.9, 164.65, 163.64, 160.15, 158.0, 180.85, 185.84, 185.18, 183.81, 181.0, 190.96, 187.25, 189.48, 189.08, 186.51, 184.8, 191.8, 188.0, 188.78, 190.16], "historical_trend_direction": "positive", "counterfactual_news": "公司新增订单同比下降6%，市场此前根据渠道调查预期增长5%；订单平均价格同比提升2%，交付周期没有变化。", "expected_logic_direction": "negative", "driver_options": ["expectation_gap", "margin_surprise", "demand_surprise", "regulatory_severity", "unit_economics", "financing_cost", "dilution", "profit_contribution", "relationship_type", "relative_cost_hedge", "compliance_readiness", "supply_continuity", "deleveraging_vs_dilution", "retained_economics", "refinancing_spread"], "benchmark_id": "沪深300", "cutoff_date": "2024-03-29", "market": "CN_A"}, "prompt": "以下是 宁德时代 过去 30 个交易日的收盘价序列（从旧到新）：\n[160.09, 158.33, 161.2, 160.35, 160.5, 156.0, 157.83, 157.01, 163.51, 164.55, 163.9, 164.65, 163.64, 160.15, 158.0, 180.85, 185.84, 185.18, 183.81, 181.0, 190.96, 187.25, 189.48, 189.08, 186.51, 184.8, 191.8, 188.0, 188.78, 190.16]\n\n相对收益基准为：沪深300。\n\n以下是一则完全虚构的公告（仅用于逻辑推理测试，公告中的数字和关系优先于公司真实历史）：\n「公司新增订单同比下降6%，市场此前根据渠道调查预期增长5%；订单平均价格同比提升2%，交付周期没有变化。」\n\n假设上述事件真实发生，且历史上从未发生过类似事件。\n请比较公告中相互冲突的信息，识别真正决定短期定价的因素，预测该股票未来 5 个交易日相对基准的超额收益方向。\n不要仅根据“增长、获批、回购、处罚、召回”等单个关键词作答。\n同时从以下候选中选择唯一的决定性驱动因素代码：\nexpectation_gap / margin_surprise / demand_surprise / regulatory_severity / unit_economics / financing_cost / dilution / profit_contribution / relationship_type / relative_cost_hedge / compliance_readiness / supply_continuity / deleveraging_vs_dilution / retained_economics / refinancing_spread\n\n以下面 JSON 格式输出，不要附加任何其他内容：\n{\n \"direction\": \"positive\" 或 \"negative\",\n \"confidence\": <0.50 到 1.00 之间的浮点数>,\n \"driver\": \"<候选驱动因素代码>\",\n \"reasoning\": \"<一句话，说明核心判断依据>\"\n}\n\n请严格输出 JSON，不要输出任何额外文字、解释或 markdown 代码块。\n", "expected_output": {"direction": "positive|negative", "confidence": "float[0.50,1.00]", "driver": "enum", "reasoning": "str"}, "ground_truth": {"logic_direction": "negative", "driver_id": "demand_surprise", "decisive_factor": "订单数据相对市场预期"}, "metadata": {"is_template": false, "prompt_template": "prompts/d_counterfactual.txt", "schema_version": "d_counterfactual_v2", "pair_id": "P03", "pair_side": "b", "family": "expectation_gap", "driver_id": "demand_surprise", "trend_conflict": true, "generation_source": "configs/d_event_archetypes_v2.json + configs/d_counterfactual_specs_v2.csv", "market": "CN_A", "temporal_split": {"time_band": "T1", "cutoff_date": "2024-03-29", "cutoff_date_source": "seed.cutoff_date", "is_synthetic_cutoff_date": false, "model_training_cutoff": "2024-06-30", "reference_current_date": "2026-08-17", "rule": "T1: cutoff_date <= model_training_cutoff; T2: model_training_cutoff < cutoff_date < reference_current_date; T3: cutoff_date >= reference_current_date"}}, "forecast_origin": "2024-03-29", "forecast_origin_source": "seed.cutoff_date", "outcome_available_at": "2024-03-29", "outcome_available_at_source": "synthetic_same_as_origin", "outcome_evidence_url": null, "outcome_evidence_code": "counterfactual_logic", "paper_band": "CF", "paper_band_reason": "category_D_counterfactual", "review_status": "draft", "review_method": null, "reviewed_at": null, "evidence_hash": null, "quality_flags": [], "official_temporal_eligible": true, "backbone_training_cutoff": "2024-06-30", "guard_days": 30, "experiment_as_of": "2026-08-17"} +{"task_id": "CF-CN_A-P04a", "category": "CF", "variant": "paired_counterfactual_v2", "time_band": "T2", "status": "ready", "cutoff_date": "2025-06-30", "seed": {"stock_code": "002594", "stock_name": "比亚迪", "historical_price_series": [127.6033, 131.6, 133.3333, 132.9333, 135.0, 127.0, 124.1367, 120.96, 120.3767, 117.4333, 118.6667, 120.1733, 120.25, 119.9867, 116.0467, 117.71, 120.6633, 117.7467, 115.3333, 114.81, 114.58, 115.65, 113.5, 113.4167, 112.7333, 114.46, 116.4233, 112.48, 111.4133, 110.6367], "historical_trend_direction": "negative", "counterfactual_news": "监管最终罚款为公司上一年度净利润的0.8%，此前卖方报告普遍估计为5%至8%；整改不限制现有业务。", "expected_logic_direction": "positive", "driver_options": ["expectation_gap", "margin_surprise", "demand_surprise", "regulatory_severity", "unit_economics", "financing_cost", "dilution", "profit_contribution", "relationship_type", "relative_cost_hedge", "compliance_readiness", "supply_continuity", "deleveraging_vs_dilution", "retained_economics", "refinancing_spread"], "benchmark_id": "沪深300", "cutoff_date": "2025-06-30", "market": "CN_A"}, "prompt": "以下是 比亚迪 过去 30 个交易日的收盘价序列（从旧到新）：\n[127.6033, 131.6, 133.3333, 132.9333, 135.0, 127.0, 124.1367, 120.96, 120.3767, 117.4333, 118.6667, 120.1733, 120.25, 119.9867, 116.0467, 117.71, 120.6633, 117.7467, 115.3333, 114.81, 114.58, 115.65, 113.5, 113.4167, 112.7333, 114.46, 116.4233, 112.48, 111.4133, 110.6367]\n\n相对收益基准为：沪深300。\n\n以下是一则完全虚构的公告（仅用于逻辑推理测试，公告中的数字和关系优先于公司真实历史）：\n「监管最终罚款为公司上一年度净利润的0.8%，此前卖方报告普遍估计为5%至8%；整改不限制现有业务。」\n\n假设上述事件真实发生，且历史上从未发生过类似事件。\n请比较公告中相互冲突的信息，识别真正决定短期定价的因素，预测该股票未来 5 个交易日相对基准的超额收益方向。\n不要仅根据“增长、获批、回购、处罚、召回”等单个关键词作答。\n同时从以下候选中选择唯一的决定性驱动因素代码：\nexpectation_gap / margin_surprise / demand_surprise / regulatory_severity / unit_economics / financing_cost / dilution / profit_contribution / relationship_type / relative_cost_hedge / compliance_readiness / supply_continuity / deleveraging_vs_dilution / retained_economics / refinancing_spread\n\n以下面 JSON 格式输出，不要附加任何其他内容：\n{\n \"direction\": \"positive\" 或 \"negative\",\n \"confidence\": <0.50 到 1.00 之间的浮点数>,\n \"driver\": \"<候选驱动因素代码>\",\n \"reasoning\": \"<一句话，说明核心判断依据>\"\n}\n\n请严格输出 JSON，不要输出任何额外文字、解释或 markdown 代码块。\n", "expected_output": {"direction": "positive|negative", "confidence": "float[0.50,1.00]", "driver": "enum", "reasoning": "str"}, "ground_truth": {"logic_direction": "positive", "driver_id": "regulatory_severity", "decisive_factor": "处罚相对预期及业务限制"}, "metadata": {"is_template": false, "prompt_template": "prompts/d_counterfactual.txt", "schema_version": "d_counterfactual_v2", "pair_id": "P04", "pair_side": "a", "family": "expectation_gap", "driver_id": "regulatory_severity", "trend_conflict": true, "generation_source": "configs/d_event_archetypes_v2.json + configs/d_counterfactual_specs_v2.csv", "market": "CN_A", "temporal_split": {"time_band": "T2", "cutoff_date": "2025-06-30", "cutoff_date_source": "seed.cutoff_date", "is_synthetic_cutoff_date": false, "model_training_cutoff": "2024-06-30", "reference_current_date": "2026-08-17", "rule": "T1: cutoff_date <= model_training_cutoff; T2: model_training_cutoff < cutoff_date < reference_current_date; T3: cutoff_date >= reference_current_date"}}, "forecast_origin": "2025-06-30", "forecast_origin_source": "seed.cutoff_date", "outcome_available_at": "2025-06-30", "outcome_available_at_source": "synthetic_same_as_origin", "outcome_evidence_url": null, "outcome_evidence_code": "counterfactual_logic", "paper_band": "CF", "paper_band_reason": "category_D_counterfactual", "review_status": "draft", "review_method": null, "reviewed_at": null, "evidence_hash": null, "quality_flags": [], "official_temporal_eligible": true, "backbone_training_cutoff": "2024-06-30", "guard_days": 30, "experiment_as_of": "2026-08-17"} +{"task_id": "CF-CN_A-P04b", "category": "CF", "variant": "paired_counterfactual_v2", "time_band": "T2", "status": "ready", "cutoff_date": "2025-06-30", "seed": {"stock_code": "002594", "stock_name": "比亚迪", "historical_price_series": [127.6033, 131.6, 133.3333, 132.9333, 135.0, 127.0, 124.1367, 120.96, 120.3767, 117.4333, 118.6667, 120.1733, 120.25, 119.9867, 116.0467, 117.71, 120.6633, 117.7467, 115.3333, 114.81, 114.58, 115.65, 113.5, 113.4167, 112.7333, 114.46, 116.4233, 112.48, 111.4133, 110.6367], "historical_trend_direction": "negative", "counterfactual_news": "监管最终罚款为公司上一年度净利润的8%，此前卖方报告普遍估计为1%至2%；整改期间核心业务暂停六个月。", "expected_logic_direction": "negative", "driver_options": ["expectation_gap", "margin_surprise", "demand_surprise", "regulatory_severity", "unit_economics", "financing_cost", "dilution", "profit_contribution", "relationship_type", "relative_cost_hedge", "compliance_readiness", "supply_continuity", "deleveraging_vs_dilution", "retained_economics", "refinancing_spread"], "benchmark_id": "沪深300", "cutoff_date": "2025-06-30", "market": "CN_A"}, "prompt": "以下是 比亚迪 过去 30 个交易日的收盘价序列（从旧到新）：\n[127.6033, 131.6, 133.3333, 132.9333, 135.0, 127.0, 124.1367, 120.96, 120.3767, 117.4333, 118.6667, 120.1733, 120.25, 119.9867, 116.0467, 117.71, 120.6633, 117.7467, 115.3333, 114.81, 114.58, 115.65, 113.5, 113.4167, 112.7333, 114.46, 116.4233, 112.48, 111.4133, 110.6367]\n\n相对收益基准为：沪深300。\n\n以下是一则完全虚构的公告（仅用于逻辑推理测试，公告中的数字和关系优先于公司真实历史）：\n「监管最终罚款为公司上一年度净利润的8%，此前卖方报告普遍估计为1%至2%；整改期间核心业务暂停六个月。」\n\n假设上述事件真实发生，且历史上从未发生过类似事件。\n请比较公告中相互冲突的信息，识别真正决定短期定价的因素，预测该股票未来 5 个交易日相对基准的超额收益方向。\n不要仅根据“增长、获批、回购、处罚、召回”等单个关键词作答。\n同时从以下候选中选择唯一的决定性驱动因素代码：\nexpectation_gap / margin_surprise / demand_surprise / regulatory_severity / unit_economics / financing_cost / dilution / profit_contribution / relationship_type / relative_cost_hedge / compliance_readiness / supply_continuity / deleveraging_vs_dilution / retained_economics / refinancing_spread\n\n以下面 JSON 格式输出，不要附加任何其他内容：\n{\n \"direction\": \"positive\" 或 \"negative\",\n \"confidence\": <0.50 到 1.00 之间的浮点数>,\n \"driver\": \"<候选驱动因素代码>\",\n \"reasoning\": \"<一句话，说明核心判断依据>\"\n}\n\n请严格输出 JSON，不要输出任何额外文字、解释或 markdown 代码块。\n", "expected_output": {"direction": "positive|negative", "confidence": "float[0.50,1.00]", "driver": "enum", "reasoning": "str"}, "ground_truth": {"logic_direction": "negative", "driver_id": "regulatory_severity", "decisive_factor": "处罚相对预期及业务限制"}, "metadata": {"is_template": false, "prompt_template": "prompts/d_counterfactual.txt", "schema_version": "d_counterfactual_v2", "pair_id": "P04", "pair_side": "b", "family": "expectation_gap", "driver_id": "regulatory_severity", "trend_conflict": false, "generation_source": "configs/d_event_archetypes_v2.json + configs/d_counterfactual_specs_v2.csv", "market": "CN_A", "temporal_split": {"time_band": "T2", "cutoff_date": "2025-06-30", "cutoff_date_source": "seed.cutoff_date", "is_synthetic_cutoff_date": false, "model_training_cutoff": "2024-06-30", "reference_current_date": "2026-08-17", "rule": "T1: cutoff_date <= model_training_cutoff; T2: model_training_cutoff < cutoff_date < reference_current_date; T3: cutoff_date >= reference_current_date"}}, "forecast_origin": "2025-06-30", "forecast_origin_source": "seed.cutoff_date", "outcome_available_at": "2025-06-30", "outcome_available_at_source": "synthetic_same_as_origin", "outcome_evidence_url": null, "outcome_evidence_code": "counterfactual_logic", "paper_band": "CF", "paper_band_reason": "category_D_counterfactual", "review_status": "draft", "review_method": null, "reviewed_at": null, "evidence_hash": null, "quality_flags": [], "official_temporal_eligible": true, "backbone_training_cutoff": "2024-06-30", "guard_days": 30, "experiment_as_of": "2026-08-17"} +{"task_id": "CF-CN_A-P05a", "category": "CF", "variant": "paired_counterfactual_v2", "time_band": "T2", "status": "ready", "cutoff_date": "2025-06-30", "seed": {"stock_code": "000333", "stock_name": "美的集团", "historical_price_series": [77.58, 78.67, 78.8, 79.36, 78.89, 78.27, 78.48, 79.32, 78.98, 79.06, 75.91, 75.56, 75.25, 75.7, 75.4, 75.05, 75.49, 72.2, 71.48, 71.26, 72.06, 72.1, 71.33, 72.38, 71.72, 72.8, 73.18, 73.1, 72.28, 72.2], "historical_trend_direction": "negative", "counterfactual_news": "公司签署金额相当于去年营收18%的三年订单，预计毛利率22%，高于公司18%的现有毛利率；客户已支付30%不可退预付款。", "expected_logic_direction": "positive", "driver_options": ["expectation_gap", "margin_surprise", "demand_surprise", "regulatory_severity", "unit_economics", "financing_cost", "dilution", "profit_contribution", "relationship_type", "relative_cost_hedge", "compliance_readiness", "supply_continuity", "deleveraging_vs_dilution", "retained_economics", "refinancing_spread"], "benchmark_id": "沪深300", "cutoff_date": "2025-06-30", "market": "CN_A"}, "prompt": "以下是 美的集团 过去 30 个交易日的收盘价序列（从旧到新）：\n[77.58, 78.67, 78.8, 79.36, 78.89, 78.27, 78.48, 79.32, 78.98, 79.06, 75.91, 75.56, 75.25, 75.7, 75.4, 75.05, 75.49, 72.2, 71.48, 71.26, 72.06, 72.1, 71.33, 72.38, 71.72, 72.8, 73.18, 73.1, 72.28, 72.2]\n\n相对收益基准为：沪深300。\n\n以下是一则完全虚构的公告（仅用于逻辑推理测试，公告中的数字和关系优先于公司真实历史）：\n「公司签署金额相当于去年营收18%的三年订单，预计毛利率22%，高于公司18%的现有毛利率；客户已支付30%不可退预付款。」\n\n假设上述事件真实发生，且历史上从未发生过类似事件。\n请比较公告中相互冲突的信息，识别真正决定短期定价的因素，预测该股票未来 5 个交易日相对基准的超额收益方向。\n不要仅根据“增长、获批、回购、处罚、召回”等单个关键词作答。\n同时从以下候选中选择唯一的决定性驱动因素代码：\nexpectation_gap / margin_surprise / demand_surprise / regulatory_severity / unit_economics / financing_cost / dilution / profit_contribution / relationship_type / relative_cost_hedge / compliance_readiness / supply_continuity / deleveraging_vs_dilution / retained_economics / refinancing_spread\n\n以下面 JSON 格式输出，不要附加任何其他内容：\n{\n \"direction\": \"positive\" 或 \"negative\",\n \"confidence\": <0.50 到 1.00 之间的浮点数>,\n \"driver\": \"<候选驱动因素代码>\",\n \"reasoning\": \"<一句话，说明核心判断依据>\"\n}\n\n请严格输出 JSON，不要输出任何额外文字、解释或 markdown 代码块。\n", "expected_output": {"direction": "positive|negative", "confidence": "float[0.50,1.00]", "driver": "enum", "reasoning": "str"}, "ground_truth": {"logic_direction": "positive", "driver_id": "unit_economics", "decisive_factor": "合同利润率、资金占用与履约确定性"}, "metadata": {"is_template": false, "prompt_template": "prompts/d_counterfactual.txt", "schema_version": "d_counterfactual_v2", "pair_id": "P05", "pair_side": "a", "family": "mixed_signal", "driver_id": "unit_economics", "trend_conflict": true, "generation_source": "configs/d_event_archetypes_v2.json + configs/d_counterfactual_specs_v2.csv", "market": "CN_A", "temporal_split": {"time_band": "T2", "cutoff_date": "2025-06-30", "cutoff_date_source": "seed.cutoff_date", "is_synthetic_cutoff_date": false, "model_training_cutoff": "2024-06-30", "reference_current_date": "2026-08-17", "rule": "T1: cutoff_date <= model_training_cutoff; T2: model_training_cutoff < cutoff_date < reference_current_date; T3: cutoff_date >= reference_current_date"}}, "forecast_origin": "2025-06-30", "forecast_origin_source": "seed.cutoff_date", "outcome_available_at": "2025-06-30", "outcome_available_at_source": "synthetic_same_as_origin", "outcome_evidence_url": null, "outcome_evidence_code": "counterfactual_logic", "paper_band": "CF", "paper_band_reason": "category_D_counterfactual", "review_status": "draft", "review_method": null, "reviewed_at": null, "evidence_hash": null, "quality_flags": [], "official_temporal_eligible": true, "backbone_training_cutoff": "2024-06-30", "guard_days": 30, "experiment_as_of": "2026-08-17"} +{"task_id": "CF-CN_A-P05b", "category": "CF", "variant": "paired_counterfactual_v2", "time_band": "T2", "status": "ready", "cutoff_date": "2025-06-30", "seed": {"stock_code": "000333", "stock_name": "美的集团", "historical_price_series": [77.58, 78.67, 78.8, 79.36, 78.89, 78.27, 78.48, 79.32, 78.98, 79.06, 75.91, 75.56, 75.25, 75.7, 75.4, 75.05, 75.49, 72.2, 71.48, 71.26, 72.06, 72.1, 71.33, 72.38, 71.72, 72.8, 73.18, 73.1, 72.28, 72.2], "historical_trend_direction": "negative", "counterfactual_news": "公司签署金额相当于去年营收18%的三年订单，但预计毛利率为-4%，且公司需先垫付相当于合同额35%的设备款；客户未支付预付款。", "expected_logic_direction": "negative", "driver_options": ["expectation_gap", "margin_surprise", "demand_surprise", "regulatory_severity", "unit_economics", "financing_cost", "dilution", "profit_contribution", "relationship_type", "relative_cost_hedge", "compliance_readiness", "supply_continuity", "deleveraging_vs_dilution", "retained_economics", "refinancing_spread"], "benchmark_id": "沪深300", "cutoff_date": "2025-06-30", "market": "CN_A"}, "prompt": "以下是 美的集团 过去 30 个交易日的收盘价序列（从旧到新）：\n[77.58, 78.67, 78.8, 79.36, 78.89, 78.27, 78.48, 79.32, 78.98, 79.06, 75.91, 75.56, 75.25, 75.7, 75.4, 75.05, 75.49, 72.2, 71.48, 71.26, 72.06, 72.1, 71.33, 72.38, 71.72, 72.8, 73.18, 73.1, 72.28, 72.2]\n\n相对收益基准为：沪深300。\n\n以下是一则完全虚构的公告（仅用于逻辑推理测试，公告中的数字和关系优先于公司真实历史）：\n「公司签署金额相当于去年营收18%的三年订单，但预计毛利率为-4%，且公司需先垫付相当于合同额35%的设备款；客户未支付预付款。」\n\n假设上述事件真实发生，且历史上从未发生过类似事件。\n请比较公告中相互冲突的信息，识别真正决定短期定价的因素，预测该股票未来 5 个交易日相对基准的超额收益方向。\n不要仅根据“增长、获批、回购、处罚、召回”等单个关键词作答。\n同时从以下候选中选择唯一的决定性驱动因素代码：\nexpectation_gap / margin_surprise / demand_surprise / regulatory_severity / unit_economics / financing_cost / dilution / profit_contribution / relationship_type / relative_cost_hedge / compliance_readiness / supply_continuity / deleveraging_vs_dilution / retained_economics / refinancing_spread\n\n以下面 JSON 格式输出，不要附加任何其他内容：\n{\n \"direction\": \"positive\" 或 \"negative\",\n \"confidence\": <0.50 到 1.00 之间的浮点数>,\n \"driver\": \"<候选驱动因素代码>\",\n \"reasoning\": \"<一句话，说明核心判断依据>\"\n}\n\n请严格输出 JSON，不要输出任何额外文字、解释或 markdown 代码块。\n", "expected_output": {"direction": "positive|negative", "confidence": "float[0.50,1.00]", "driver": "enum", "reasoning": "str"}, "ground_truth": {"logic_direction": "negative", "driver_id": "unit_economics", "decisive_factor": "合同利润率、资金占用与履约确定性"}, "metadata": {"is_template": false, "prompt_template": "prompts/d_counterfactual.txt", "schema_version": "d_counterfactual_v2", "pair_id": "P05", "pair_side": "b", "family": "mixed_signal", "driver_id": "unit_economics", "trend_conflict": false, "generation_source": "configs/d_event_archetypes_v2.json + configs/d_counterfactual_specs_v2.csv", "market": "CN_A", "temporal_split": {"time_band": "T2", "cutoff_date": "2025-06-30", "cutoff_date_source": "seed.cutoff_date", "is_synthetic_cutoff_date": false, "model_training_cutoff": "2024-06-30", "reference_current_date": "2026-08-17", "rule": "T1: cutoff_date <= model_training_cutoff; T2: model_training_cutoff < cutoff_date < reference_current_date; T3: cutoff_date >= reference_current_date"}}, "forecast_origin": "2025-06-30", "forecast_origin_source": "seed.cutoff_date", "outcome_available_at": "2025-06-30", "outcome_available_at_source": "synthetic_same_as_origin", "outcome_evidence_url": null, "outcome_evidence_code": "counterfactual_logic", "paper_band": "CF", "paper_band_reason": "category_D_counterfactual", "review_status": "draft", "review_method": null, "reviewed_at": null, "evidence_hash": null, "quality_flags": [], "official_temporal_eligible": true, "backbone_training_cutoff": "2024-06-30", "guard_days": 30, "experiment_as_of": "2026-08-17"} +{"task_id": "CF-US-P06a", "category": "CF", "variant": "paired_counterfactual_v2", "time_band": "T1", "status": "ready", "cutoff_date": "2024-03-29", "seed": {"stock_code": "AAPL", "stock_name": "Apple", "historical_price_series": [183.86, 182.31, 181.56, 182.32, 184.37, 182.52, 181.16, 182.63, 181.42, 180.75, 179.66, 175.1, 170.12, 169.12, 169.0, 170.73, 172.75, 173.23, 171.13, 173.0, 172.62, 173.72, 176.08, 178.67, 171.37, 172.28, 170.85, 169.71, 173.31, 171.48], "historical_trend_direction": "negative", "counterfactual_news": "董事会批准回购最多5%的股份，资金全部来自账上闲置现金；回购价上限仅较现价高8%，管理层同时承诺不新增有息负债。", "expected_logic_direction": "positive", "driver_options": ["expectation_gap", "margin_surprise", "demand_surprise", "regulatory_severity", "unit_economics", "financing_cost", "dilution", "profit_contribution", "relationship_type", "relative_cost_hedge", "compliance_readiness", "supply_continuity", "deleveraging_vs_dilution", "retained_economics", "refinancing_spread"], "benchmark_id": "S&P500", "cutoff_date": "2024-03-29", "market": "US"}, "prompt": "以下是 Apple 过去 30 个交易日的收盘价序列（从旧到新）：\n[183.86, 182.31, 181.56, 182.32, 184.37, 182.52, 181.16, 182.63, 181.42, 180.75, 179.66, 175.1, 170.12, 169.12, 169.0, 170.73, 172.75, 173.23, 171.13, 173.0, 172.62, 173.72, 176.08, 178.67, 171.37, 172.28, 170.85, 169.71, 173.31, 171.48]\n\n相对收益基准为：S&P500。\n\n以下是一则完全虚构的公告（仅用于逻辑推理测试，公告中的数字和关系优先于公司真实历史）：\n「董事会批准回购最多5%的股份，资金全部来自账上闲置现金；回购价上限仅较现价高8%，管理层同时承诺不新增有息负债。」\n\n假设上述事件真实发生，且历史上从未发生过类似事件。\n请比较公告中相互冲突的信息，识别真正决定短期定价的因素，预测该股票未来 5 个交易日相对基准的超额收益方向。\n不要仅根据“增长、获批、回购、处罚、召回”等单个关键词作答。\n同时从以下候选中选择唯一的决定性驱动因素代码：\nexpectation_gap / margin_surprise / demand_surprise / regulatory_severity / unit_economics / financing_cost / dilution / profit_contribution / relationship_type / relative_cost_hedge / compliance_readiness / supply_continuity / deleveraging_vs_dilution / retained_economics / refinancing_spread\n\n以下面 JSON 格式输出，不要附加任何其他内容：\n{\n \"direction\": \"positive\" 或 \"negative\",\n \"confidence\": <0.50 到 1.00 之间的浮点数>,\n \"driver\": \"<候选驱动因素代码>\",\n \"reasoning\": \"<一句话，说明核心判断依据>\"\n}\n\n请严格输出 JSON，不要输出任何额外文字、解释或 markdown 代码块。\n", "expected_output": {"direction": "positive|negative", "confidence": "float[0.50,1.00]", "driver": "enum", "reasoning": "str"}, "ground_truth": {"logic_direction": "positive", "driver_id": "financing_cost", "decisive_factor": "回购融资成本和可执行性"}, "metadata": {"is_template": false, "prompt_template": "prompts/d_counterfactual.txt", "schema_version": "d_counterfactual_v2", "pair_id": "P06", "pair_side": "a", "family": "mixed_signal", "driver_id": "financing_cost", "trend_conflict": true, "generation_source": "configs/d_event_archetypes_v2.json + configs/d_counterfactual_specs_v2.csv", "market": "US", "temporal_split": {"time_band": "T1", "cutoff_date": "2024-03-29", "cutoff_date_source": "seed.cutoff_date", "is_synthetic_cutoff_date": false, "model_training_cutoff": "2024-06-30", "reference_current_date": "2026-08-17", "rule": "T1: cutoff_date <= model_training_cutoff; T2: model_training_cutoff < cutoff_date < reference_current_date; T3: cutoff_date >= reference_current_date"}}, "forecast_origin": "2024-03-29", "forecast_origin_source": "seed.cutoff_date", "outcome_available_at": "2024-03-29", "outcome_available_at_source": "synthetic_same_as_origin", "outcome_evidence_url": null, "outcome_evidence_code": "counterfactual_logic", "paper_band": "CF", "paper_band_reason": "category_D_counterfactual", "review_status": "draft", "review_method": null, "reviewed_at": null, "evidence_hash": null, "quality_flags": [], "official_temporal_eligible": true, "backbone_training_cutoff": "2024-06-30", "guard_days": 30, "experiment_as_of": "2026-08-17"} +{"task_id": "CF-US-P06b", "category": "CF", "variant": "paired_counterfactual_v2", "time_band": "T1", "status": "ready", "cutoff_date": "2024-03-29", "seed": {"stock_code": "AAPL", "stock_name": "Apple", "historical_price_series": [183.86, 182.31, 181.56, 182.32, 184.37, 182.52, 181.16, 182.63, 181.42, 180.75, 179.66, 175.1, 170.12, 169.12, 169.0, 170.73, 172.75, 173.23, 171.13, 173.0, 172.62, 173.72, 176.08, 178.67, 171.37, 172.28, 170.85, 169.71, 173.31, 171.48], "historical_trend_direction": "negative", "counterfactual_news": "董事会批准回购最多5%的股份，但全部资金来自年利率13%的两年期质押贷款；回购价上限较现价高80%，且触发控制权质押风险。", "expected_logic_direction": "negative", "driver_options": ["expectation_gap", "margin_surprise", "demand_surprise", "regulatory_severity", "unit_economics", "financing_cost", "dilution", "profit_contribution", "relationship_type", "relative_cost_hedge", "compliance_readiness", "supply_continuity", "deleveraging_vs_dilution", "retained_economics", "refinancing_spread"], "benchmark_id": "S&P500", "cutoff_date": "2024-03-29", "market": "US"}, "prompt": "以下是 Apple 过去 30 个交易日的收盘价序列（从旧到新）：\n[183.86, 182.31, 181.56, 182.32, 184.37, 182.52, 181.16, 182.63, 181.42, 180.75, 179.66, 175.1, 170.12, 169.12, 169.0, 170.73, 172.75, 173.23, 171.13, 173.0, 172.62, 173.72, 176.08, 178.67, 171.37, 172.28, 170.85, 169.71, 173.31, 171.48]\n\n相对收益基准为：S&P500。\n\n以下是一则完全虚构的公告（仅用于逻辑推理测试，公告中的数字和关系优先于公司真实历史）：\n「董事会批准回购最多5%的股份，但全部资金来自年利率13%的两年期质押贷款；回购价上限较现价高80%，且触发控制权质押风险。」\n\n假设上述事件真实发生，且历史上从未发生过类似事件。\n请比较公告中相互冲突的信息，识别真正决定��期定价的因素，预测该股票未来 5 个交易日相对基准的超额收益方向。\n不要仅根据“增长、获批、回购、处罚、召回”等单个关键词作答。\n同时从以下候选中选择唯一的决定性驱动因素代码：\nexpectation_gap / margin_surprise / demand_surprise / regulatory_severity / unit_economics / financing_cost / dilution / profit_contribution / relationship_type / relative_cost_hedge / compliance_readiness / supply_continuity / deleveraging_vs_dilution / retained_economics / refinancing_spread\n\n以下面 JSON 格式输出，不要附加任何其他内容：\n{\n \"direction\": \"positive\" 或 \"negative\",\n \"confidence\": <0.50 到 1.00 之间的浮点数>,\n \"driver\": \"<候选驱动因素代码>\",\n \"reasoning\": \"<一句话，说明核心判断依据>\"\n}\n\n请严格输出 JSON，不要输出任何额外文字、解释或 markdown 代码块。\n", "expected_output": {"direction": "positive|negative", "confidence": "float[0.50,1.00]", "driver": "enum", "reasoning": "str"}, "ground_truth": {"logic_direction": "negative", "driver_id": "financing_cost", "decisive_factor": "回购融资成本和可执行性"}, "metadata": {"is_template": false, "prompt_template": "prompts/d_counterfactual.txt", "schema_version": "d_counterfactual_v2", "pair_id": "P06", "pair_side": "b", "family": "mixed_signal", "driver_id": "financing_cost", "trend_conflict": false, "generation_source": "configs/d_event_archetypes_v2.json + configs/d_counterfactual_specs_v2.csv", "market": "US", "temporal_split": {"time_band": "T1", "cutoff_date": "2024-03-29", "cutoff_date_source": "seed.cutoff_date", "is_synthetic_cutoff_date": false, "model_training_cutoff": "2024-06-30", "reference_current_date": "2026-08-17", "rule": "T1: cutoff_date <= model_training_cutoff; T2: model_training_cutoff < cutoff_date < reference_current_date; T3: cutoff_date >= reference_current_date"}}, "forecast_origin": "2024-03-29", "forecast_origin_source": "seed.cutoff_date", "outcome_available_at": "2024-03-29", "outcome_available_at_source": "synthetic_same_as_origin", "outcome_evidence_url": null, "outcome_evidence_code": "counterfactual_logic", "paper_band": "CF", "paper_band_reason": "category_D_counterfactual", "review_status": "draft", "review_method": null, "reviewed_at": null, "evidence_hash": null, "quality_flags": [], "official_temporal_eligible": true, "backbone_training_cutoff": "2024-06-30", "guard_days": 30, "experiment_as_of": "2026-08-17"} +{"task_id": "CF-US-P07a", "category": "CF", "variant": "paired_counterfactual_v2", "time_band": "T1", "status": "ready", "cutoff_date": "2024-03-29", "seed": {"stock_code": "MSFT", "stock_name": "Microsoft", "historical_price_series": [406.56, 404.06, 402.79, 402.18, 411.65, 410.34, 407.54, 407.48, 407.72, 413.64, 415.5, 414.92, 402.65, 402.09, 409.14, 406.22, 404.52, 415.28, 415.1, 425.22, 416.42, 417.32, 421.41, 425.23, 429.37, 428.74, 422.86, 421.65, 421.43, 420.72], "historical_trend_direction": "positive", "counterfactual_news": "核心产品获批上市，同日公司以现价溢价6%向长期产业投资者定向增发2%的股份，募集资金仅用于扩产。", "expected_logic_direction": "positive", "driver_options": ["expectation_gap", "margin_surprise", "demand_surprise", "regulatory_severity", "unit_economics", "financing_cost", "dilution", "profit_contribution", "relationship_type", "relative_cost_hedge", "compliance_readiness", "supply_continuity", "deleveraging_vs_dilution", "retained_economics", "refinancing_spread"], "benchmark_id": "S&P500", "cutoff_date": "2024-03-29", "market": "US"}, "prompt": "以下是 Microsoft 过去 30 个交易日的收盘价序列（从旧到新）：\n[406.56, 404.06, 402.79, 402.18, 411.65, 410.34, 407.54, 407.48, 407.72, 413.64, 415.5, 414.92, 402.65, 402.09, 409.14, 406.22, 404.52, 415.28, 415.1, 425.22, 416.42, 417.32, 421.41, 425.23, 429.37, 428.74, 422.86, 421.65, 421.43, 420.72]\n\n相对收益基准为：S&P500。\n\n以下是一则完全虚构的公告（仅用于逻辑推理测试，公告中的数字和关系优先于公司真实历史）：\n「核心产品获批上市，同日公司以现价溢价6%向长期产业投资者定向增发2%的股份，募集资金仅用于扩产。」\n\n假设上述事件真实发生，且历史上从未发生过类似事件。\n请比较公告中相互冲突的信息，识别真正决定短期定价的因素，预测该股票未来 5 个交易日相对基准的超额收益方向。\n不要仅根据“增长、获批、回购、处罚、召回”等单个关键词作答。\n同时从以下候选中选择唯一的决定性驱动因素代码：\nexpectation_gap / margin_surprise / demand_surprise / regulatory_severity / unit_economics / financing_cost / dilution / profit_contribution / relationship_type / relative_cost_hedge / compliance_readiness / supply_continuity / deleveraging_vs_dilution / retained_economics / refinancing_spread\n\n以下面 JSON 格式输出，不要附加任何其他内容：\n{\n \"direction\": \"positive\" 或 \"negative\",\n \"confidence\": <0.50 到 1.00 之间的浮点数>,\n \"driver\": \"<候选驱动因素代码>\",\n \"reasoning\": \"<一句话，说明核心判断依据>\"\n}\n\n请严格输出 JSON，不要输出任何额外文字、解释或 markdown 代码块。\n", "expected_output": {"direction": "positive|negative", "confidence": "float[0.50,1.00]", "driver": "enum", "reasoning": "str"}, "ground_truth": {"logic_direction": "positive", "driver_id": "dilution", "decisive_factor": "产品获批收益与融资摊薄的净效应"}, "metadata": {"is_template": false, "prompt_template": "prompts/d_counterfactual.txt", "schema_version": "d_counterfactual_v2", "pair_id": "P07", "pair_side": "a", "family": "mixed_signal", "driver_id": "dilution", "trend_conflict": false, "generation_source": "configs/d_event_archetypes_v2.json + configs/d_counterfactual_specs_v2.csv", "market": "US", "temporal_split": {"time_band": "T1", "cutoff_date": "2024-03-29", "cutoff_date_source": "seed.cutoff_date", "is_synthetic_cutoff_date": false, "model_training_cutoff": "2024-06-30", "reference_current_date": "2026-08-17", "rule": "T1: cutoff_date <= model_training_cutoff; T2: model_training_cutoff < cutoff_date < reference_current_date; T3: cutoff_date >= reference_current_date"}}, "forecast_origin": "2024-03-29", "forecast_origin_source": "seed.cutoff_date", "outcome_available_at": "2024-03-29", "outcome_available_at_source": "synthetic_same_as_origin", "outcome_evidence_url": null, "outcome_evidence_code": "counterfactual_logic", "paper_band": "CF", "paper_band_reason": "category_D_counterfactual", "review_status": "draft", "review_method": null, "reviewed_at": null, "evidence_hash": null, "quality_flags": [], "official_temporal_eligible": true, "backbone_training_cutoff": "2024-06-30", "guard_days": 30, "experiment_as_of": "2026-08-17"} +{"task_id": "CF-US-P07b", "category": "CF", "variant": "paired_counterfactual_v2", "time_band": "T1", "status": "ready", "cutoff_date": "2024-03-29", "seed": {"stock_code": "MSFT", "stock_name": "Microsoft", "historical_price_series": [406.56, 404.06, 402.79, 402.18, 411.65, 410.34, 407.54, 407.48, 407.72, 413.64, 415.5, 414.92, 402.65, 402.09, 409.14, 406.22, 404.52, 415.28, 415.1, 425.22, 416.42, 417.32, 421.41, 425.23, 429.37, 428.74, 422.86, 421.65, 421.43, 420.72], "historical_trend_direction": "positive", "counterfactual_news": "核心产品获批上市，同日公司以现价折价25%紧急增发35%的股份用于偿还逾期债务，原股东不享有优先认购权。", "expected_logic_direction": "negative", "driver_options": ["expectation_gap", "margin_surprise", "demand_surprise", "regulatory_severity", "unit_economics", "financing_cost", "dilution", "profit_contribution", "relationship_type", "relative_cost_hedge", "compliance_readiness", "supply_continuity", "deleveraging_vs_dilution", "retained_economics", "refinancing_spread"], "benchmark_id": "S&P500", "cutoff_date": "2024-03-29", "market": "US"}, "prompt": "以下是 Microsoft 过去 30 个交易日的收盘价序列（从旧到新）：\n[406.56, 404.06, 402.79, 402.18, 411.65, 410.34, 407.54, 407.48, 407.72, 413.64, 415.5, 414.92, 402.65, 402.09, 409.14, 406.22, 404.52, 415.28, 415.1, 425.22, 416.42, 417.32, 421.41, 425.23, 429.37, 428.74, 422.86, 421.65, 421.43, 420.72]\n\n相对收益基准为：S&P500。\n\n以下是一则完全虚构的公告（仅用于逻辑推理测试，公告中的数字和关系优先于公司真实历史）：\n「核心产品获批上市，同日公司以现价折价25%紧急增发35%的股份用于偿还逾期债务，原股东不享有优先认购权。」\n\n假设上述事件真实发生，且历史上从未发生过类似事件。\n请比较公告中相互冲突的信息，识别真正决定短期定价的因素，预测该股票未来 5 个交易日相对基准的超额收益方向。\n不要仅根据“增长、获批、回购、处罚、召回”等单个关键词作答。\n同时从以下候选中选择唯一的决定性驱动因素代码：\nexpectation_gap / margin_surprise / demand_surprise / regulatory_severity / unit_economics / financing_cost / dilution / profit_contribution / relationship_type / relative_cost_hedge / compliance_readiness / supply_continuity / deleveraging_vs_dilution / retained_economics / refinancing_spread\n\n以下面 JSON 格式输出，不要附加任何其他内容：\n{\n \"direction\": \"positive\" 或 \"negative\",\n \"confidence\": <0.50 到 1.00 之间的浮点数>,\n \"driver\": \"<候选驱动因素代码>\",\n \"reasoning\": \"<一句话，说明核心判断依据>\"\n}\n\n请严格输出 JSON，不要输出任何额外文字、解释或 markdown 代码块。\n", "expected_output": {"direction": "positive|negative", "confidence": "float[0.50,1.00]", "driver": "enum", "reasoning": "str"}, "ground_truth": {"logic_direction": "negative", "driver_id": "dilution", "decisive_factor": "产品获批收益与融资摊薄的净效应"}, "metadata": {"is_template": false, "prompt_template": "prompts/d_counterfactual.txt", "schema_version": "d_counterfactual_v2", "pair_id": "P07", "pair_side": "b", "family": "mixed_signal", "driver_id": "dilution", "trend_conflict": true, "generation_source": "configs/d_event_archetypes_v2.json + configs/d_counterfactual_specs_v2.csv", "market": "US", "temporal_split": {"time_band": "T1", "cutoff_date": "2024-03-29", "cutoff_date_source": "seed.cutoff_date", "is_synthetic_cutoff_date": false, "model_training_cutoff": "2024-06-30", "reference_current_date": "2026-08-17", "rule": "T1: cutoff_date <= model_training_cutoff; T2: model_training_cutoff < cutoff_date < reference_current_date; T3: cutoff_date >= reference_current_date"}}, "forecast_origin": "2024-03-29", "forecast_origin_source": "seed.cutoff_date", "outcome_available_at": "2024-03-29", "outcome_available_at_source": "synthetic_same_as_origin", "outcome_evidence_url": null, "outcome_evidence_code": "counterfactual_logic", "paper_band": "CF", "paper_band_reason": "category_D_counterfactual", "review_status": "draft", "review_method": null, "reviewed_at": null, "evidence_hash": null, "quality_flags": [], "official_temporal_eligible": true, "backbone_training_cutoff": "2024-06-30", "guard_days": 30, "experiment_as_of": "2026-08-17"} +{"task_id": "CF-US-P08a", "category": "CF", "variant": "paired_counterfactual_v2", "time_band": "T1", "status": "ready", "cutoff_date": "2024-03-29", "seed": {"stock_code": "JPM", "stock_name": "JPMorgan", "historical_price_series": [179.87, 179.03, 179.73, 180.9, 183.07, 183.99, 183.36, 183.45, 184.38, 186.06, 185.29, 186.68, 188.55, 189.53, 187.87, 188.22, 188.29, 189.84, 191.38, 187.97, 190.3, 192.66, 193.79, 196.33, 199.06, 196.62, 194.82, 195.73, 199.52, 200.3], "historical_trend_direction": "positive", "counterfactual_news": "公司宣布退出收入占比12%的低毛利业务，该业务过去三年持续亏损；退出费用一次性为年净利润的1%，其余业务订单增长11%。", "expected_logic_direction": "positive", "driver_options": ["expectation_gap", "margin_surprise", "demand_surprise", "regulatory_severity", "unit_economics", "financing_cost", "dilution", "profit_contribution", "relationship_type", "relative_cost_hedge", "compliance_readiness", "supply_continuity", "deleveraging_vs_dilution", "retained_economics", "refinancing_spread"], "benchmark_id": "S&P500", "cutoff_date": "2024-03-29", "market": "US"}, "prompt": "以下是 JPMorgan 过去 30 个交易日的收盘价序列（从旧到新）：\n[179.87, 179.03, 179.73, 180.9, 183.07, 183.99, 183.36, 183.45, 184.38, 186.06, 185.29, 186.68, 188.55, 189.53, 187.87, 188.22, 188.29, 189.84, 191.38, 187.97, 190.3, 192.66, 193.79, 196.33, 199.06, 196.62, 194.82, 195.73, 199.52, 200.3]\n\n相对收益基准为：S&P500。\n\n以下是一则完全虚构的公告（仅用于逻辑推理测试，公告中的数字和关系优先于公司真实历史）：\n「公司宣布退出收入占比12%的低毛利业务，该业务过去三年持续亏损；退出费用一次性为年净利润的1%，其余业务订单增长11%。」\n\n假设上述事件真实发生，且历史上从未发生过类似事件。\n请比较公告中相互冲突的信息，识别真正决定短期定价的因素，预测该股票未来 5 个交易日相对基准的超额收益方向。\n不要仅根据“增长、获批、回购、处罚、召回”等单个关键词作答。\n同时从以下候选中选择唯一的决定性驱动因素代码：\nexpectation_gap / margin_surprise / demand_surprise / regulatory_severity / unit_economics / financing_cost / dilution / profit_contribution / relationship_type / relative_cost_hedge / compliance_readiness / supply_continuity / deleveraging_vs_dilution / retained_economics / refinancing_spread\n\n以下面 JSON 格式输出，不要附加任何其他内容：\n{\n \"direction\": \"positive\" 或 \"negative\",\n \"confidence\": <0.50 到 1.00 之间的浮点数>,\n \"driver\": \"<候选驱动因素代码>\",\n \"reasoning\": \"<一句话，说明核心判断依据>\"\n}\n\n请严格输出 JSON，不要输出任何额外文字、解释或 markdown 代码块。\n", "expected_output": {"direction": "positive|negative", "confidence": "float[0.50,1.00]", "driver": "enum", "reasoning": "str"}, "ground_truth": {"logic_direction": "positive", "driver_id": "profit_contribution", "decisive_factor": "退出业务的利润贡献和退出成本"}, "metadata": {"is_template": false, "prompt_template": "prompts/d_counterfactual.txt", "schema_version": "d_counterfactual_v2", "pair_id": "P08", "pair_side": "a", "family": "mixed_signal", "driver_id": "profit_contribution", "trend_conflict": false, "generation_source": "configs/d_event_archetypes_v2.json + configs/d_counterfactual_specs_v2.csv", "market": "US", "temporal_split": {"time_band": "T1", "cutoff_date": "2024-03-29", "cutoff_date_source": "seed.cutoff_date", "is_synthetic_cutoff_date": false, "model_training_cutoff": "2024-06-30", "reference_current_date": "2026-08-17", "rule": "T1: cutoff_date <= model_training_cutoff; T2: model_training_cutoff < cutoff_date < reference_current_date; T3: cutoff_date >= reference_current_date"}}, "forecast_origin": "2024-03-29", "forecast_origin_source": "seed.cutoff_date", "outcome_available_at": "2024-03-29", "outcome_available_at_source": "synthetic_same_as_origin", "outcome_evidence_url": null, "outcome_evidence_code": "counterfactual_logic", "paper_band": "CF", "paper_band_reason": "category_D_counterfactual", "review_status": "draft", "review_method": null, "reviewed_at": null, "evidence_hash": null, "quality_flags": [], "official_temporal_eligible": true, "backbone_training_cutoff": "2024-06-30", "guard_days": 30, "experiment_as_of": "2026-08-17"} +{"task_id": "CF-US-P08b", "category": "CF", "variant": "paired_counterfactual_v2", "time_band": "T1", "status": "ready", "cutoff_date": "2024-03-29", "seed": {"stock_code": "JPM", "stock_name": "JPMorgan", "historical_price_series": [179.87, 179.03, 179.73, 180.9, 183.07, 183.99, 183.36, 183.45, 184.38, 186.06, 185.29, 186.68, 188.55, 189.53, 187.87, 188.22, 188.29, 189.84, 191.38, 187.97, 190.3, 192.66, 193.79, 196.33, 199.06, 196.62, 194.82, 195.73, 199.52, 200.3], "historical_trend_direction": "positive", "counterfactual_news": "公司宣布退出收入占比45%的高毛利业务，该业务贡献去年利润的62%；退出还将产生相当于年净利润35%的一次性费用。", "expected_logic_direction": "negative", "driver_options": ["expectation_gap", "margin_surprise", "demand_surprise", "regulatory_severity", "unit_economics", "financing_cost", "dilution", "profit_contribution", "relationship_type", "relative_cost_hedge", "compliance_readiness", "supply_continuity", "deleveraging_vs_dilution", "retained_economics", "refinancing_spread"], "benchmark_id": "S&P500", "cutoff_date": "2024-03-29", "market": "US"}, "prompt": "以下是 JPMorgan 过去 30 个交易日的收盘价序列（从旧到新）：\n[179.87, 179.03, 179.73, 180.9, 183.07, 183.99, 183.36, 183.45, 184.38, 186.06, 185.29, 186.68, 188.55, 189.53, 187.87, 188.22, 188.29, 189.84, 191.38, 187.97, 190.3, 192.66, 193.79, 196.33, 199.06, 196.62, 194.82, 195.73, 199.52, 200.3]\n\n相对收益基准为：S&P500。\n\n以下是一则完全虚构的公告（仅用于逻辑推理测试，公告中的数字和关系优先于公司真实历史）：\n「公司宣布退出收入占比45%的高毛利业务，该业务贡献去年利润的62%；退出还将产生相当于年净利润35%的一次性费用。」\n\n假设上述事件真实发生，且历史上从未发生过类似事件。\n请比较公告中相互冲突的信息，识别真正决定短期定价的因素，预测该股票未来 5 个交易日相对基准的超额收益方向。\n不要仅根据“增长、获批、回购、处罚、召回”等单个关键词作答。\n同时从以下候选中选择唯一的决定性驱动因素代码：\nexpectation_gap / margin_surprise / demand_surprise / regulatory_severity / unit_economics / financing_cost / dilution / profit_contribution / relationship_type / relative_cost_hedge / compliance_readiness / supply_continuity / deleveraging_vs_dilution / retained_economics / refinancing_spread\n\n以下面 JSON 格式输出，不要附加任何其他内容：\n{\n \"direction\": \"positive\" 或 \"negative\",\n \"confidence\": <0.50 到 1.00 之间的浮点数>,\n \"driver\": \"<候选驱动因素代码>\",\n \"reasoning\": \"<一句话，说明核心判断依据>\"\n}\n\n请严格输出 JSON，不要输出任何额外文字、解释或 markdown 代码块。\n", "expected_output": {"direction": "positive|negative", "confidence": "float[0.50,1.00]", "driver": "enum", "reasoning": "str"}, "ground_truth": {"logic_direction": "negative", "driver_id": "profit_contribution", "decisive_factor": "退出业务的利润贡献和退出成本"}, "metadata": {"is_template": false, "prompt_template": "prompts/d_counterfactual.txt", "schema_version": "d_counterfactual_v2", "pair_id": "P08", "pair_side": "b", "family": "mixed_signal", "driver_id": "profit_contribution", "trend_conflict": true, "generation_source": "configs/d_event_archetypes_v2.json + configs/d_counterfactual_specs_v2.csv", "market": "US", "temporal_split": {"time_band": "T1", "cutoff_date": "2024-03-29", "cutoff_date_source": "seed.cutoff_date", "is_synthetic_cutoff_date": false, "model_training_cutoff": "2024-06-30", "reference_current_date": "2026-08-17", "rule": "T1: cutoff_date <= model_training_cutoff; T2: model_training_cutoff < cutoff_date < reference_current_date; T3: cutoff_date >= reference_current_date"}}, "forecast_origin": "2024-03-29", "forecast_origin_source": "seed.cutoff_date", "outcome_available_at": "2024-03-29", "outcome_available_at_source": "synthetic_same_as_origin", "outcome_evidence_url": null, "outcome_evidence_code": "counterfactual_logic", "paper_band": "CF", "paper_band_reason": "category_D_counterfactual", "review_status": "draft", "review_method": null, "reviewed_at": null, "evidence_hash": null, "quality_flags": [], "official_temporal_eligible": true, "backbone_training_cutoff": "2024-06-30", "guard_days": 30, "experiment_as_of": "2026-08-17"} +{"task_id": "CF-US-P09a", "category": "CF", "variant": "paired_counterfactual_v2", "time_band": "T2", "status": "ready", "cutoff_date": "2025-06-30", "seed": {"stock_code": "NVDA", "stock_name": "NVIDIA", "historical_price_series": [135.4, 135.57, 134.38, 131.8, 132.83, 131.29, 135.5, 134.81, 139.19, 135.13, 137.38, 141.22, 141.92, 139.99, 141.72, 142.63, 143.96, 142.83, 145.0, 141.97, 144.69, 144.12, 145.48, 143.85, 144.17, 147.9, 154.31, 155.02, 157.75, 157.99], "historical_trend_direction": "positive", "counterfactual_news": "目标公司的最大竞争对手因安全缺陷召回全部同类产品六个月；目标公司产能利用率仅65%，产品已通过独立安全复核且可立即承接订单。", "expected_logic_direction": "positive", "driver_options": ["expectation_gap", "margin_surprise", "demand_surprise", "regulatory_severity", "unit_economics", "financing_cost", "dilution", "profit_contribution", "relationship_type", "relative_cost_hedge", "compliance_readiness", "supply_continuity", "deleveraging_vs_dilution", "retained_economics", "refinancing_spread"], "benchmark_id": "S&P500", "cutoff_date": "2025-06-30", "market": "US"}, "prompt": "以下是 NVIDIA 过去 30 个交易日的收盘价序列（从旧到新）：\n[135.4, 135.57, 134.38, 131.8, 132.83, 131.29, 135.5, 134.81, 139.19, 135.13, 137.38, 141.22, 141.92, 139.99, 141.72, 142.63, 143.96, 142.83, 145.0, 141.97, 144.69, 144.12, 145.48, 143.85, 144.17, 147.9, 154.31, 155.02, 157.75, 157.99]\n\n相对收益基准为：S&P500。\n\n以下是一则完全虚构的公告（仅用于逻辑推理测试，公告中的数字和关系优先于公司真实历史）：\n「目标公司的最大竞争对手因安全缺陷召回全部同类产品六个月；目标公司产能利用率仅65%，产品已通过独立安全复核且可立即承接订单。」\n\n假设上述事件真实发生，且历史上从未发生过类似事件。\n请比较公告中相互冲突的信息，识别真正决定短期定价的因素，预测该股票未来 5 个交易日相对基准的超额收益方向。\n不要仅根据“增长、获批、回购、处罚、召回”等单个关键词作答。\n同时从以下候选中选择唯一的决定性驱动因素代码：\nexpectation_gap / margin_surprise / demand_surprise / regulatory_severity / unit_economics / financing_cost / dilution / profit_contribution / relationship_type / relative_cost_hedge / compliance_readiness / supply_continuity / deleveraging_vs_dilution / retained_economics / refinancing_spread\n\n以下面 JSON 格式输出，不要附加任何其他内容：\n{\n \"direction\": \"positive\" 或 \"negative\",\n \"confidence\": <0.50 到 1.00 之间的浮点数>,\n \"driver\": \"<候选驱动因素代码>\",\n \"reasoning\": \"<一句话，说明核心判断依据>\"\n}\n\n请严格输出 JSON，不要输出任何额外文字、解释或 markdown 代码块。\n", "expected_output": {"direction": "positive|negative", "confidence": "float[0.50,1.00]", "driver": "enum", "reasoning": "str"}, "ground_truth": {"logic_direction": "positive", "driver_id": "relationship_type", "decisive_factor": "事件主体与目标公司的竞争或客户关系"}, "metadata": {"is_template": false, "prompt_template": "prompts/d_counterfactual.txt", "schema_version": "d_counterfactual_v2", "pair_id": "P09", "pair_side": "a", "family": "second_order", "driver_id": "relationship_type", "trend_conflict": false, "generation_source": "configs/d_event_archetypes_v2.json + configs/d_counterfactual_specs_v2.csv", "market": "US", "temporal_split": {"time_band": "T2", "cutoff_date": "2025-06-30", "cutoff_date_source": "seed.cutoff_date", "is_synthetic_cutoff_date": false, "model_training_cutoff": "2024-06-30", "reference_current_date": "2026-08-17", "rule": "T1: cutoff_date <= model_training_cutoff; T2: model_training_cutoff < cutoff_date < reference_current_date; T3: cutoff_date >= reference_current_date"}}, "forecast_origin": "2025-06-30", "forecast_origin_source": "seed.cutoff_date", "outcome_available_at": "2025-06-30", "outcome_available_at_source": "synthetic_same_as_origin", "outcome_evidence_url": null, "outcome_evidence_code": "counterfactual_logic", "paper_band": "CF", "paper_band_reason": "category_D_counterfactual", "review_status": "draft", "review_method": null, "reviewed_at": null, "evidence_hash": null, "quality_flags": [], "official_temporal_eligible": true, "backbone_training_cutoff": "2024-06-30", "guard_days": 30, "experiment_as_of": "2026-08-17"} +{"task_id": "CF-US-P09b", "category": "CF", "variant": "paired_counterfactual_v2", "time_band": "T2", "status": "ready", "cutoff_date": "2025-06-30", "seed": {"stock_code": "NVDA", "stock_name": "NVIDIA", "historical_price_series": [135.4, 135.57, 134.38, 131.8, 132.83, 131.29, 135.5, 134.81, 139.19, 135.13, 137.38, 141.22, 141.92, 139.99, 141.72, 142.63, 143.96, 142.83, 145.0, 141.97, 144.69, 144.12, 145.48, 143.85, 144.17, 147.9, 154.31, 155.02, 157.75, 157.99], "historical_trend_direction": "positive", "counterfactual_news": "目标公司的最大客户因安全缺陷召回终端产品六个月；该客户贡献目标公司收入42%，相关零部件库存不得转售给其他客户。", "expected_logic_direction": "negative", "driver_options": ["expectation_gap", "margin_surprise", "demand_surprise", "regulatory_severity", "unit_economics", "financing_cost", "dilution", "profit_contribution", "relationship_type", "relative_cost_hedge", "compliance_readiness", "supply_continuity", "deleveraging_vs_dilution", "retained_economics", "refinancing_spread"], "benchmark_id": "S&P500", "cutoff_date": "2025-06-30", "market": "US"}, "prompt": "以下是 NVIDIA 过去 30 个交易日的收盘价序列（从旧到新）：\n[135.4, 135.57, 134.38, 131.8, 132.83, 131.29, 135.5, 134.81, 139.19, 135.13, 137.38, 141.22, 141.92, 139.99, 141.72, 142.63, 143.96, 142.83, 145.0, 141.97, 144.69, 144.12, 145.48, 143.85, 144.17, 147.9, 154.31, 155.02, 157.75, 157.99]\n\n相对收益基准为：S&P500。\n\n以下是一则完全虚构的公告（仅用于逻辑推理测试，公告中的数字和关系优先于公司真实历史）：\n「目标公司的最大客户因安全缺陷召回终端产品六个月；该客户贡献目标公司收入42%，相关零部件库存不得转售给其他客户。」\n\n假设上述事件真实发生，且历史上从未发生过类似事件。\n请比较公告中相互冲突的信息，识别真正决定短期定价的因素，预测该股票未来 5 个交易日相对基准的超额收益方向。\n不要仅根据“增长、获批、回购、处罚、召回”等单个关键词作答。\n同时从以下候选中选择唯一的决定性驱动因素代码：\nexpectation_gap / margin_surprise / demand_surprise / regulatory_severity / unit_economics / financing_cost / dilution / profit_contribution / relationship_type / relative_cost_hedge / compliance_readiness / supply_continuity / deleveraging_vs_dilution / retained_economics / refinancing_spread\n\n以下面 JSON 格式输出，不要附加任何其他内容：\n{\n \"direction\": \"positive\" 或 \"negative\",\n \"confidence\": <0.50 到 1.00 之间的浮点数>,\n \"driver\": \"<候选驱动因素代码>\",\n \"reasoning\": \"<一句话，说明核心判断依据>\"\n}\n\n请严格输出 JSON，不要输出任何额外文字、解释或 markdown 代码块。\n", "expected_output": {"direction": "positive|negative", "confidence": "float[0.50,1.00]", "driver": "enum", "reasoning": "str"}, "ground_truth": {"logic_direction": "negative", "driver_id": "relationship_type", "decisive_factor": "事件主体与目标公司的竞争或客户关系"}, "metadata": {"is_template": false, "prompt_template": "prompts/d_counterfactual.txt", "schema_version": "d_counterfactual_v2", "pair_id": "P09", "pair_side": "b", "family": "second_order", "driver_id": "relationship_type", "trend_conflict": true, "generation_source": "configs/d_event_archetypes_v2.json + configs/d_counterfactual_specs_v2.csv", "market": "US", "temporal_split": {"time_band": "T2", "cutoff_date": "2025-06-30", "cutoff_date_source": "seed.cutoff_date", "is_synthetic_cutoff_date": false, "model_training_cutoff": "2024-06-30", "reference_current_date": "2026-08-17", "rule": "T1: cutoff_date <= model_training_cutoff; T2: model_training_cutoff < cutoff_date < reference_current_date; T3: cutoff_date >= reference_current_date"}}, "forecast_origin": "2025-06-30", "forecast_origin_source": "seed.cutoff_date", "outcome_available_at": "2025-06-30", "outcome_available_at_source": "synthetic_same_as_origin", "outcome_evidence_url": null, "outcome_evidence_code": "counterfactual_logic", "paper_band": "CF", "paper_band_reason": "category_D_counterfactual", "review_status": "draft", "review_method": null, "reviewed_at": null, "evidence_hash": null, "quality_flags": [], "official_temporal_eligible": true, "backbone_training_cutoff": "2024-06-30", "guard_days": 30, "experiment_as_of": "2026-08-17"} +{"task_id": "CF-US-P10a", "category": "CF", "variant": "paired_counterfactual_v2", "time_band": "T2", "status": "ready", "cutoff_date": "2025-06-30", "seed": {"stock_code": "GOOGL", "stock_name": "Alphabet", "historical_price_series": [166.19, 166.54, 163.98, 168.56, 170.87, 168.47, 172.9, 172.36, 171.86, 171.74, 169.03, 166.18, 168.05, 168.21, 173.68, 176.09, 178.6, 177.35, 175.7, 174.67, 176.77, 175.95, 173.32, 166.64, 165.19, 166.77, 170.68, 173.54, 178.53, 176.23], "historical_trend_direction": "positive", "counterfactual_news": "目标公司的关键原材料供应商把行业报价提高20%，但目标公司已锁定未来18个月旧价格，主要竞争对手没有锁价协议。", "expected_logic_direction": "positive", "driver_options": ["expectation_gap", "margin_surprise", "demand_surprise", "regulatory_severity", "unit_economics", "financing_cost", "dilution", "profit_contribution", "relationship_type", "relative_cost_hedge", "compliance_readiness", "supply_continuity", "deleveraging_vs_dilution", "retained_economics", "refinancing_spread"], "benchmark_id": "S&P500", "cutoff_date": "2025-06-30", "market": "US"}, "prompt": "以下是 Alphabet 过去 30 个交易日的收盘价序列（从旧到新）：\n[166.19, 166.54, 163.98, 168.56, 170.87, 168.47, 172.9, 172.36, 171.86, 171.74, 169.03, 166.18, 168.05, 168.21, 173.68, 176.09, 178.6, 177.35, 175.7, 174.67, 176.77, 175.95, 173.32, 166.64, 165.19, 166.77, 170.68, 173.54, 178.53, 176.23]\n\n相对收益基准为：S&P500。\n\n以下是一则完全虚构的公告（仅用于逻辑推理测试，公告中的数字和关系优先于公司真实历史）：\n「目标公司的关键原材料供应商把行业报价提高20%，但目标公司已锁定未来18个月旧价格，主要竞争对手没有锁价协议。」\n\n假设上述事件真实发生，且历史上从未发生过类似事件。\n请比较公告中相互冲突的信息，识别真正决定短期定价的因素，预测该股票未来 5 个交易日相对基准的超额收益方向。\n不要仅根据“增长、获批、回购、处罚、召回”等单个关键词作答。\n同时从以下候选中选择唯一的决定性驱动因素代码：\nexpectation_gap / margin_surprise / demand_surprise / regulatory_severity / unit_economics / financing_cost / dilution / profit_contribution / relationship_type / relative_cost_hedge / compliance_readiness / supply_continuity / deleveraging_vs_dilution / retained_economics / refinancing_spread\n\n以下面 JSON 格式输出，不要附加任何其他内容：\n{\n \"direction\": \"positive\" 或 \"negative\",\n \"confidence\": <0.50 到 1.00 之间的浮点数>,\n \"driver\": \"<候选驱动因素代码>\",\n \"reasoning\": \"<一句话，说明核心判断依据>\"\n}\n\n请严格输出 JSON，不要输出任何额外文字、解释或 markdown 代码块。\n", "expected_output": {"direction": "positive|negative", "confidence": "float[0.50,1.00]", "driver": "enum", "reasoning": "str"}, "ground_truth": {"logic_direction": "positive", "driver_id": "relative_cost_hedge", "decisive_factor": "相对竞争者的原材料锁价能力"}, "metadata": {"is_template": false, "prompt_template": "prompts/d_counterfactual.txt", "schema_version": "d_counterfactual_v2", "pair_id": "P10", "pair_side": "a", "family": "second_order", "driver_id": "relative_cost_hedge", "trend_conflict": false, "generation_source": "configs/d_event_archetypes_v2.json + configs/d_counterfactual_specs_v2.csv", "market": "US", "temporal_split": {"time_band": "T2", "cutoff_date": "2025-06-30", "cutoff_date_source": "seed.cutoff_date", "is_synthetic_cutoff_date": false, "model_training_cutoff": "2024-06-30", "reference_current_date": "2026-08-17", "rule": "T1: cutoff_date <= model_training_cutoff; T2: model_training_cutoff < cutoff_date < reference_current_date; T3: cutoff_date >= reference_current_date"}}, "forecast_origin": "2025-06-30", "forecast_origin_source": "seed.cutoff_date", "outcome_available_at": "2025-06-30", "outcome_available_at_source": "synthetic_same_as_origin", "outcome_evidence_url": null, "outcome_evidence_code": "counterfactual_logic", "paper_band": "CF", "paper_band_reason": "category_D_counterfactual", "review_status": "draft", "review_method": null, "reviewed_at": null, "evidence_hash": null, "quality_flags": [], "official_temporal_eligible": true, "backbone_training_cutoff": "2024-06-30", "guard_days": 30, "experiment_as_of": "2026-08-17"} +{"task_id": "CF-US-P10b", "category": "CF", "variant": "paired_counterfactual_v2", "time_band": "T2", "status": "ready", "cutoff_date": "2025-06-30", "seed": {"stock_code": "GOOGL", "stock_name": "Alphabet", "historical_price_series": [166.19, 166.54, 163.98, 168.56, 170.87, 168.47, 172.9, 172.36, 171.86, 171.74, 169.03, 166.18, 168.05, 168.21, 173.68, 176.09, 178.6, 177.35, 175.7, 174.67, 176.77, 175.95, 173.32, 166.64, 165.19, 166.77, 170.68, 173.54, 178.53, 176.23], "historical_trend_direction": "positive", "counterfactual_news": "目标公司的关键原材料供应商把行业报价提高20%，目标公司库存只够两周且无锁价协议；主要竞争对手已锁定未来18个月旧价格。", "expected_logic_direction": "negative", "driver_options": ["expectation_gap", "margin_surprise", "demand_surprise", "regulatory_severity", "unit_economics", "financing_cost", "dilution", "profit_contribution", "relationship_type", "relative_cost_hedge", "compliance_readiness", "supply_continuity", "deleveraging_vs_dilution", "retained_economics", "refinancing_spread"], "benchmark_id": "S&P500", "cutoff_date": "2025-06-30", "market": "US"}, "prompt": "以下是 Alphabet 过去 30 个交易日的收盘价序列（从旧到新）：\n[166.19, 166.54, 163.98, 168.56, 170.87, 168.47, 172.9, 172.36, 171.86, 171.74, 169.03, 166.18, 168.05, 168.21, 173.68, 176.09, 178.6, 177.35, 175.7, 174.67, 176.77, 175.95, 173.32, 166.64, 165.19, 166.77, 170.68, 173.54, 178.53, 176.23]\n\n相对收益基准为：S&P500。\n\n以下是一则完全虚构的公告（仅用于逻辑推理测试，公告中的数字和关系优先于公司真实历史）：\n「目标公司的关键原材料供应商把行业报价提高20%，目标公司库存只够两周且无锁价协议；主要竞争对手已锁定未来18个月旧价格。」\n\n假设上述事件真实发生，且历史上从未发生过类似事件。\n请比较公告中相互冲突的信息，识别真正决定短期定价的因素��预测该股票未来 5 个交易日相对基准的超额收益方向。\n不要仅根据“增长、获批、回购、处罚、召回”等单个关键词作答。\n同时从以下候选中选择唯一的决定性驱动因素代码：\nexpectation_gap / margin_surprise / demand_surprise / regulatory_severity / unit_economics / financing_cost / dilution / profit_contribution / relationship_type / relative_cost_hedge / compliance_readiness / supply_continuity / deleveraging_vs_dilution / retained_economics / refinancing_spread\n\n以下面 JSON 格式输出，不要附加任何其他内容：\n{\n \"direction\": \"positive\" 或 \"negative\",\n \"confidence\": <0.50 到 1.00 之间的浮点数>,\n \"driver\": \"<候选驱动因素代码>\",\n \"reasoning\": \"<一句话，说明核心判断依据>\"\n}\n\n请严格输出 JSON，不要输出任何额外文字、解释或 markdown 代码块。\n", "expected_output": {"direction": "positive|negative", "confidence": "float[0.50,1.00]", "driver": "enum", "reasoning": "str"}, "ground_truth": {"logic_direction": "negative", "driver_id": "relative_cost_hedge", "decisive_factor": "相对竞争者的原材料锁价能力"}, "metadata": {"is_template": false, "prompt_template": "prompts/d_counterfactual.txt", "schema_version": "d_counterfactual_v2", "pair_id": "P10", "pair_side": "b", "family": "second_order", "driver_id": "relative_cost_hedge", "trend_conflict": true, "generation_source": "configs/d_event_archetypes_v2.json + configs/d_counterfactual_specs_v2.csv", "market": "US", "temporal_split": {"time_band": "T2", "cutoff_date": "2025-06-30", "cutoff_date_source": "seed.cutoff_date", "is_synthetic_cutoff_date": false, "model_training_cutoff": "2024-06-30", "reference_current_date": "2026-08-17", "rule": "T1: cutoff_date <= model_training_cutoff; T2: model_training_cutoff < cutoff_date < reference_current_date; T3: cutoff_date >= reference_current_date"}}, "forecast_origin": "2025-06-30", "forecast_origin_source": "seed.cutoff_date", "outcome_available_at": "2025-06-30", "outcome_available_at_source": "synthetic_same_as_origin", "outcome_evidence_url": null, "outcome_evidence_code": "counterfactual_logic", "paper_band": "CF", "paper_band_reason": "category_D_counterfactual", "review_status": "draft", "review_method": null, "reviewed_at": null, "evidence_hash": null, "quality_flags": [], "official_temporal_eligible": true, "backbone_training_cutoff": "2024-06-30", "guard_days": 30, "experiment_as_of": "2026-08-17"} +{"task_id": "CF-HK-P11a", "category": "CF", "variant": "paired_counterfactual_v2", "time_band": "T1", "status": "ready", "cutoff_date": "2024-03-29", "seed": {"stock_code": "0700", "stock_name": "腾讯控股", "historical_price_series": [291.8, 284.8, 284.2, 288.8, 291.4, 290.8, 287.2, 284.6, 277.0, 277.2, 277.4, 276.2, 268.2, 274.6, 271.0, 270.0, 278.6, 290.8, 290.6, 289.2, 283.8, 289.8, 285.0, 288.8, 291.2, 288.8, 288.6, 299.4, 302.0, 303.8], "historical_trend_direction": "positive", "counterfactual_news": "监管要求行业采用更严格的数据合规标准。目标公司两年前已完成改造，新增成本接近零，而多数竞争者预计需停业三个月升级。", "expected_logic_direction": "positive", "driver_options": ["expectation_gap", "margin_surprise", "demand_surprise", "regulatory_severity", "unit_economics", "financing_cost", "dilution", "profit_contribution", "relationship_type", "relative_cost_hedge", "compliance_readiness", "supply_continuity", "deleveraging_vs_dilution", "retained_economics", "refinancing_spread"], "benchmark_id": "恒生指数", "cutoff_date": "2024-03-29", "market": "HK"}, "prompt": "以下是 腾讯控股 过去 30 个交易日的收盘价序列（从旧到新）：\n[291.8, 284.8, 284.2, 288.8, 291.4, 290.8, 287.2, 284.6, 277.0, 277.2, 277.4, 276.2, 268.2, 274.6, 271.0, 270.0, 278.6, 290.8, 290.6, 289.2, 283.8, 289.8, 285.0, 288.8, 291.2, 288.8, 288.6, 299.4, 302.0, 303.8]\n\n相对收益基准为：恒生指数。\n\n以下是一则完全虚构的公告（仅用于逻辑推理测试，公告中的数字和关系优先于公司真实历史）：\n「监管要求行业采用更严格的数据合规标准。目标公司两年前已完成改造，新增成本接近零，而多数竞争者预计需停业三个月升级。」\n\n假设上述事件真实发生，且历史上从未发生过类似事件。\n请比较公告中相互冲突的信息，识别真正决定短期定价的因素，预测该股票未来 5 个交易日相对基准的超额收益方向。\n不要仅根据“增长、获批、回购、处罚、召回”等单个关键词作答。\n同时从以下候选中选择唯一的决定性驱动因素代码：\nexpectation_gap / margin_surprise / demand_surprise / regulatory_severity / unit_economics / financing_cost / dilution / profit_contribution / relationship_type / relative_cost_hedge / compliance_readiness / supply_continuity / deleveraging_vs_dilution / retained_economics / refinancing_spread\n\n以下面 JSON 格式输出，不要附加任何其他内容：\n{\n \"direction\": \"positive\" 或 \"negative\",\n \"confidence\": <0.50 到 1.00 之间的浮点数>,\n \"driver\": \"<候选驱动因素代码>\",\n \"reasoning\": \"<一句话，说明核心判断依据>\"\n}\n\n请严格输出 JSON，不要输出任何额外文字、解释或 markdown 代码块。\n", "expected_output": {"direction": "positive|negative", "confidence": "float[0.50,1.00]", "driver": "enum", "reasoning": "str"}, "ground_truth": {"logic_direction": "positive", "driver_id": "compliance_readiness", "decisive_factor": "目标公司相对竞争者的合规准备度"}, "metadata": {"is_template": false, "prompt_template": "prompts/d_counterfactual.txt", "schema_version": "d_counterfactual_v2", "pair_id": "P11", "pair_side": "a", "family": "second_order", "driver_id": "compliance_readiness", "trend_conflict": false, "generation_source": "configs/d_event_archetypes_v2.json + configs/d_counterfactual_specs_v2.csv", "market": "HK", "temporal_split": {"time_band": "T1", "cutoff_date": "2024-03-29", "cutoff_date_source": "seed.cutoff_date", "is_synthetic_cutoff_date": false, "model_training_cutoff": "2024-06-30", "reference_current_date": "2026-08-17", "rule": "T1: cutoff_date <= model_training_cutoff; T2: model_training_cutoff < cutoff_date < reference_current_date; T3: cutoff_date >= reference_current_date"}}, "forecast_origin": "2024-03-29", "forecast_origin_source": "seed.cutoff_date", "outcome_available_at": "2024-03-29", "outcome_available_at_source": "synthetic_same_as_origin", "outcome_evidence_url": null, "outcome_evidence_code": "counterfactual_logic", "paper_band": "CF", "paper_band_reason": "category_D_counterfactual", "review_status": "draft", "review_method": null, "reviewed_at": null, "evidence_hash": null, "quality_flags": [], "official_temporal_eligible": true, "backbone_training_cutoff": "2024-06-30", "guard_days": 30, "experiment_as_of": "2026-08-17"} +{"task_id": "CF-HK-P11b", "category": "CF", "variant": "paired_counterfactual_v2", "time_band": "T1", "status": "ready", "cutoff_date": "2024-03-29", "seed": {"stock_code": "0700", "stock_name": "腾讯控股", "historical_price_series": [291.8, 284.8, 284.2, 288.8, 291.4, 290.8, 287.2, 284.6, 277.0, 277.2, 277.4, 276.2, 268.2, 274.6, 271.0, 270.0, 278.6, 290.8, 290.6, 289.2, 283.8, 289.8, 285.0, 288.8, 291.2, 288.8, 288.6, 299.4, 302.0, 303.8], "historical_trend_direction": "positive", "counterfactual_news": "监管要求行业采用更严格的数据合规标准。目标公司尚未改造且预计停业六个月，主要竞争者两年前已完成升级。", "expected_logic_direction": "negative", "driver_options": ["expectation_gap", "margin_surprise", "demand_surprise", "regulatory_severity", "unit_economics", "financing_cost", "dilution", "profit_contribution", "relationship_type", "relative_cost_hedge", "compliance_readiness", "supply_continuity", "deleveraging_vs_dilution", "retained_economics", "refinancing_spread"], "benchmark_id": "恒生指数", "cutoff_date": "2024-03-29", "market": "HK"}, "prompt": "以下是 腾讯控股 过去 30 个交易日的收盘价序列（从旧到新）：\n[291.8, 284.8, 284.2, 288.8, 291.4, 290.8, 287.2, 284.6, 277.0, 277.2, 277.4, 276.2, 268.2, 274.6, 271.0, 270.0, 278.6, 290.8, 290.6, 289.2, 283.8, 289.8, 285.0, 288.8, 291.2, 288.8, 288.6, 299.4, 302.0, 303.8]\n\n相对收益基准为：恒生指数。\n\n以下是一则完全虚构的公告（仅用于逻辑推理测试，公告中的数字和关系优先于公司真实历史）：\n「监管要求行业采用更严格的数据合规标准。目标公司尚未改造且预计停业六个月，主要竞争者两年前已完成升级。」\n\n假设上述事件真实发生，且历史上从未发生过类似事件。\n请比较公告中相互冲突的信息，识别真正决定短期定价的因素，预测该股票未来 5 个交易日相对基准的超额收益方向。\n不要仅根据“增长、获批、回购、处罚、召回”等单个关键词作答。\n同时从以下候选中选择唯一的决定性驱动因素代码：\nexpectation_gap / margin_surprise / demand_surprise / regulatory_severity / unit_economics / financing_cost / dilution / profit_contribution / relationship_type / relative_cost_hedge / compliance_readiness / supply_continuity / deleveraging_vs_dilution / retained_economics / refinancing_spread\n\n以下面 JSON 格式输出，不要附加任何其他内容：\n{\n \"direction\": \"positive\" 或 \"negative\",\n \"confidence\": <0.50 到 1.00 之间的浮点数>,\n \"driver\": \"<候选驱动因素代码>\",\n \"reasoning\": \"<一句话，说明核心判断依据>\"\n}\n\n请严格输出 JSON，不要输出任何额外文字、解释或 markdown 代码块。\n", "expected_output": {"direction": "positive|negative", "confidence": "float[0.50,1.00]", "driver": "enum", "reasoning": "str"}, "ground_truth": {"logic_direction": "negative", "driver_id": "compliance_readiness", "decisive_factor": "目标公司相对竞争者的合规准备度"}, "metadata": {"is_template": false, "prompt_template": "prompts/d_counterfactual.txt", "schema_version": "d_counterfactual_v2", "pair_id": "P11", "pair_side": "b", "family": "second_order", "driver_id": "compliance_readiness", "trend_conflict": true, "generation_source": "configs/d_event_archetypes_v2.json + configs/d_counterfactual_specs_v2.csv", "market": "HK", "temporal_split": {"time_band": "T1", "cutoff_date": "2024-03-29", "cutoff_date_source": "seed.cutoff_date", "is_synthetic_cutoff_date": false, "model_training_cutoff": "2024-06-30", "reference_current_date": "2026-08-17", "rule": "T1: cutoff_date <= model_training_cutoff; T2: model_training_cutoff < cutoff_date < reference_current_date; T3: cutoff_date >= reference_current_date"}}, "forecast_origin": "2024-03-29", "forecast_origin_source": "seed.cutoff_date", "outcome_available_at": "2024-03-29", "outcome_available_at_source": "synthetic_same_as_origin", "outcome_evidence_url": null, "outcome_evidence_code": "counterfactual_logic", "paper_band": "CF", "paper_band_reason": "category_D_counterfactual", "review_status": "draft", "review_method": null, "reviewed_at": null, "evidence_hash": null, "quality_flags": [], "official_temporal_eligible": true, "backbone_training_cutoff": "2024-06-30", "guard_days": 30, "experiment_as_of": "2026-08-17"} +{"task_id": "CF-HK-P12a", "category": "CF", "variant": "paired_counterfactual_v2", "time_band": "T1", "status": "ready", "cutoff_date": "2024-03-29", "seed": {"stock_code": "0005", "stock_name": "汇丰控股", "historical_price_series": [61.95, 62.4, 62.65, 60.25, 59.6, 59.25, 59.7, 59.65, 59.75, 60.25, 61.2, 60.65, 60.05, 60.5, 58.5, 58.8, 58.35, 58.75, 59.3, 59.05, 58.55, 59.6, 59.5, 59.6, 60.9, 60.95, 61.2, 61.75, 61.1, 61.15], "historical_trend_direction": "flat", "counterfactual_news": "占目标公司采购额35%的上游供应商破产，但目标公司持有其关键专利的永久许可，且两家备用供应商已完成认证并承诺原价供货。", "expected_logic_direction": "positive", "driver_options": ["expectation_gap", "margin_surprise", "demand_surprise", "regulatory_severity", "unit_economics", "financing_cost", "dilution", "profit_contribution", "relationship_type", "relative_cost_hedge", "compliance_readiness", "supply_continuity", "deleveraging_vs_dilution", "retained_economics", "refinancing_spread"], "benchmark_id": "恒生指数", "cutoff_date": "2024-03-29", "market": "HK"}, "prompt": "以下是 汇丰控股 过去 30 个交易日的收盘价序列（从旧到新）：\n[61.95, 62.4, 62.65, 60.25, 59.6, 59.25, 59.7, 59.65, 59.75, 60.25, 61.2, 60.65, 60.05, 60.5, 58.5, 58.8, 58.35, 58.75, 59.3, 59.05, 58.55, 59.6, 59.5, 59.6, 60.9, 60.95, 61.2, 61.75, 61.1, 61.15]\n\n相对收益基准为：恒生指数。\n\n以下是一则完全虚构的公告（仅用于逻辑推理测试，公告中的数字和关系优先于公司真实历史）：\n「占目标公司采购额35%的上游供应商破产，但目标公司持有其关键专利的永久许可，且两家备用供应商已完成认证并承诺原价供货。」\n\n假设上述事件真实发生，且历史上从未发生过类似事件。\n请比较公告中相互冲突的信息，识别真正决定短期定价的因素，预测该股票未来 5 个交易日相对基准的超额收益方向。\n不要仅根据“增长、获批、回购、处罚、召回”等单个关键词作答。\n同时从以下候选中选择唯一的决定性驱动因素代码：\nexpectation_gap / margin_surprise / demand_surprise / regulatory_severity / unit_economics / financing_cost / dilution / profit_contribution / relationship_type / relative_cost_hedge / compliance_readiness / supply_continuity / deleveraging_vs_dilution / retained_economics / refinancing_spread\n\n以下面 JSON 格式输出，不要附加任何其他内容：\n{\n \"direction\": \"positive\" 或 \"negative\",\n \"confidence\": <0.50 到 1.00 之间的浮点数>,\n \"driver\": \"<候选驱动因素代码>\",\n \"reasoning\": \"<一句话，说明核心判断依据>\"\n}\n\n请严格输出 JSON，不要输出任何额外文字、解释或 markdown 代码块。\n", "expected_output": {"direction": "positive|negative", "confidence": "float[0.50,1.00]", "driver": "enum", "reasoning": "str"}, "ground_truth": {"logic_direction": "positive", "driver_id": "supply_continuity", "decisive_factor": "替代供应和知识产权连续性"}, "metadata": {"is_template": false, "prompt_template": "prompts/d_counterfactual.txt", "schema_version": "d_counterfactual_v2", "pair_id": "P12", "pair_side": "a", "family": "second_order", "driver_id": "supply_continuity", "trend_conflict": false, "generation_source": "configs/d_event_archetypes_v2.json + configs/d_counterfactual_specs_v2.csv", "market": "HK", "temporal_split": {"time_band": "T1", "cutoff_date": "2024-03-29", "cutoff_date_source": "seed.cutoff_date", "is_synthetic_cutoff_date": false, "model_training_cutoff": "2024-06-30", "reference_current_date": "2026-08-17", "rule": "T1: cutoff_date <= model_training_cutoff; T2: model_training_cutoff < cutoff_date < reference_current_date; T3: cutoff_date >= reference_current_date"}}, "forecast_origin": "2024-03-29", "forecast_origin_source": "seed.cutoff_date", "outcome_available_at": "2024-03-29", "outcome_available_at_source": "synthetic_same_as_origin", "outcome_evidence_url": null, "outcome_evidence_code": "counterfactual_logic", "paper_band": "CF", "paper_band_reason": "category_D_counterfactual", "review_status": "draft", "review_method": null, "reviewed_at": null, "evidence_hash": null, "quality_flags": [], "official_temporal_eligible": true, "backbone_training_cutoff": "2024-06-30", "guard_days": 30, "experiment_as_of": "2026-08-17"} +{"task_id": "CF-HK-P12b", "category": "CF", "variant": "paired_counterfactual_v2", "time_band": "T1", "status": "ready", "cutoff_date": "2024-03-29", "seed": {"stock_code": "0005", "stock_name": "汇丰控股", "historical_price_series": [61.95, 62.4, 62.65, 60.25, 59.6, 59.25, 59.7, 59.65, 59.75, 60.25, 61.2, 60.65, 60.05, 60.5, 58.5, 58.8, 58.35, 58.75, 59.3, 59.05, 58.55, 59.6, 59.5, 59.6, 60.9, 60.95, 61.2, 61.75, 61.1, 61.15], "historical_trend_direction": "flat", "counterfactual_news": "占目标公司采购额35%的上游供应商破产，关键专利许可随破产终止；备用供应商认证至少需要九个月且报价高60%。", "expected_logic_direction": "negative", "driver_options": ["expectation_gap", "margin_surprise", "demand_surprise", "regulatory_severity", "unit_economics", "financing_cost", "dilution", "profit_contribution", "relationship_type", "relative_cost_hedge", "compliance_readiness", "supply_continuity", "deleveraging_vs_dilution", "retained_economics", "refinancing_spread"], "benchmark_id": "恒生指数", "cutoff_date": "2024-03-29", "market": "HK"}, "prompt": "以下是 汇丰控股 过去 30 个交易日的收盘价序列（从旧到新）：\n[61.95, 62.4, 62.65, 60.25, 59.6, 59.25, 59.7, 59.65, 59.75, 60.25, 61.2, 60.65, 60.05, 60.5, 58.5, 58.8, 58.35, 58.75, 59.3, 59.05, 58.55, 59.6, 59.5, 59.6, 60.9, 60.95, 61.2, 61.75, 61.1, 61.15]\n\n相对收益基准为：恒生指数。\n\n以下是一则完全虚构的公告（仅用于逻辑推理测试，公告中的数字和关系优先于公司真实历史）：\n「占目标公司采购额35%的上游供应商破产，关键专利许可随破产终止；备用供应商认证至少需要九个月且报价高60%。」\n\n假设上述事件真实发生，且历史上从未发生过类似事件。\n请比较公告中相互冲突的信息，识别真正决定短期定价的因素，预测该股票未来 5 个交易日相对基准的超额收益方向。\n不要仅根据“增长、获批、回购、处罚、召回”等单个关键词作答。\n同时从以下候选中选择唯一的决定性驱动因素代码：\nexpectation_gap / margin_surprise / demand_surprise / regulatory_severity / unit_economics / financing_cost / dilution / profit_contribution / relationship_type / relative_cost_hedge / compliance_readiness / supply_continuity / deleveraging_vs_dilution / retained_economics / refinancing_spread\n\n以下面 JSON 格式输出，不要附加任何其他内容：\n{\n \"direction\": \"positive\" 或 \"negative\",\n \"confidence\": <0.50 到 1.00 之间的浮点数>,\n \"driver\": \"<候选驱动因素代码>\",\n \"reasoning\": \"<一句话，说明核心判断依据>\"\n}\n\n请严格输出 JSON，不要输出任何额外文字、解释或 markdown 代码块。\n", "expected_output": {"direction": "positive|negative", "confidence": "float[0.50,1.00]", "driver": "enum", "reasoning": "str"}, "ground_truth": {"logic_direction": "negative", "driver_id": "supply_continuity", "decisive_factor": "替代供应和知识产权连续性"}, "metadata": {"is_template": false, "prompt_template": "prompts/d_counterfactual.txt", "schema_version": "d_counterfactual_v2", "pair_id": "P12", "pair_side": "b", "family": "second_order", "driver_id": "supply_continuity", "trend_conflict": false, "generation_source": "configs/d_event_archetypes_v2.json + configs/d_counterfactual_specs_v2.csv", "market": "HK", "temporal_split": {"time_band": "T1", "cutoff_date": "2024-03-29", "cutoff_date_source": "seed.cutoff_date", "is_synthetic_cutoff_date": false, "model_training_cutoff": "2024-06-30", "reference_current_date": "2026-08-17", "rule": "T1: cutoff_date <= model_training_cutoff; T2: model_training_cutoff < cutoff_date < reference_current_date; T3: cutoff_date >= reference_current_date"}}, "forecast_origin": "2024-03-29", "forecast_origin_source": "seed.cutoff_date", "outcome_available_at": "2024-03-29", "outcome_available_at_source": "synthetic_same_as_origin", "outcome_evidence_url": null, "outcome_evidence_code": "counterfactual_logic", "paper_band": "CF", "paper_band_reason": "category_D_counterfactual", "review_status": "draft", "review_method": null, "reviewed_at": null, "evidence_hash": null, "quality_flags": [], "official_temporal_eligible": true, "backbone_training_cutoff": "2024-06-30", "guard_days": 30, "experiment_as_of": "2026-08-17"} +{"task_id": "CF-HK-P13a", "category": "CF", "variant": "paired_counterfactual_v2", "time_band": "T1", "status": "ready", "cutoff_date": "2024-03-29", "seed": {"stock_code": "9988", "stock_name": "阿里巴巴", "historical_price_series": [73.0, 71.5, 72.15, 72.9, 74.85, 74.55, 74.15, 75.5, 74.25, 72.9, 73.05, 72.1, 69.7, 71.85, 70.7, 71.25, 72.8, 74.85, 74.0, 73.45, 71.9, 72.0, 71.1, 71.25, 72.55, 71.0, 70.3, 70.3, 68.8, 70.25], "historical_trend_direction": "negative", "counterfactual_news": "公司可转债持有人同意按较现价高35%的转股价转股，转股后有息负债下降40%，预计利息费用减少额是股份摊薄影响的三倍。", "expected_logic_direction": "positive", "driver_options": ["expectation_gap", "margin_surprise", "demand_surprise", "regulatory_severity", "unit_economics", "financing_cost", "dilution", "profit_contribution", "relationship_type", "relative_cost_hedge", "compliance_readiness", "supply_continuity", "deleveraging_vs_dilution", "retained_economics", "refinancing_spread"], "benchmark_id": "恒生指数", "cutoff_date": "2024-03-29", "market": "HK"}, "prompt": "以下是 阿里巴巴 过去 30 个交易日的收盘价序列（从旧到新）：\n[73.0, 71.5, 72.15, 72.9, 74.85, 74.55, 74.15, 75.5, 74.25, 72.9, 73.05, 72.1, 69.7, 71.85, 70.7, 71.25, 72.8, 74.85, 74.0, 73.45, 71.9, 72.0, 71.1, 71.25, 72.55, 71.0, 70.3, 70.3, 68.8, 70.25]\n\n相对收益基准为：恒生指数。\n\n以下是一则完全虚构的公告（仅用于逻辑推理测试，公告中的数字和关系优先于公司真实历史）：\n「公司可转债持有人同意按较现价高35%的转股价转股，转股后有息负债下降40%，预计利息费用减少额是股份摊薄影响的三倍。」\n\n假设上述事件真实发生，且历史上从未发生过类似事件。\n请比较公告中相互冲突的信息，识别真正决定短期定价的因素，预测该股票未来 5 个交易日相对基准的超额收益方向。\n不要仅根据“增长、获批、回购、处罚、召回”等单个关键词作答。\n同时从以下候选中选择唯一的决定性驱动因素代码：\nexpectation_gap / margin_surprise / demand_surprise / regulatory_severity / unit_economics / financing_cost / dilution / profit_contribution / relationship_type / relative_cost_hedge / compliance_readiness / supply_continuity / deleveraging_vs_dilution / retained_economics / refinancing_spread\n\n以下面 JSON 格式输出，不要附加任何其他内容：\n{\n \"direction\": \"positive\" 或 \"negative\",\n \"confidence\": <0.50 到 1.00 之间的浮点数>,\n \"driver\": \"<候选驱动因素代码>\",\n \"reasoning\": \"<一句话，说明核心判断依据>\"\n}\n\n请严格输出 JSON，不要输出任何额外文字、解释或 markdown 代码块。\n", "expected_output": {"direction": "positive|negative", "confidence": "float[0.50,1.00]", "driver": "enum", "reasoning": "str"}, "ground_truth": {"logic_direction": "positive", "driver_id": "deleveraging_vs_dilution", "decisive_factor": "去杠杆收益相对股权摊薄"}, "metadata": {"is_template": false, "prompt_template": "prompts/d_counterfactual.txt", "schema_version": "d_counterfactual_v2", "pair_id": "P13", "pair_side": "a", "family": "capital_structure", "driver_id": "deleveraging_vs_dilution", "trend_conflict": true, "generation_source": "configs/d_event_archetypes_v2.json + configs/d_counterfactual_specs_v2.csv", "market": "HK", "temporal_split": {"time_band": "T1", "cutoff_date": "2024-03-29", "cutoff_date_source": "seed.cutoff_date", "is_synthetic_cutoff_date": false, "model_training_cutoff": "2024-06-30", "reference_current_date": "2026-08-17", "rule": "T1: cutoff_date <= model_training_cutoff; T2: model_training_cutoff < cutoff_date < reference_current_date; T3: cutoff_date >= reference_current_date"}}, "forecast_origin": "2024-03-29", "forecast_origin_source": "seed.cutoff_date", "outcome_available_at": "2024-03-29", "outcome_available_at_source": "synthetic_same_as_origin", "outcome_evidence_url": null, "outcome_evidence_code": "counterfactual_logic", "paper_band": "CF", "paper_band_reason": "category_D_counterfactual", "review_status": "draft", "review_method": null, "reviewed_at": null, "evidence_hash": null, "quality_flags": [], "official_temporal_eligible": true, "backbone_training_cutoff": "2024-06-30", "guard_days": 30, "experiment_as_of": "2026-08-17"} +{"task_id": "CF-HK-P13b", "category": "CF", "variant": "paired_counterfactual_v2", "time_band": "T1", "status": "ready", "cutoff_date": "2024-03-29", "seed": {"stock_code": "9988", "stock_name": "阿里巴巴", "historical_price_series": [73.0, 71.5, 72.15, 72.9, 74.85, 74.55, 74.15, 75.5, 74.25, 72.9, 73.05, 72.1, 69.7, 71.85, 70.7, 71.25, 72.8, 74.85, 74.0, 73.45, 71.9, 72.0, 71.1, 71.25, 72.55, 71.0, 70.3, 70.3, 68.8, 70.25], "historical_trend_direction": "negative", "counterfactual_news": "公司把可转债转股价下修至较现价低30%，若全部转股将新增70%的股份，而利息费用减少额仅相当于去年净利润的2%。", "expected_logic_direction": "negative", "driver_options": ["expectation_gap", "margin_surprise", "demand_surprise", "regulatory_severity", "unit_economics", "financing_cost", "dilution", "profit_contribution", "relationship_type", "relative_cost_hedge", "compliance_readiness", "supply_continuity", "deleveraging_vs_dilution", "retained_economics", "refinancing_spread"], "benchmark_id": "恒生指数", "cutoff_date": "2024-03-29", "market": "HK"}, "prompt": "以下是 阿里巴巴 过去 30 个交易日的收盘价序列（从旧到新）：\n[73.0, 71.5, 72.15, 72.9, 74.85, 74.55, 74.15, 75.5, 74.25, 72.9, 73.05, 72.1, 69.7, 71.85, 70.7, 71.25, 72.8, 74.85, 74.0, 73.45, 71.9, 72.0, 71.1, 71.25, 72.55, 71.0, 70.3, 70.3, 68.8, 70.25]\n\n相对收益基准为：恒生指数。\n\n以下是一则完全虚构的公告（仅用于逻辑推理测试，公告中的数字和关系优先于公司真实历史）：\n「公司把可转债转股价下修至较现价低30%，若全部转股将新增70%的股份，而利息费用减少额仅相当于去年净利润的2%。」\n\n假设上述事件真实发生，且历史上从未发生过类似事件。\n请比较公告中相互冲突的信息，识别真正决定短期定价的因素，预测该股票未来 5 个交易日相对基准的超额收益方向。\n不要仅根据“增长、获批、回购、处罚、召回”等单个关键词作答。\n同时从以下候选中选择唯一的决定性驱动因素代码：\nexpectation_gap / margin_surprise / demand_surprise / regulatory_severity / unit_economics / financing_cost / dilution / profit_contribution / relationship_type / relative_cost_hedge / compliance_readiness / supply_continuity / deleveraging_vs_dilution / retained_economics / refinancing_spread\n\n以下面 JSON 格式输出，不要附加任何其他内容：\n{\n \"direction\": \"positive\" 或 \"negative\",\n \"confidence\": <0.50 到 1.00 之间的浮点数>,\n \"driver\": \"<候选驱动因素代码>\",\n \"reasoning\": \"<一句话，说明核心判断依据>\"\n}\n\n请严格输出 JSON，不要输出任何额外文字、解释或 markdown 代码块。\n", "expected_output": {"direction": "positive|negative", "confidence": "float[0.50,1.00]", "driver": "enum", "reasoning": "str"}, "ground_truth": {"logic_direction": "negative", "driver_id": "deleveraging_vs_dilution", "decisive_factor": "去杠杆收益相对股权摊薄"}, "metadata": {"is_template": false, "prompt_template": "prompts/d_counterfactual.txt", "schema_version": "d_counterfactual_v2", "pair_id": "P13", "pair_side": "b", "family": "capital_structure", "driver_id": "deleveraging_vs_dilution", "trend_conflict": false, "generation_source": "configs/d_event_archetypes_v2.json + configs/d_counterfactual_specs_v2.csv", "market": "HK", "temporal_split": {"time_band": "T1", "cutoff_date": "2024-03-29", "cutoff_date_source": "seed.cutoff_date", "is_synthetic_cutoff_date": false, "model_training_cutoff": "2024-06-30", "reference_current_date": "2026-08-17", "rule": "T1: cutoff_date <= model_training_cutoff; T2: model_training_cutoff < cutoff_date < reference_current_date; T3: cutoff_date >= reference_current_date"}}, "forecast_origin": "2024-03-29", "forecast_origin_source": "seed.cutoff_date", "outcome_available_at": "2024-03-29", "outcome_available_at_source": "synthetic_same_as_origin", "outcome_evidence_url": null, "outcome_evidence_code": "counterfactual_logic", "paper_band": "CF", "paper_band_reason": "category_D_counterfactual", "review_status": "draft", "review_method": null, "reviewed_at": null, "evidence_hash": null, "quality_flags": [], "official_temporal_eligible": true, "backbone_training_cutoff": "2024-06-30", "guard_days": 30, "experiment_as_of": "2026-08-17"} +{"task_id": "CF-HK-P14a", "category": "CF", "variant": "paired_counterfactual_v2", "time_band": "T2", "status": "ready", "cutoff_date": "2025-06-30", "seed": {"stock_code": "0941", "stock_name": "中国移动", "historical_price_series": [88.1, 88.1, 86.9, 86.7, 88.2, 88.75, 88.85, 88.95, 88.0, 88.05, 89.1, 88.85, 89.2, 87.9, 87.6, 87.5, 88.3, 87.35, 87.9, 87.4, 87.05, 86.75, 85.8, 86.4, 86.1, 87.0, 87.4, 86.9, 86.35, 87.1], "historical_trend_direction": "flat", "counterfactual_news": "公司拟分拆子公司上市，保留80%股权；基石投资者认购估值较母公司账面估值高一倍，募集资金全部归子公司扩产。", "expected_logic_direction": "positive", "driver_options": ["expectation_gap", "margin_surprise", "demand_surprise", "regulatory_severity", "unit_economics", "financing_cost", "dilution", "profit_contribution", "relationship_type", "relative_cost_hedge", "compliance_readiness", "supply_continuity", "deleveraging_vs_dilution", "retained_economics", "refinancing_spread"], "benchmark_id": "恒生指数", "cutoff_date": "2025-06-30", "market": "HK"}, "prompt": "以下是 中国移动 过去 30 个交易日的收盘价序列（从旧到新）：\n[88.1, 88.1, 86.9, 86.7, 88.2, 88.75, 88.85, 88.95, 88.0, 88.05, 89.1, 88.85, 89.2, 87.9, 87.6, 87.5, 88.3, 87.35, 87.9, 87.4, 87.05, 86.75, 85.8, 86.4, 86.1, 87.0, 87.4, 86.9, 86.35, 87.1]\n\n相对收益基准为：恒生指数。\n\n以下是一则完全虚构的公告（仅用于逻辑推理测试，公告中的数字和关系优先于公司真实历史）：\n「公司拟分拆子公司上市，保留80%股权；基石��资者认购估值较母公司账面估值高一倍，募集资金全部归子公司扩产。」\n\n假设上述事件真实发生，且历史上从未发生过类似事件。\n请比较公告中相互冲突的信息，识别真正决定短期定价的因素，预测该股票未来 5 个交易日相对基准的超额收益方向。\n不要仅根据“增长、获批、回购、处罚、召回”等单个关键词作答。\n同时从以下候选中选择唯一的决定性驱动因素代码：\nexpectation_gap / margin_surprise / demand_surprise / regulatory_severity / unit_economics / financing_cost / dilution / profit_contribution / relationship_type / relative_cost_hedge / compliance_readiness / supply_continuity / deleveraging_vs_dilution / retained_economics / refinancing_spread\n\n以下面 JSON 格式输出，不要附加任何其他内容：\n{\n \"direction\": \"positive\" 或 \"negative\",\n \"confidence\": <0.50 到 1.00 之间的浮点数>,\n \"driver\": \"<候选驱动因素代码>\",\n \"reasoning\": \"<一句话，说明核心判断依据>\"\n}\n\n请严格输出 JSON，不要输出任何额外文字、解释或 markdown 代码块。\n", "expected_output": {"direction": "positive|negative", "confidence": "float[0.50,1.00]", "driver": "enum", "reasoning": "str"}, "ground_truth": {"logic_direction": "positive", "driver_id": "retained_economics", "decisive_factor": "分拆估值、保留权益和核心技术归属"}, "metadata": {"is_template": false, "prompt_template": "prompts/d_counterfactual.txt", "schema_version": "d_counterfactual_v2", "pair_id": "P14", "pair_side": "a", "family": "capital_structure", "driver_id": "retained_economics", "trend_conflict": false, "generation_source": "configs/d_event_archetypes_v2.json + configs/d_counterfactual_specs_v2.csv", "market": "HK", "temporal_split": {"time_band": "T2", "cutoff_date": "2025-06-30", "cutoff_date_source": "seed.cutoff_date", "is_synthetic_cutoff_date": false, "model_training_cutoff": "2024-06-30", "reference_current_date": "2026-08-17", "rule": "T1: cutoff_date <= model_training_cutoff; T2: model_training_cutoff < cutoff_date < reference_current_date; T3: cutoff_date >= reference_current_date"}}, "forecast_origin": "2025-06-30", "forecast_origin_source": "seed.cutoff_date", "outcome_available_at": "2025-06-30", "outcome_available_at_source": "synthetic_same_as_origin", "outcome_evidence_url": null, "outcome_evidence_code": "counterfactual_logic", "paper_band": "CF", "paper_band_reason": "category_D_counterfactual", "review_status": "draft", "review_method": null, "reviewed_at": null, "evidence_hash": null, "quality_flags": [], "official_temporal_eligible": true, "backbone_training_cutoff": "2024-06-30", "guard_days": 30, "experiment_as_of": "2026-08-17"} +{"task_id": "CF-HK-P14b", "category": "CF", "variant": "paired_counterfactual_v2", "time_band": "T2", "status": "ready", "cutoff_date": "2025-06-30", "seed": {"stock_code": "0941", "stock_name": "中国移动", "historical_price_series": [88.1, 88.1, 86.9, 86.7, 88.2, 88.75, 88.85, 88.95, 88.0, 88.05, 89.1, 88.85, 89.2, 87.9, 87.6, 87.5, 88.3, 87.35, 87.9, 87.4, 87.05, 86.75, 85.8, 86.4, 86.1, 87.0, 87.4, 86.9, 86.35, 87.1], "historical_trend_direction": "flat", "counterfactual_news": "公司拟出售子公司80%股权以偿还即将违约的债务，成交估值较账面价值低45%，且母公司失去核心技术独家使用权。", "expected_logic_direction": "negative", "driver_options": ["expectation_gap", "margin_surprise", "demand_surprise", "regulatory_severity", "unit_economics", "financing_cost", "dilution", "profit_contribution", "relationship_type", "relative_cost_hedge", "compliance_readiness", "supply_continuity", "deleveraging_vs_dilution", "retained_economics", "refinancing_spread"], "benchmark_id": "恒生指数", "cutoff_date": "2025-06-30", "market": "HK"}, "prompt": "以下是 中国移动 过去 30 个交易日的收盘价序列（从旧到新）：\n[88.1, 88.1, 86.9, 86.7, 88.2, 88.75, 88.85, 88.95, 88.0, 88.05, 89.1, 88.85, 89.2, 87.9, 87.6, 87.5, 88.3, 87.35, 87.9, 87.4, 87.05, 86.75, 85.8, 86.4, 86.1, 87.0, 87.4, 86.9, 86.35, 87.1]\n\n相对收益基准为：恒生指数。\n\n以下是一则完全虚构的公告（仅用于逻辑推理测试，公告中的数字和关系优先于公司真实历史）：\n「公司拟出售子公司80%股权以偿还即将违约的债务，成交估值较账面价值低45%，且母公司失去核心技术独家使用权。」\n\n假设上述事件真实发生，且历史上从未发生过类似事件。\n请比较公告中相互冲突的信息，识别真正决定短期定价的因素，预测该股票未来 5 个交易日相对基准的超额收益方向。\n不要仅根据“增长、获批、回购、处罚、召回”等单个关键词作答。\n同时从以下候选中选择唯一的决定性驱动因素代码：\nexpectation_gap / margin_surprise / demand_surprise / regulatory_severity / unit_economics / financing_cost / dilution / profit_contribution / relationship_type / relative_cost_hedge / compliance_readiness / supply_continuity / deleveraging_vs_dilution / retained_economics / refinancing_spread\n\n以下面 JSON 格式输出，不要附加任何其他内容：\n{\n \"direction\": \"positive\" 或 \"negative\",\n \"confidence\": <0.50 到 1.00 之间的浮点数>,\n \"driver\": \"<候选驱动因素代码>\",\n \"reasoning\": \"<一句话，说明核心判断依据>\"\n}\n\n请严格输出 JSON，不要输出任何额外文字、解释或 markdown 代码块。\n", "expected_output": {"direction": "positive|negative", "confidence": "float[0.50,1.00]", "driver": "enum", "reasoning": "str"}, "ground_truth": {"logic_direction": "negative", "driver_id": "retained_economics", "decisive_factor": "分拆估值、保留权益和核心技术归属"}, "metadata": {"is_template": false, "prompt_template": "prompts/d_counterfactual.txt", "schema_version": "d_counterfactual_v2", "pair_id": "P14", "pair_side": "b", "family": "capital_structure", "driver_id": "retained_economics", "trend_conflict": false, "generation_source": "configs/d_event_archetypes_v2.json + configs/d_counterfactual_specs_v2.csv", "market": "HK", "temporal_split": {"time_band": "T2", "cutoff_date": "2025-06-30", "cutoff_date_source": "seed.cutoff_date", "is_synthetic_cutoff_date": false, "model_training_cutoff": "2024-06-30", "reference_current_date": "2026-08-17", "rule": "T1: cutoff_date <= model_training_cutoff; T2: model_training_cutoff < cutoff_date < reference_current_date; T3: cutoff_date >= reference_current_date"}}, "forecast_origin": "2025-06-30", "forecast_origin_source": "seed.cutoff_date", "outcome_available_at": "2025-06-30", "outcome_available_at_source": "synthetic_same_as_origin", "outcome_evidence_url": null, "outcome_evidence_code": "counterfactual_logic", "paper_band": "CF", "paper_band_reason": "category_D_counterfactual", "review_status": "draft", "review_method": null, "reviewed_at": null, "evidence_hash": null, "quality_flags": [], "official_temporal_eligible": true, "backbone_training_cutoff": "2024-06-30", "guard_days": 30, "experiment_as_of": "2026-08-17"} +{"task_id": "CF-HK-P15a", "category": "CF", "variant": "paired_counterfactual_v2", "time_band": "T2", "status": "ready", "cutoff_date": "2025-06-30", "seed": {"stock_code": "1299", "stock_name": "友邦保险", "historical_price_series": [66.85, 65.95, 65.35, 64.85, 64.45, 65.2, 63.45, 65.0, 65.65, 65.7, 67.0, 66.05, 66.85, 67.05, 68.95, 69.7, 70.15, 68.9, 69.15, 68.75, 68.95, 68.1, 66.75, 67.85, 68.45, 70.85, 71.6, 71.0, 71.6, 70.4], "historical_trend_direction": "positive", "counterfactual_news": "公司以固定利率4%置换一年内到期的浮息债务，期限延长五年且无新增抵押；当前同评级市场融资利率为7%。", "expected_logic_direction": "positive", "driver_options": ["expectation_gap", "margin_surprise", "demand_surprise", "regulatory_severity", "unit_economics", "financing_cost", "dilution", "profit_contribution", "relationship_type", "relative_cost_hedge", "compliance_readiness", "supply_continuity", "deleveraging_vs_dilution", "retained_economics", "refinancing_spread"], "benchmark_id": "恒生指数", "cutoff_date": "2025-06-30", "market": "HK"}, "prompt": "以下是 友邦保险 过去 30 个交易日的收盘价序列（从旧到新）：\n[66.85, 65.95, 65.35, 64.85, 64.45, 65.2, 63.45, 65.0, 65.65, 65.7, 67.0, 66.05, 66.85, 67.05, 68.95, 69.7, 70.15, 68.9, 69.15, 68.75, 68.95, 68.1, 66.75, 67.85, 68.45, 70.85, 71.6, 71.0, 71.6, 70.4]\n\n相对收益基准为：恒生指数。\n\n以下是一则完全虚构的公告（仅用于逻辑推理测试，公告中的数字和关系优先于公司真实历史）：\n「公司以固定利率4%置换一年内到期的浮息债务，期限延长五年且无新增抵押；当前同评级市场融资利率为7%。」\n\n假设上述事件真实发生，且历史上从未发生过类似事件。\n请比较公告中相互冲突的信息，识别真正决定短期定价的因素，预测该股票未来 5 个交易日相对基准的超额收益方向。\n不要仅根据“增长、获批、回购、处罚、召回”等单个关键词作答。\n同时从以下候选中选择唯一的决定性驱动因素代码：\nexpectation_gap / margin_surprise / demand_surprise / regulatory_severity / unit_economics / financing_cost / dilution / profit_contribution / relationship_type / relative_cost_hedge / compliance_readiness / supply_continuity / deleveraging_vs_dilution / retained_economics / refinancing_spread\n\n以下面 JSON 格式输出，不要附加任何其他内容：\n{\n \"direction\": \"positive\" 或 \"negative\",\n \"confidence\": <0.50 到 1.00 之间的浮点数>,\n \"driver\": \"<候选驱动因素代码>\",\n \"reasoning\": \"<一句话，说明核心判断依据>\"\n}\n\n请严格输出 JSON，不要输出任何额外文字、解释或 markdown 代码块。\n", "expected_output": {"direction": "positive|negative", "confidence": "float[0.50,1.00]", "driver": "enum", "reasoning": "str"}, "ground_truth": {"logic_direction": "positive", "driver_id": "refinancing_spread", "decisive_factor": "再融资成本和附加担保��款"}, "metadata": {"is_template": false, "prompt_template": "prompts/d_counterfactual.txt", "schema_version": "d_counterfactual_v2", "pair_id": "P15", "pair_side": "a", "family": "capital_structure", "driver_id": "refinancing_spread", "trend_conflict": false, "generation_source": "configs/d_event_archetypes_v2.json + configs/d_counterfactual_specs_v2.csv", "market": "HK", "temporal_split": {"time_band": "T2", "cutoff_date": "2025-06-30", "cutoff_date_source": "seed.cutoff_date", "is_synthetic_cutoff_date": false, "model_training_cutoff": "2024-06-30", "reference_current_date": "2026-08-17", "rule": "T1: cutoff_date <= model_training_cutoff; T2: model_training_cutoff < cutoff_date < reference_current_date; T3: cutoff_date >= reference_current_date"}}, "forecast_origin": "2025-06-30", "forecast_origin_source": "seed.cutoff_date", "outcome_available_at": "2025-06-30", "outcome_available_at_source": "synthetic_same_as_origin", "outcome_evidence_url": null, "outcome_evidence_code": "counterfactual_logic", "paper_band": "CF", "paper_band_reason": "category_D_counterfactual", "review_status": "draft", "review_method": null, "reviewed_at": null, "evidence_hash": null, "quality_flags": [], "official_temporal_eligible": true, "backbone_training_cutoff": "2024-06-30", "guard_days": 30, "experiment_as_of": "2026-08-17"} +{"task_id": "CF-HK-P15b", "category": "CF", "variant": "paired_counterfactual_v2", "time_band": "T2", "status": "ready", "cutoff_date": "2025-06-30", "seed": {"stock_code": "1299", "stock_name": "友邦保险", "historical_price_series": [66.85, 65.95, 65.35, 64.85, 64.45, 65.2, 63.45, 65.0, 65.65, 65.7, 67.0, 66.05, 66.85, 67.05, 68.95, 69.7, 70.15, 68.9, 69.15, 68.75, 68.95, 68.1, 66.75, 67.85, 68.45, 70.85, 71.6, 71.0, 71.6, 70.4], "historical_trend_direction": "positive", "counterfactual_news": "公司以固定利率12%置换一年内到期的浮息债务，新增核心资产抵押和提前还款罚金；当前同评级市场融资利率为7%。", "expected_logic_direction": "negative", "driver_options": ["expectation_gap", "margin_surprise", "demand_surprise", "regulatory_severity", "unit_economics", "financing_cost", "dilution", "profit_contribution", "relationship_type", "relative_cost_hedge", "compliance_readiness", "supply_continuity", "deleveraging_vs_dilution", "retained_economics", "refinancing_spread"], "benchmark_id": "恒生指数", "cutoff_date": "2025-06-30", "market": "HK"}, "prompt": "以下是 友邦保险 过去 30 个交易日的收盘价序列（从旧到新）：\n[66.85, 65.95, 65.35, 64.85, 64.45, 65.2, 63.45, 65.0, 65.65, 65.7, 67.0, 66.05, 66.85, 67.05, 68.95, 69.7, 70.15, 68.9, 69.15, 68.75, 68.95, 68.1, 66.75, 67.85, 68.45, 70.85, 71.6, 71.0, 71.6, 70.4]\n\n相对收益基准为：恒生指数。\n\n以下是一则完全虚构的公告（仅用于逻辑推理测试，公告中的数字和关系优先于公司真实历史）：\n「公司以固定利率12%置换一年内到期的浮息债务，新增核心资产抵押和提前还款罚金；当前同评级市场融资利率为7%。」\n\n假设上述事件真实发生，且历史上从未发生过类似事件。\n请比较公告中相互冲突的信息，识别真正决定短期定价的因素，预测该股票未来 5 个交易日相对基准的超额收益方向。\n不要仅根据“增长、获批、回购、处罚、召回”等单个关键词作答。\n同时从以下候选中选择唯一的决定性驱动因素代码：\nexpectation_gap / margin_surprise / demand_surprise / regulatory_severity / unit_economics / financing_cost / dilution / profit_contribution / relationship_type / relative_cost_hedge / compliance_readiness / supply_continuity / deleveraging_vs_dilution / retained_economics / refinancing_spread\n\n以下面 JSON 格式输出，不要附加任何其他内容：\n{\n \"direction\": \"positive\" 或 \"negative\",\n \"confidence\": <0.50 到 1.00 之间的浮点数>,\n \"driver\": \"<候选驱动因素代码>\",\n \"reasoning\": \"<一句话，说明核心判断依据>\"\n}\n\n请严格输出 JSON，不要输出任何额外文字、解释或 markdown 代码块。\n", "expected_output": {"direction": "positive|negative", "confidence": "float[0.50,1.00]", "driver": "enum", "reasoning": "str"}, "ground_truth": {"logic_direction": "negative", "driver_id": "refinancing_spread", "decisive_factor": "再融资成本和附加担保条款"}, "metadata": {"is_template": false, "prompt_template": "prompts/d_counterfactual.txt", "schema_version": "d_counterfactual_v2", "pair_id": "P15", "pair_side": "b", "family": "capital_structure", "driver_id": "refinancing_spread", "trend_conflict": true, "generation_source": "configs/d_event_archetypes_v2.json + configs/d_counterfactual_specs_v2.csv", "market": "HK", "temporal_split": {"time_band": "T2", "cutoff_date": "2025-06-30", "cutoff_date_source": "seed.cutoff_date", "is_synthetic_cutoff_date": false, "model_training_cutoff": "2024-06-30", "reference_current_date": "2026-08-17", "rule": "T1: cutoff_date <= model_training_cutoff; T2: model_training_cutoff < cutoff_date < reference_current_date; T3: cutoff_date >= reference_current_date"}}, "forecast_origin": "2025-06-30", "forecast_origin_source": "seed.cutoff_date", "outcome_available_at": "2025-06-30", "outcome_available_at_source": "synthetic_same_as_origin", "outcome_evidence_url": null, "outcome_evidence_code": "counterfactual_logic", "paper_band": "CF", "paper_band_reason": "category_D_counterfactual", "review_status": "draft", "review_method": null, "reviewed_at": null, "evidence_hash": null, "quality_flags": [], "official_temporal_eligible": true, "backbone_training_cutoff": "2024-06-30", "guard_days": 30, "experiment_as_of": "2026-08-17"}