148 lines
6.8 KiB
JSON
148 lines
6.8 KiB
JSON
{
|
||
"calibration_version": "0.3-draft",
|
||
"status": "draft_owner_review_needed",
|
||
"notes": "These inputs calibrate selector behavior only. They are not model regression cases and do not replace owner-supplied raw QPI case materials.",
|
||
"inputs": [
|
||
{
|
||
"case_id": "selector_calibration_fact_lookup_001",
|
||
"category": "fact_lookup",
|
||
"input": "查一下 QPI 这个缩写是什么意思，只要事实。",
|
||
"expected_selector_behavior": "no_call",
|
||
"expected_notes": "Hard no-call because the user asks for lookup only."
|
||
},
|
||
{
|
||
"case_id": "selector_calibration_fact_lookup_002",
|
||
"category": "fact_lookup",
|
||
"input": "这篇文章的发布日期是哪天？不要展开。",
|
||
"expected_selector_behavior": "no_call",
|
||
"expected_notes": "Date lookup does not need QPI or IA."
|
||
},
|
||
{
|
||
"case_id": "selector_calibration_rewrite_001",
|
||
"category": "rewrite_polish_translation",
|
||
"input": "请直接润色这段话，不要深入分析。",
|
||
"expected_selector_behavior": "no_call",
|
||
"expected_notes": "Pure polish request."
|
||
},
|
||
{
|
||
"case_id": "selector_calibration_translation_001",
|
||
"category": "rewrite_polish_translation",
|
||
"input": "翻译这句话：The model is provisional.",
|
||
"expected_selector_behavior": "no_call",
|
||
"expected_notes": "Translation request."
|
||
},
|
||
{
|
||
"case_id": "selector_calibration_direct_execution_001",
|
||
"category": "direct_execution",
|
||
"input": "把这个 JSON 缩进格式化一下，马上执行，不用解释。",
|
||
"expected_selector_behavior": "no_call",
|
||
"expected_notes": "Direct execution request."
|
||
},
|
||
{
|
||
"case_id": "selector_calibration_qpi_only_001",
|
||
"category": "qpi_only",
|
||
"input": "我们现在的问题到底是缺数据、缺路径，还是缺共识？",
|
||
"expected_selector_behavior": "select_qpi",
|
||
"expected_notes": "Explicit QPI problem-definition gate."
|
||
},
|
||
{
|
||
"case_id": "selector_calibration_qpi_only_002",
|
||
"category": "qpi_only",
|
||
"input": "这个需求反复卡住，我不确定该怎么定义这个问题。",
|
||
"expected_selector_behavior": "select_qpi",
|
||
"expected_notes": "Problem definition is incomplete."
|
||
},
|
||
{
|
||
"case_id": "selector_calibration_qpi_insufficient_context_001",
|
||
"category": "mixed_qpi_cases",
|
||
"input": "如何提高流量？",
|
||
"expected_selector_behavior": "select_qpi_low_confidence",
|
||
"expected_notes": "Selector may route to QPI, but QPI output must be provisional because subject and scenario are missing."
|
||
},
|
||
{
|
||
"case_id": "selector_calibration_qpi_intra_frame_mixed_001",
|
||
"category": "mixed_qpi_cases",
|
||
"input": "我是集团营销负责人，既缺流量增长方法，也担心销售、库存、客服和激励机制跟不上。",
|
||
"expected_selector_behavior": "select_qpi",
|
||
"expected_notes": "Same subject and scenario contain path/resource and governance scarcity."
|
||
},
|
||
{
|
||
"case_id": "selector_calibration_qpi_governance_load_001",
|
||
"category": "mixed_qpi_cases",
|
||
"input": "这个流程改造表面只是工具不熟，但实际牵涉真实调用、上下文裁剪、审计边界和后续复用。",
|
||
"expected_selector_behavior": "select_qpi",
|
||
"expected_notes": "Governance load raises Issue weight even if the owner is one person."
|
||
},
|
||
{
|
||
"case_id": "selector_calibration_ia_after_qpi_001",
|
||
"category": "ia_after_qpi",
|
||
"input": "QPI 已判断这是中重型课题，请用思想考古分析它的底层假设和核心机制。",
|
||
"expected_selector_behavior": "select_intellectual_archaeology",
|
||
"expected_notes": "QPI-before-IA gate is satisfied."
|
||
},
|
||
{
|
||
"case_id": "selector_calibration_ia_after_qpi_002",
|
||
"category": "ia_after_qpi",
|
||
"input": "QPI 已判断这是 P/I mixed，请下潜到 purpose 或 core_mechanism，说明不用到哲学基岩的理由。",
|
||
"expected_selector_behavior": "select_intellectual_archaeology",
|
||
"expected_notes": "Depth should be gated, not default full seven-layer analysis."
|
||
},
|
||
{
|
||
"case_id": "selector_calibration_ia_model_extraction_001",
|
||
"category": "ia_explicit_model_extraction",
|
||
"input": "请从这篇文章抽取一个可复用模型，并检查它的底层假设是否稳定。",
|
||
"expected_selector_behavior": "select_intellectual_archaeology",
|
||
"expected_notes": "Explicit model extraction and heavy-depth signal."
|
||
},
|
||
{
|
||
"case_id": "selector_calibration_ia_model_extraction_002",
|
||
"category": "ia_explicit_model_extraction",
|
||
"input": "我想把这套专家直觉沉淀成模型，需要识别隐性知识和验证点。",
|
||
"expected_selector_behavior": "select_intellectual_archaeology",
|
||
"expected_notes": "Expert intuition and tacit knowledge are IA triggers."
|
||
},
|
||
{
|
||
"case_id": "selector_calibration_false_positive_deep_word_001",
|
||
"category": "false_positive_traps",
|
||
"input": "把这句话里的“底层逻辑”换成更通俗的说法，不要深入分析。",
|
||
"expected_selector_behavior": "no_call",
|
||
"expected_notes": "Deep word appears, but hard no-call rewrite intent wins."
|
||
},
|
||
{
|
||
"case_id": "selector_calibration_false_positive_model_word_001",
|
||
"category": "false_positive_traps",
|
||
"input": "翻译 model 这个词，不要展开。",
|
||
"expected_selector_behavior": "no_call",
|
||
"expected_notes": "The word model is not a model-extraction request."
|
||
},
|
||
{
|
||
"case_id": "selector_calibration_false_positive_philosophy_001",
|
||
"category": "false_positive_traps",
|
||
"input": "查一下哲学基岩这个词出现在哪篇文章，只查位置。",
|
||
"expected_selector_behavior": "no_call",
|
||
"expected_notes": "Lookup request wins over IA keyword."
|
||
},
|
||
{
|
||
"case_id": "selector_calibration_override_001",
|
||
"category": "direct_execution_with_analysis_override",
|
||
"input": "这看起来只是润色，但不要只执行，请分析背后的问题结构。",
|
||
"expected_selector_behavior": "select_qpi",
|
||
"expected_notes": "Explicit analysis override allows QPI despite a rewrite-like surface."
|
||
},
|
||
{
|
||
"case_id": "selector_calibration_pipeline_001",
|
||
"category": "pipeline",
|
||
"input": "问题定义还没完成，先不要思想考古，先判断这是 Q、P 还是 I。",
|
||
"expected_selector_behavior": "select_qpi_reject_ia",
|
||
"expected_notes": "QPI-before-IA gate."
|
||
},
|
||
{
|
||
"case_id": "selector_calibration_pipeline_002",
|
||
"category": "pipeline",
|
||
"input": "QPI 已判断不是事实查询，而是系统性课题；下一步检查它是否值得沉淀成可复用模型。",
|
||
"expected_selector_behavior": "select_intellectual_archaeology",
|
||
"expected_notes": "QPI has completed the front gate."
|
||
}
|
||
]
|
||
}
|