editorial_intents.yml

schema_version: 1

# 外部材料只匹配这些已拥有的母题，不能自行创造 Apodex 叙事。
intents:
  EI1_official_answer_changed:
    name: "正式答案因新证据发生变化"
    status: active
    columns: [C1, C3]
    problem_shapes: [PS1_evidence_shift, PS6_moving_target, PS8_auditable_conclusion]
    theses: [T3_asynchronous_verification, T6_auditable_history, T8_search_judgment_loop]
    event_triggers: [issued, revised, withdrew, reopened, overturned, warning added, warning removed, updated guidance, label change]
    source_types: [regulator, standards_body, court, scientific_institution, company_primary_record]
    product_backing: [website, technical_report]
    positive_examples: ["监管者因新安全信息修改警示、标签或正式建议", "权威机构公开撤回、修订或重新开启既有结论"]
    negative_examples: ["只发表了一篇新论文，没有改变现实中的正式判断", "泛泛讨论 AI 会犯错或需要 verification"]
    retrieval:
      x_queries:
        - '("revised guidance" OR "withdrew guidance" OR "reopened review" OR "label change")'
        - '("warning added" OR "warning removed" OR "updated official assessment")'

  EI2_authorities_disagree_on_action:
    name: "可信机构对同一现实决策给出不同答案"
    status: active
    columns: [C1]
    problem_shapes: [PS2_credible_conflict, PS3_conditional_decision]
    theses: [T2_independent_verification, T4_evidence_graph, T5_source_diversity]
    event_triggers: [rejected, disputed, split vote, conflicting recommendations, opposed, reversed decision]
    source_types: [regulator, advisory_committee, standards_body, court, scientific_institution]
    product_backing: [website, technical_report]
    positive_examples: ["两个可核验的权威主体对同一治疗、政策或标准给出冲突行动建议", "委员会表决与工作人员审查结论发生可解释的分歧"]
    negative_examples: ["学者之间的抽象观点争论", "两篇论文结果不一致但没有现实决策或后果"]
    retrieval:
      x_queries:
        - '("split vote" OR "rejected the recommendation" OR "disputed the finding")'
        - '("conflicting recommendations" OR "reversed the decision") (agency OR committee OR court)'

  EI3_open_decision_window:
    name: "答案仍在形成，外部证据现在可以改变它"
    status: active
    columns: [C3]
    problem_shapes: [PS5_definition_boundary, PS6_moving_target]
    theses: [T3_asynchronous_verification, T6_auditable_history]
    event_triggers: [proposed rule, draft guidance, request for comment, call for evidence, public consultation, advisory committee, comment deadline]
    source_types: [regulator, standards_body, legislature, scientific_funder, public_agency]
    product_backing: [website, technical_report]
    positive_examples: ["仍开放的正式征求意见、证据征集或拟议规则", "明确给出截止日期且结论尚未定稿的公共决策"]
    negative_examples: ["已经结束、没有现实参与窗口的旧 consultation", "活动预告或普通 conference call for papers"]
    retrieval:
      x_queries:
        - '("request for comment" OR "call for evidence" OR "public consultation")'
        - '("proposed rule" OR "draft guidance") (deadline OR comments OR hearing)'

  EI4_many_judgments_one_decision:
    name: "多个有偏判断怎样汇聚成更可靠的决策"
    status: active
    columns: [C1, C2]
    problem_shapes: [PS3_conditional_decision, PS4_fragmented_record, PS7_branching_investigation]
    theses: [T1_single_pass_ceiling, T4_evidence_graph, T7_agent_level_scaling]
    event_triggers: [forecast aggregation, expert elicitation, ensemble decision, collective forecast, decision tournament, crowd forecast]
    source_types: [primary_research, forecasting_institution, public_agency, scientific_institution]
    product_backing: [technical_report]
    positive_examples: ["真实项目展示多个有偏预测经过聚合后优于依赖单一判断", "系统设计改变了独立判断的汇聚、挑战或加权方式"]
    negative_examples: ["纯模型 ensemble benchmark", "没有现实决策含义的算法精度提升"]
    retrieval:
      x_queries:
        - '("forecast aggregation" OR "expert elicitation" OR "collective forecast") (study OR project)'
        - '("decision tournament" OR "crowd forecast") (accuracy OR calibration OR outcome)'

  EI5_fragmented_record_changes_case:
    name: "分散记录拼在一起后，案件或调查的含义改变"
    status: active
    columns: [C1]
    problem_shapes: [PS4_fragmented_record, PS7_branching_investigation, PS9_source_collapse]
    theses: [T1_single_pass_ceiling, T4_evidence_graph, T5_source_diversity, T8_search_judgment_loop]
    event_triggers: [investigation found, records revealed, newly disclosed, documents show, across jurisdictions, multiple registries, unreported case]
    source_types: [official_record, investigative_report, court_record, registry, scientific_institution]
    product_backing: [website, technical_report]
    positive_examples: ["多个独立记录或辖区拼接后揭示此前未公开的重要事实", "调查需要跨来源追踪支持、冲突与缺失证据"]
    negative_examples: ["单一新闻稿被多家媒体重复转载", "只总结一篇论文或一份报告"]
    retrieval:
      x_queries:
        - '("newly disclosed" OR "records revealed" OR "documents show") (investigation OR case)'
        - '("unreported case" OR "across jurisdictions" OR "multiple registries")'

  # ---------------------------------------------------------------------------
  # 常青引擎（evergreen）。EI1–EI5 全部事件驱动，没新闻的一周就断供。
  # EI6 / EI7 的主入口是 curated 母表，不是抓取器；X query 只提供 timing，
  # 不提供选题。这是"源源不断"的结构保证。
  #
  # 注意：collect.py:432 会跳过没有 x_queries 的 active intent，而
  # tests/test_pipeline_contract.py::test_each_active_intent_contributes_one_rotating_query
  # 断言每个 active intent 必须产出一条 query。所以两者都必须带 x_queries。
  # ---------------------------------------------------------------------------

  EI6_missing_control:
    name: "报告出来的指标没有排除掉替代解释"
    status: active
    intake: curated
    intake_source: config/confound_library.yml
    columns: [C1, C2]
    problem_shapes: [PS10_missing_control]
    theses: [T2_independent_verification]
    event_triggers: [reporting standard published, reporting guidelines updated, author checklist required, round robin study, interlaboratory comparison, checklist adopted by journal, minimum reporting set]
    source_types: [standards_body, scientific_institution, primary_research]
    product_backing: [technical_report]
    positive_examples:
      - "某学科发布或更新了一份公开的 reporting standard / author checklist，其中某一条报告项正是排除某个具名混杂因素所必需"
      - "多实验室 round-robin 显示统一材料下性能差异来自装配或协议参数，而非被研究的对象"
    negative_examples:
      - "泛泛讨论科学不可复现、复现性危机或同行评议产能问题（docs/01 排除边界已明文排除复现性泛叙事）"
      - "只是介绍一篇学术方法论文，没有具名指标与具名混杂因素"
      - "任何指向具体研究者或具体论文的诚信指控（pillars.yml exclude 硬拦）"
    hard_gate_ref: "pillars.yml → problem_shapes.PS10_missing_control.hard_gate"
    retrieval:
      x_queries:
        - '("reporting standard" OR "reporting guidelines" OR "author checklist") (journal OR editors OR required)'
        - '("round robin" OR "interlaboratory") (reproducibility OR variability OR protocol)'

  EI8_verification_role_gap:
    name: "某一类检查者系统性漏掉了什么，或漏掉之后代价落在谁身上"
    status: active
    intake: curated
    intake_source: config/checker_library.yml
    columns: [C1, C2]
    problem_shapes: [PS8_auditable_conclusion, PS9_source_collapse, PS10_missing_control]
    theses: [T2_independent_verification, T5_source_diversity, T6_auditable_history]
    event_triggers: [peer review study, replication study, citation network analysis, cost of irreproducibility, outcome switching, reporting discrepancy, correction of the record]
    source_types: [primary_research, scientific_institution, standards_body]
    product_backing: [technical_report]
    positive_examples:
      - "一份已发表研究量化了某一类检查环节的检出率、遗漏率或纠错渠道的实际通过率"
      - "一份已发表分析量化了结论不可靠之后由谁承担、承担多少（成本核算或成功率估计）"
    negative_examples:
      - "泛泛讨论同行评议不可靠、复现性危机或学术界体制问题（docs/01 排除边界已明文排除）"
      - "任何指向具体研究者、具体实验室或具体论文的诚信指控（pillars.yml exclude 硬拦）"
      - "以研究不当行为 / fraud 为主题的材料——已排除赛道，即使结构上贴题也不收"
    hard_gate_ref: "checker_library.yml 六必填：checker / systematic_gap / documented_finding / evidence_anchor / problem_shape_id / thesis_id"
    retrieval:
      x_queries:
        - '("peer review" OR "replication") ("detection rate" OR "how many" OR "failed to detect")'
        - '("cost of" OR "wasted") ("irreproducible" OR "unreliable") research'

  EI7_cross_field_isomorphism:
    name: "两个学科的难题在计算范式上是同一个问题"
    status: active
    intake: curated
    intake_source: "战略文件 A1 学科配对；Selene 供料"
    columns: [C1, C2]
    problem_shapes: [PS11_cross_field_isomorphism]
    theses: [T9_cross_domain_synthesis]
    event_triggers: [method transferred, borrowed from another field, same computational problem, structure property optimization, cross disciplinary method]
    source_types: [primary_research, scientific_institution]
    product_backing: [website, technical_report]
    positive_examples:
      - "A 学科的一个难题与 B 学科的难题在计算范式上同构，且 B 学科已有更成熟的解法而 A 学科尚未采用"
      - "一个方法从某领域迁移到另一个领域并给出可核证的结果"
    negative_examples:
      - "只是并列两个领域的表面类比，没有说明计算范式上哪里同构"
      - "把 Apodex 说成已在某学科验证过 —— 落地页 Where We've Gone Deepest 只有生命科学与量化研究"
    retrieval:
      x_queries:
        - '("same problem" OR "borrowed from" OR "transfers directly") (materials OR catalysis OR protein OR battery)'
        - '("structure-property" OR "high-dimensional search") (optimization OR design OR screening)'


  # 唯一一条不需要外部世界发生任何事就能出货的选题理由。前八条全部要么等外部事件
  # （EI1-EI5），要么锚在第三方已发表的记录上（EI6/EI8/EI7）。这一条锚在我们自己的
  # 技术报告上：一个已确认的设计选择 + 一个可引用的数字，讲"我们为什么这样设计"。
  # 因为锚是自家材料，它不构成独立证据 —— 硬闸门里的 boundary 字段就是为此存在。
  EI9_design_choice_from_report:
    name: "我们为什么这样设计：技术报告里一个已确认的设计选择，配一个可引用的数字"
    status: active
    intake: curated
    intake_source: "config/education_library.yml"
    columns: [C2]
    problem_shapes: [PS7_branching_investigation, PS8_auditable_conclusion, PS10_missing_control]
    theses:
      - T2_independent_verification
      - T3_asynchronous_verification
      - T6_auditable_history
      - T7_agent_level_scaling
    event_triggers: []
    source_types: [owned_material]
    product_backing: [technical_report]
    positive_examples:
      - "技术报告里一个写明了的设计选择，配一个能被别人引用的数值或坐标，并说清这个数字不成立的边界"
      - "一个我们主动放弃的做法，以及放弃它之后测出来的差别"
    negative_examples:
      - "只有主张没有数字（『我们更可靠』），或只有数字没有机制（贴一张成绩单）"
      - "把自家评测结果写成行业公认结论 —— 锚是自家材料，必须在文案里标明来源"
      - "对比类数字不标明对比的是哪一版对手 —— 对手更新后该说法即失效"
      - "报告里标注 still in flight / 未完成的数字"
    hard_gate_ref: "education_library.yml 五必填：claim / number（必须含阿拉伯数字）/ mechanism / report_section / boundary"
    retrieval:
      x_queries: []