eduevidence 6.0.0 → 6.2.0

package/skill/sub-skills/evidence-review/SKILL.md

@@ -1,18 +1,37 @@
 ---
 name: evidence-review
 description: "Synthesizes extracted claims and evidence nodes into the project Evidence Graph with meta-analysis pooling."
+capability: claim_linking + evidence_synthesis
 ---
+
 # Evidence Review Skill
 
-## 1. When to Use
-Trigger to aggregate all validated evidence nodes into the project's single source of truth (SSOT) Claim Graph and execute quantitative synthesis.
-
-## 2. Process
-1. **Evidence Graph State Machine**:
-   - Construct directed graph of Claims, Sources, and Findings.
-   - Determine claim status: SUPPORTED, CONTRADICTED, MIXED, or UNCERTAIN.
-2. **Meta-Analysis Pooling**:
-   - Fixed-effect inverse-variance pooling and DerSimonian-Laird random-effects pooling.
-   - Cochran's Q, degrees of freedom, and I^2 heterogeneity index.
-   - Egger regression and Rosenthal Fail-Safe N publication bias checks.
-   - Leave-one-out study sensitivity analysis.
+## When to Use
+Trigger to aggregate validated evidence nodes into the project's single source of truth (SSOT) Claim Graph and execute quantitative synthesis.
+
+## Inputs
+- `evidence.jsonl` + `methodology.json`
+- 项目 Evidence Graph 当前 revision
+
+## Process
+1. **Evidence Graph State Machine**: build the directed graph of claims, sources, findings; determine claim status (SUPPORTED / CONTRADICTED / MIXED / UNCERTAIN).
+2. **Meta-Analysis Pooling** (`engine/meta_analysis.py`): fixed-effect inverse-variance pooling and DerSimonian–Laird random-effects pooling; Cochran's Q, df, I².
+3. **Publication Bias & Robustness** (`engine/bias.py` / `engine/robustness.py`): Egger regression, Rosenthal fail-safe N, leave-one-out sensitivity.
+4. **Counting rule**: pool by independent **study**, never by finding count.
+
+## Output Contract
+Graph revision (append-only) plus synthesis artifacts; projections (`result.json`, HTML) are derived, never the source of truth.
+
+## Quality Gates
+- [ ] 按独立研究计数，非按 finding 计数。
+- [ ] 异质性（I²、Q）与偏倚检验结果一并报告。
+- [ ] 三个方向列未被合并隐藏。
+
+## Anti-Patterns
+- 同一研究的多个 outcome 当作多项独立证据；对不可合并的结果强行做标准化合并。
+
+## Worked Example
+8 来源 12 条发现 → 仅部分可合并；不可合并者以叙述式证据矩阵呈现并标注原因。
+
+## References
+- `references/evidence-quality.md`、`docs/evidence-synthesis.md`、`engine/meta_analysis.py`

package/skill/sub-skills/gap-analysis/SKILL.md

@@ -1,25 +1,47 @@
 ---
 name: gap-analysis
-description: "Identifies population, measurement, and methodological gaps in the Evidence Graph and diagnoses cross-study empirical contradictions using BioGapLens PICO taxonomy."
+description: "Identifies population, measurement, and methodological gaps in the Evidence Graph and diagnoses cross-study empirical contradictions."
+capability: knowledge_gap_detection
 ---
 
-# gap-analysis — Research Gap Discovery & Contradiction Lens Sub-Skill
+# gap-analysis — Research Gap Discovery & Contradiction Lens
 
 ## When to Use
-Triggered after Evidence Extraction and Meta-Analysis (between Step 8 and Step 9) to ensure that trial designs are strictly grounded on verified empirical gaps rather than generic templates.
+Triggered after evidence extraction and meta-analysis, before study design: trial designs must be grounded on verified empirical gaps, never on generic templates.
 
-## Execution Rules
-1. **Measurement & Retention Gap Audit**: If evidence only measures immediate task speed (OutcomeDimension: PROCEDURAL_EFFICIENCY), flag a missing delayed unassisted retention gap.
-2. **Population Heterogeneity Audit**: Check if studies exclusively evaluate elite CS majors or introductory cohorts; flag advanced algorithmic transfer gaps.
-3. **Contradiction Lens**: When studies report divergent effect directions ($g > +0.3$ vs $g < -0.1$), isolate the moderating variable (e.g. Socratic scaffolding vs unguided copy-pasting).
+## Inputs
+- Evidence Graph（含 claim 状态与 outcome 维度）
+- 矛盾诊断（方向冲突、异质性来源）
+
+## Process
+1. **Measurement & Retention Gap Audit**: if evidence only measures immediate task speed, flag the missing delayed unassisted retention measurement.
+2. **Population Heterogeneity Audit**: check whether studies cover only elite CS majors or introductory cohorts; flag advanced-transfer gaps.
+3. **Contradiction Lens**: when directions diverge (g > +0.3 vs g < −0.1), isolate the moderating variable (e.g. scaffolded vs unguided use).
+4. **Priority**: rank gaps by decision-materiality so study design is not spent on immaterial questions.
+
+## Output Contract
+`GapNode` list written directly to the SSOT Evidence Graph, each with `gap_id`, `gap_type`, `description`, `target_outcome`, `recommended_trial_design`.
 
-## Output Contract (`GapNode` list written directly to SSOT `EvidenceGraph`)
 ```json
 {
   "gap_id": "GAP-RETENTION-001",
   "gap_type": "Measurement/Retention Gap",
   "description": "Lack of 12-week longitudinal retention data measuring unassisted transfer in CS1.",
   "target_outcome": "Delayed Unassisted Problem Solving",
-  "recommended_trial_design": "12-Week Cluster Randomized Trial with 4-week delayed post-test without AI access"
+  "recommended_trial_design": "12-week cluster randomized trial with 4-week delayed post-test without AI access"
 }
 ```
+
+## Quality Gates
+- [ ] 每个 gap 绑定具体未满足的测量/人群/方法维度。
+- [ ] gap 与决策材料性关联，而非"文献少"。
+- [ ] 不把"论文数量少"当作研究缺口。
+
+## Anti-Patterns
+- 用"研究不足"概括一切；把已有证据能回答的问题标成缺口。
+
+## Worked Example
+现有证据多为即测任务表现 → 产出 GAP-RETENTION-001，约束后续 StudyDesign 必须包含无 AI 的延迟后测。
+
+## References
+- `engine/gap_lens.py`、`engine/gaps.py`、`references/scientific-invariants.md`

package/skill/sub-skills/literature-review/SKILL.md

@@ -1,21 +1,42 @@
 ---
 name: literature-review
-description: "Executes multi-source academic and web retrieval across OpenAlex, Semantic Scholar, CrossRef, AIHot, AgentSearch, and user-configured providers (Tavily, Brave, SerpAPI)."
+description: "Executes multi-source academic and web retrieval across OpenAlex, Semantic Scholar, CrossRef, AIHot, AgentSearch, Sciverse, and user-configured providers."
+capability: literature_search + counter_evidence_search + source_fetch + source_validation
 ---
+
 # Literature Review Skill
 
-## 1. When to Use
+## When to Use
 Trigger after research intent is established to gather candidate empirical studies, peer-reviewed papers, and verified grey literature.
 
-## 2. Multi-Channel Search Routing
-1. **Zero-Config Academic Providers**:
-   - OpenAlex: Search 250M+ scholarly works for peer-reviewed studies with DOIs.
-   - Semantic Scholar: Graph-based citation and abstract retrieval.
-   - CrossRef: Official DOI resolution and metadata extraction.
-   - AIHot: Real-time AI research trends and technical reports.
-   - AgentSearch: SciPhi open academic and vector search.
-2. **Configured API Key Providers**:
-   - Tavily, Brave Search, SerpAPI, Exa, Bocha.
-3. **Fetch & Validation**:
-   - Pass candidate URLs through retrieval/fetch.py and validate via retrieval/validate.py.
-   - Strictly reject ungrounded snippet hallucinations.
+## Inputs
+- `frame.json`（检索边界、纳排标准）
+- 检索预算（S/M/L 决定）与可选 key：`SCIVERSE_API_TOKEN` / `TAVILY_API_KEY` / `BRAVE_API_KEY`
+
+## Process
+1. **Plan first** — 写 `SearchPlan`（core / expansion / **counter_evidence**），经 `retrieval/audit.py` 执行并导出审计四件套。
+2. **Zero-Config Academic Providers**: OpenAlex (250M+ works with DOIs), Semantic Scholar (graph citations + abstracts), CrossRef (DOI registry), AIHot (real-time AI/EdTech feed), AgentSearch / ArXiv.
+3. **Key-based academic channel**: Sciverse (`SCIVERSE_API_TOKEN`) — `/meta-search` 产出 Source 级命中；`/agentic-search` 产出 chunk 定位子，必须经 `/content` 读原文并过校验门，定位写入 `chunks.jsonl`。
+4. **Configured web providers**: Tavily, Brave.
+5. **Fetch & Validation**: 候选 URL 走 `retrieval/fetch.py` 降级链并由 `retrieval/validate.py` 校验；严格拒绝 snippet 幻觉。
+6. **Compliance**: 遵守 `references/retrieval-compliance.md`（robots / 限速 / paywall / 署名与缓存）。
+
+## Output Contract
+- `sources.jsonl`（`schemas/source.schema.json`）+ `fetch/`（raw + clean + provenance + fallback_chain）
+- 审计导出：`search-provenance.json` / `search-attempts.jsonl` / `source-screening.csv` / `exclusion-log.csv`
+- Sciverse 定位：`chunks.jsonl`（`locator_state=discovery_only_requires_content_fetch`）
+
+## Quality Gates
+- [ ] 反方查询独立构造且计数 > 0。
+- [ ] 每条来源为 FETCH_VALID 或经确认的 FETCH_PARTIAL。
+- [ ] 无 DOI/URL 记录标 `needs_manual_location`，未伪造定位。
+- [ ] 排除记录写明原因。
+
+## Anti-Patterns
+- 用支持证据的检索式找反证；把 abstract 当证据内容；为凑数填充无关来源。
+
+## Worked Example
+`python scripts/search_provenance.py "first-year CS generative AI coding assistant" --out runs/x/provenance --concept "AI coding assistant"` → 审计四件套 + 候选来源表，随后逐条 fetch/validate。
+
+## References
+- `references/retrieval-protocol.md`、`references/retrieval-compliance.md`、`references/source-validity.md`、`skill/task-briefs/retrieve.md`

package/skill/sub-skills/methodology-audit/SKILL.md

@@ -1,20 +1,37 @@
 ---
 name: methodology-audit
 description: "Audits empirical studies against WWC 5.0, GRADE risk-of-bias frameworks, and social science pitfalls."
+capability: methodology_appraisal
 ---
+
 # Methodology Audit Skill (Methodology Tribunal)
 
-## 1. When to Use
+## When to Use
 Trigger before claim synthesis to evaluate threats to internal and external validity, applying methodological confidence scoring.
 
-## 2. Audit Frameworks
-1. **WWC 5.0 Rating**:
-   - Tier 1: Meets WWC Standards Without Reservations (clean RCT, low attrition).
-   - Tier 2: Meets WWC Standards With Reservations (QED with baseline equivalence).
-   - Tier 3: Correlational / Promising.
-2. **Social Science 4 Critical Pitfalls**:
-   - Task Performance != Genuine Learning.
-   - Short-Term Score != Long-Term Retention (4+ weeks).
-   - AI-Assisted Performance != Independent Transfer.
-   - Correlation != Causation.
-3. **GRADE Certainty Rating**: High, Moderate, Low, Very Low.
+## Inputs
+- `evidence.jsonl` + `fetch/` 原文
+- `references/wwc_standards.md`、`references/grade_framework.md`、`references/social_science_pitfalls.md`
+
+## Process
+1. **WWC 5.0 Rating**: Tier 1 (meets standards without reservations — clean RCT, low attrition), Tier 2 (with reservations — QED with baseline equivalence), Tier 3 (correlational / promising).
+2. **Social Science Pitfalls**: task performance ≠ genuine learning; short-term score ≠ retention (4+ weeks); AI-assisted performance ≠ independent transfer; correlation ≠ causation.
+3. **GRADE Certainty**: High / Moderate / Low / Very Low at the body-of-evidence level.
+4. **15-item checklist** in fixed order (control_group … dropout), each `met|partial|missing|not_applicable`.
+
+## Output Contract
+`methodology.json` per `schemas/methodology.schema.json`, including `task_vs_learning_guard.equates_task_with_learning`.
+
+## Quality Gates
+- [ ] 15 项齐全且仅用四值枚举。
+- [ ] guard 结论明确。
+- [ ] 审计只判"证据是否成立"，不判"证据说什么"。
+
+## Anti-Patterns
+- 用样本量大掩盖无对照；把相关性研究列入因果结论；缺项直接记 `met`。
+
+## Worked Example
+某准实验无前测等价性 → Tier 2 (with reservations) + `pre_test: missing`；其结论只可用于"提示可能"，不进强支持列。
+
+## References
+- `references/methodology-audit.md`、`references/tribunal-policy.md`、`skill/agents/method-reviewer.md`、`skill/task-briefs/audit.md`

package/skill/sub-skills/report-generation/SKILL.md

@@ -1,6 +1,7 @@
 ---
 name: report-generation
 description: "Renders 5 baked-theme single-file bilingual HTML reports, executive Visual Briefs, and Markdown reports, powered by Lieflat Charts editorial visualization standards and AI-composed, data-driven chart galleries."
+capability: report_projection + report_rendering
 ---
 # Report Generation Skill
 
@@ -55,3 +56,30 @@ Dark themes draw on the theme's `card_bg`; text contrast is checked against the
 
 ## 4. Web Studio Sync
 The Local Web Studio (`scripts/dashboard_server.py`) serves the baked HTML reports and Lieflat figures directly.
+
+## Inputs
+- `result.json` / `result.zh.json`（叙述字段先过语言门禁 `check_language_parallel`）
+- 当前 Graph Revision 与 decision snapshot 标识
+
+## Quality Gates
+- [ ] 双语语义对齐（数字 / ID / 枚举 / URL 不变）。
+- [ ] 渲染完整性门通过：显示数值可回溯到 `result.json`，探针无 `REPORT_INVALID`。
+- [ ] 布局不变量通过 `scripts/lint_report_layout.py`（390 / 768 / 1280 × brief/full）。
+- [ ] 溯源表格保留（来源表 / 证据矩阵 / Claim Trace），图表只作补充。
+- [ ] `artifact_manifest.json` 记录来源 revision 与各产物哈希。
+
+## Anti-Patterns
+- 由模型写入图表数值（数值必须来自 `scripts/charts_data.py` 提取器）。
+- 为了"更可视化"而删除可核验表格；用占位或虚构数据补齐缺失图表。
+- 报告与快照不一致时改报告不改结论；在 HTML 内提供运行时换肤。
+
+## Failure Handling
+| 失败 | 处理 |
+|---|---|
+| `REPORT_INVALID` | 阻断发布并重跑渲染，不带缺陷投放。 |
+| 双语不对齐 | 修复 `result.zh.json` 后重新烘焙。 |
+| 数据不足以支撑某图 | 抑制该图并记录原因，不补造数据。 |
+| 快照缺失 | 回到上游科学阶段补齐；禁止用默认值生成报告。 |
+
+## References
+- `skill/task-briefs/present.md`、`skill/task-briefs/projection.md`、`visualization/eduevidence-report/references/layout-constraints.md`

package/skill/sub-skills/research-planning/SKILL.md

@@ -1,21 +1,48 @@
 ---
 name: research-planning
-description: "Extracts the structured Education Research Frame (PICO + decision target) from user natural language and determines execution mode (S/M/L) via the Complexity Gate."
+description: "Extracts the structured Research Frame (PICO + decision target, in the selected domain's vocabulary) from user natural language and determines execution mode (S/M/L) via the Complexity Gate."
+capability: research_framing
 ---
+
 # Research Planning Skill
 
-## 1. When to Use
-Trigger this skill when the user initiates a new empirical educational or social science inquiry or updates the research scope. This is the **Frame** stage of the Canonical Protocol (docs/architecture.md).
+## When to Use
+Trigger this skill when the user initiates a new empirical inquiry — education, policy, organisational practice, or any other registered domain — or updates the research scope. This is the **Frame** stage of the Canonical Protocol (`docs/architecture.md`).
+
+## Inputs
+- 用户原始问题（自然语言，可能含隐含假设）
+- 可选结构化线索：population / intervention / comparison / outcomes / constraints / depth
+- 领域（决定 frame 词汇）：education（learner/course）/ policy（decision_object/population/stakeholders），或其他注册在 `domains/` 下的领域
 
-## 2. Process
-1. **Frame Formulation (Education Research Frame)**:
-   - **Population (P)**: Target learner/cohort demographics, grade level, domain.
-   - **Intervention (I)**: Specific pedagogical technique, tool, AI system, or curriculum change.
+## Process
+1. **Frame Formulation (Research Frame，字段随领域而定)**
+   - **Population (P)**: Who or what the decision acts on — learners and courses in education, affected groups and stakeholders in policy, customers or staff in organisations. Use the field vocabulary of the selected domain.
+   - **Intervention (I)**: The specific practice under decision — a teaching method, tool, AI system, curriculum change, regulation, process or programme.
    - **Comparison (C)**: Active control, business-as-usual, or non-intervention baseline.
-   - **Outcomes (O)**: Primary and secondary outcome metrics (task performance vs delayed transfer must be separated).
-   - **decision_target + scope + inclusion/exclusion criteria**: required by education-frame.schema.json.
-2. **Complexity Gating (S/M/L)** (scripts/complexity_gate.py, 全程唯一权威分级):
-   - S (Quick Fact): Fact-checking single claim (k=3-5).
-   - M (Standard Review): Full multi-source evidence review (k=8-15).
-   - L (Deep Causal Cycle): Synthesis + Trial Design + Empirical DID Regression.
-3. **Output Contract**: Frame 本体按 schemas/education-frame.schema.json（V1）校验；研究意图投影用 schemas/v2/research-intent.schema.json（V2 契约）。框架完整前禁止任何教学建议（领域层第一道闸门）。
+   - **Outcomes (O)**: Primary and secondary outcome metrics; task performance and delayed transfer must stay separate.
+   - **decision target + scope + inclusion/exclusion criteria**: required by the selected domain's frame schema.
+2. **Complexity Gating (S/M/L)** — `scripts/complexity_gate.py` is the single authority:
+   - S (Quick Fact): fact-check a single claim (k=3–5).
+   - M (Standard Review): full multi-source evidence review (k=8–15).
+   - L (Deep Causal Cycle): synthesis + trial design + empirical DID regression.
+3. **Missing inputs**: ask the user (`NEEDS_USER_CONTEXT`); never invent a default population, comparator, or outcome.
+
+## Output Contract
+- Frame body validates against the selected domain's frame schema (`domains/<id>/manifest.json` → `frame_schema`): `schemas/education-frame.schema.json` for education, `domains/policy/frame.schema.json` for policy.
+- Research-intent projection uses `schemas/v2/research-intent.schema.json` (V2 contract).
+- **No intervention advice may be produced before framing completes** — the domain layer's first gate.
+
+## Quality Gates
+- [ ] PICO 五要素齐全，comparator 明确，且字段词汇与所选领域一致。
+- [ ] primary outcome 与 secondary/risk outcome 分离。
+- [ ] scope 与纳排标准可操作，非"相关研究"式表述。
+- [ ] 随 frame 输出 S/M/L 等级建议。
+
+## Anti-Patterns
+- 把"相关研究"当作 scope；把任务完成速度当作学习效果；为缺失人群编造默认值；frame 未完成即给干预建议；把某个领域的字段强制套到另一个领域。
+
+## Worked Example
+输入"大一 C 语言课程要不要允许用 AI 编程助手？" → education 域：population=CS1 学生；intervention=生成式 AI 编程助手（含护栏）；comparison=无 AI 常规教学；primary outcome=独立考试成绩（learning）；decision_target=teaching_decision。同一协议在 policy 域写成：population=客服团队与受影响客户；intervention=带人工审核的 AI 助手；decision_object=adopt。
+
+## References
+- `schemas/education-frame.schema.json`、`references/education-framing.md`、`references/outcome-taxonomy.md`、`skill/task-briefs/frame.md`

package/skill/sub-skills/study-design/SKILL.md

@@ -1,16 +1,37 @@
 ---
 name: study-design
 description: "Generates pre-registered quasi-experimental DID and RCT trial designs to fill identified knowledge gaps."
+capability: study_design + measurement_design
 ---
+
 # Study Design Skill
 
-## 1. When to Use
-Trigger automatically when Evidence Graph yields MIXED, UNCERTAIN, or INSUFFICIENT verdicts, converting knowledge gaps into actionable empirical protocols.
+## When to Use
+Trigger when the Evidence Graph yields MIXED, UNCERTAIN, or INSUFFICIENT outcomes — converting knowledge gaps into actionable empirical protocols. **A new design requires an explicit evidence-grounded KnowledgeGap ID.**
+
+## Inputs
+- `GapNode`（`GAP-*`）+ claim 状态
+- 目标人群、可用课时、伦理审查结论
+
+## Process
+1. **Trial Specification**: sampling frame; cluster randomization or matched control classes; 4-phase rollout — baseline pre-test (W1), treatment exposure (W2–W9), immediate post-test (W10), 4-week delayed transfer test (W14).
+2. **Pre-Registration Manifest**: power calculation (e.g. N > 120 for power ≥ 0.80 at g = 0.35); primary/secondary instruments; stopping rules; pre-specified regression model.
+3. **Measurement plan**: hand over to evaluation design so thresholds are fixed before enrolment.
+
+## Output Contract
+StudyDesign + measurement plan validated against the grounded-study gate (`scripts/complexity_gate.py` + engine study-design checks). "Few papers found" alone never authorises a pilot.
+
+## Quality Gates
+- [ ] 设计引用显式 KnowledgeGap ID。
+- [ ] 功效计算与停止规则齐备。
+- [ ] 包含无 AI 的延迟迁移测量。
+- [ ] 伦理审查已通过。
+
+## Anti-Patterns
+- 因"文献少"直接开研究；事后补功效计算；把即测当作迁移。
+
+## Worked Example
+GAP-RETENTION-001 → 12 周集群随机设计：W1 前测 / W10 即测 / W14 无 AI 迁移测，N≈160，预先注册分析模型。
 
-## 2. Process
-1. **12-Week Trial Specification**:
-   - Sampling frame, cluster randomization / matched control classrooms.
-   - 4-phase rollout: Baseline pre-test (W1), Treatment exposure (W2-W9), Immediate post-test (W10), 4-week delayed transfer test (W14).
-2. **Pre-Registration Manifest**:
-   - Statistical power calculation (N > 120 for power >= 0.80 at effect size g = 0.35).
-   - Primary/secondary outcome instruments, stopping rules, and pre-specified regression model.
+## References
+- `engine/study_design.py`、`references/intervention-design.md`、`skill/sub-skills/ethics-review/SKILL.md`

package/skill/task-briefs/adjudicate.md

@@ -4,14 +4,39 @@
 整合 Frame + Evidence Matrix + Skeptic Findings + Method Reviews，产出四态裁决
 （ADOPT/PILOT/REJECT/INSUFFICIENT EVIDENCE）+ Can/Cannot Claim + 证据边界。
 
-## 输入
-- frame.json、evidence.jsonl、skeptic.json、methodology.json
+## 前置输入
+- `frame.json`、`evidence.jsonl`、`skeptic.json`、`methodology.json`
+- 确定性置信度脚本 `scripts/compute_confidence.py` 与裁决前闸门 `scripts/pre_verdict_gate.py`
 
 ## 产出
-- raw_verdict.json（模型裁决）→ orchestrator 跑 Pre-Verdict Gate + 确定性置信度
-  → final_verdict.json（schemas/verdict.schema.json）。
+- `raw_verdict.json`（模型裁决）→ orchestrator 跑 Pre-Verdict Gate + 确定性置信度
+  → `final_verdict.json`（`schemas/verdict.schema.json`）。
 
-## 规则（人话化硬标准，第一页决策语言）
-- decision_rationale 必须为 ≤4 句面向读者的流畅散文（en/zh 分写）；
+## 执行规则（人话化硬标准，第一页决策语言）
+- `decision_rationale` 必须为 ≤4 句面向读者的流畅散文（en/zh 分写）；
 - 禁证据 ID 列表（E-xxx/EV-xxx）、禁 schema 键（overall_risk= 等）、禁截断残留（null）；
-- what_can/cannot_be_claimed 等列表同样人话化；统计数字可保留但以自然表达呈现。
+- what_can/cannot_be_claimed 等列表同样人话化；统计数字可保留但以自然表达呈现。
+- **先过 Pre-Verdict Gate 再定稿**：critical 失败会封顶置信度并禁止高置信 ADOPT。
+- 置信度由规则化公式给出（0.30 质量 + 0.25 一致性 + 0.20 直接性 + 0.25 独立研究数 − 双罚分），不由模型自评。
+- 证据不足时输出 INSUFFICIENT EVIDENCE，并写明"什么证据会改变这个决定"。
+- 语言人话化规则见 `references/scientific-invariants.md` 与角色提示词。
+
+## 质量门
+- [ ] `final_verdict.json` 通过 verdict schema，action 取自四态枚举。
+- [ ] Pre-Verdict Gate 已执行并留痕（`gate_report.json`）。
+- [ ] 置信度可复算，分解项可审计。
+- [ ] 反证与 threats_to_validity 均被显式回应（采纳、限定或驳回并说明理由）。
+
+## 失败模式与回退
+| 失败 | 处理 |
+|---|---|
+| `PRE_VERDICT_FAILED` | 修复前置产物（schema / 交叉评审 / 方法学）后重跑闸门。 |
+| `GATE_CRITICAL_FAILURE` | 封顶置信度，强制降至 PILOT 或 INSUFFICIENT EVIDENCE。 |
+| `CONFLICT_UNRESOLVED` | 保持不确定，不强行裁决。 |
+| 证据仅支持任务表现 | 不得据以产出学习效果类结论。 |
+
+## 语言与呈现契约
+面向决策者的第一页语言；ID、枚举、URL 保留可追溯性，但不出现在叙述句中。
+
+## 交接说明
+`final_verdict.json` 是 Applicability / Intervene / Evaluate 与一切投影的唯一依据。

package/skill/task-briefs/applicability.md

@@ -1,3 +1,38 @@
-# Applicability stage
+# Task Brief — stage: applicability（角色：evidence-judge）
 
-Assess transportability to the target population and setting. State supported population, implementation conditions, excluded populations, outcome limits, and uncertainty. Do not infer applicability from the presence of a positive result.
+## 目标
+判定证据可迁移到目标人群与场景的程度：明确支持人群、实施条件、被排除人群、结局边界与不确定性。
+
+## 前置输入
+- `evidence.jsonl`（含人群、场景、剂量与结局信息）
+- `final_verdict.json`（裁决与边界）
+- `references/applicability-policy.md`
+
+## 产出
+- `applicability.json`：supported / unsupported populations、conditions、outcome limits、
+  boundary of transfer、uncertainty statement；供 Intervene 判断试点人群是否落在支持范围内。
+
+## 执行规则
+- **阳性结果不自动迁移**：不能因为"研究有效"就推定目标人群同样有效。
+- 逐条对齐：目标人群 vs 研究人群、目标场景 vs 研究场景、目标 outcome vs 已测量 outcome。
+- 明确写出被排除人群（例如被排除的补习班学生、特殊需求学习者）与理由。
+- 剂量与实施条件（师资培训、课时、工具版本、护栏）必须显式列出——这些常常是效果成立的前提。
+- 不确定之处如实标注为不确定，不用"一般来说"这类泛化措辞掩盖缺口。
+
+## 质量门
+- [ ] 支持人群、条件、结局边界、排除人群四类信息齐全。
+- [ ] 每条边界都追溯到具体证据，而非泛泛陈述。
+- [ ] 不确定性显式声明，且与裁决置信度一致。
+
+## 失败模式与回退
+| 失败 | 处理 |
+|---|---|
+| `SCOPE_MISMATCH` | 缩小结论范围而非放宽适用条件。 |
+| 关键人群信息缺失 | 记入不确定项；不得默认"与目标人群相同"。 |
+| 边界与裁决口径冲突 | 以证据为准修正边界，并回退到 Adjudicate 复核。 |
+
+## 语言与呈现契约
+面向决策者的人话表述：谁适用、在什么条件下、对哪些结果、到什么程度为止。
+
+## 交接说明
+`applicability.json` 约束 Intervene 的试点人群；越界设计必须被拒绝或缩回支持范围。

package/skill/task-briefs/audit.md

@@ -3,13 +3,38 @@
 ## 目标
 按审计清单审查每个研究的方法学质量，强制执行"任务完成表现 ≠ 学习效果"最高优先级规则。
 
-## 输入
-- evidence.jsonl + 来源 fetch 内容
+## 前置输入
+- `evidence.jsonl` + 来源 fetch 内容
+- 审计框架：`references/wwc_standards.md`、`references/grade_framework.md`、`references/methodology-audit.md`
 
 ## 产出
-- methodology.json：audit_items（含 task_vs_learning_guard）+ 每条 PASS/CONCERN/FAIL verdict
-  （显示层经 zh_labels 映射中文）。
+- `methodology.json`：15 项 `audit_items`（control_group / randomization / pre_test / post_test /
+  retention_test / transfer_test / sample_bias / self_selection / measurement_validity / confounders /
+  instructor_effect / novelty_effect / tool_version_effect / ai_usage_policy / dropout，每项 status:
+  met|partial|missing|not_applicable）+ 每条 PASS/CONCERN/FAIL verdict + `task_vs_learning_guard`
+  （显示层经 zh_labels 映射中文），须通过 `schemas/methodology.schema.json`。
 
-## 规则
-- 只审"证据站不站得住"，不审"证据说什么"；审计说明（note/summary）为人话叙述，
-  枚举/代号（PASS/CONCERN/FAIL）只作标签。
+## 执行规则
+- 只审"证据站不站得住"，不审"证据说什么"——审计结论不得夹带对效果的判断。
+- `task_vs_learning_guard.equates_task_with_learning` 必须显式给出；为 true 时该研究不得支撑任何学习效果声明。
+- 审计说明（note/summary）为人话叙述，枚举/代号（PASS/CONCERN/FAIL）只作标签。
+- 独立性：Method Reviewer 与内容判断角色分离（`independence_required: true`），且需强上下文能力。
+
+## 质量门
+- [ ] 15 项审计项齐全，无缺项、无自造项。
+- [ ] 每项状态取自四值枚举，`not_applicable` 必须说明为何不适用。
+- [ ] `task_vs_learning_guard` 存在且结论明确。
+- [ ] 每条 PASS/CONCERN/FAIL 有可核验依据（对应原文特征）。
+
+## 失败模式与回退
+| 失败 | 处理 |
+|---|---|
+| 关键信息缺失（如未报告随机化） | 记 `missing` 并在说明中写清缺什么，不猜测。 |
+| 结论依赖任务表现 | 触发 guard，剥夺其学习效果支撑资格。 |
+| schema 校验失败 | 修复后重跑审计。 |
+
+## 语言与呈现契约
+审计叙述为人话；状态枚举只作标签显示。
+
+## 交接说明
+`methodology.json` 直接决定每条证据能否支持 claim；Adjudicate 必须读取 `task_vs_learning_guard` 与各条 verdict。

package/skill/task-briefs/challenge.md

@@ -5,11 +5,40 @@
 novelty effect、alternative explanation；禁止虚构反方证据；没有反方证据时输出
 NO CONTRADICTORY EVIDENCE FOUND。
 
-## 输入
-- evidence.jsonl
+## 前置输入
+- `evidence.jsonl`
+- `frame.json`（用于判断结论是否超出研究范围）
 
 ## 产出
-- skeptic.json：search_performed=true；findings（counter_evidence/null_results/confounders）。
+- `skeptic.json`：`search_performed=true`；`skeptic_findings[]`（九项检查，字段名稳定：
+  `check` / `status`（found|not_found）/ `detail` / `related_evidence_ids`）；
+  `contradictory_evidence_found`；`no_contradictory_evidence_statement`；`threats_to_validity`。
+- 作为独立交叉审核角色时的输出须符合 `schemas/cross-model-review.schema.json`
+  （必需字段 `agreement`、`final_recommendation`）。
 
-## 规则（人话化硬标准）
-- 叙述字段为面向研究者的流畅中文；禁证据 ID 堆砌（用"作者-年份 + 人话描述"替代）。
+## 执行规则（固定 9 项检查，缺一不可）
+1. null result　2. negative result　3. contradictory evidence　4. alternative explanation
+5. measurement mismatch（测的是任务完成还是学习）　6. sampling bias
+7. novelty effect　8. AI dependency / over-reliance　9. scope overreach
+
+- 反方检索必须独立构造查询，不复用支持证据的检索式。
+- 找不到反证就是 "not_found"，并输出标准语句；**不得为了凑数虚构反方文献**。
+- 独立性：Skeptic 不得与主分析使用同一模型家族（`independence_required: true`）。
+
+## 质量门
+- [ ] 九项检查全部有 `status`，无遗漏项。
+- [ ] 每条 `found` 的 finding 都绑定 `related_evidence_ids` 或明确来源。
+- [ ] `no_contradictory_evidence_statement` 与 `contradictory_evidence_found` 语义一致。
+
+## 失败模式与回退
+| 失败 | 处理 |
+|---|---|
+| 反证检索为空 | 记录 negative-search record；输出标准语句，不填充虚构来源。 |
+| 反证与支持证据冲突 | 保留双方，交 Adjudicate 处理；不在此阶段裁决。 |
+| 交叉审核不可用（无独立模型） | 降级为原生自审并显式标注，不得伪装为独立审核。 |
+
+## 语言与呈现契约（人话化硬标准）
+叙述字段为面向研究者的流畅中文；禁证据 ID 堆砌（用"作者-年份 + 人话描述"替代）。
+
+## 交接说明
+`skeptic.json` 是 Adjudicate 的必读输入；其 `threats_to_validity` 必须被裁决明确回应或采纳。

package/skill/task-briefs/evaluate.md

@@ -3,11 +3,36 @@
 ## 目标
 为 PILOT/ADOPT 配套评价方案：基线/后测/保持/迁移 + 过程/学习/风险指标 + 成功阈值与停止条件。
 
-## 输入
-- intervention.json + final_verdict.json
+## 前置输入
+- `intervention.json` + `final_verdict.json`
+- 评价框架：`references/evaluation-design.md`、`references/evaluation-policy.md`、`references/outcome-taxonomy.md`
 
 ## 产出
-- evaluation.json（schemas/evaluation.schema.json）。
+- `evaluation.json`（`schemas/evaluation.schema.json`）：baseline / post / retention / transfer 测量点，
+  过程与风险指标，成功与失败阈值，分析计划（可被 `scripts/did_regression.py` 直接执行）。
 
-## 规则
-- 分离 Task Performance 与 Learning Effect；叙述为人话。
+## 执行规则
+- 分离 Task Performance 与 Learning Effect；两者指标不得合并成一个"效果"。
+- 保持（retention）与迁移（transfer）必须单独设计测量点，不能只做即测。
+- 阈值在数据到达前固定；事后改阈值属协议偏离，须记录而非吸收。
+- 分析计划要具体到可执行（设计类型、比较组、协变量、缺失处理），使 DID/效应量脚本无需再做决定。
+- 叙述为人话。
+
+## 质量门
+- [ ] `evaluation.json` 通过 evaluation schema。
+- [ ] 基线、即测、保持、迁移四类测量点齐全。
+- [ ] 成功、失败、停止阈值三者都在，且先于数据固定。
+- [ ] 分析计划可被确定性脚本执行（无残留人工决策）。
+
+## 失败模式与回退
+| 失败 | 处理 |
+|---|---|
+| 无法做对照 | 改为单组前后测并显式标注设计局限，不得仍按因果口径表述。 |
+| 阈值事后调整 | 记录为协议偏离并降级结论强度。 |
+| 样本量不足 | 报告功效局限；不得以不显著当作"无效果"。 |
+
+## 语言与呈现契约
+人话评价方案；统计符号与阈值数字保留原样以便复算。
+
+## 交接说明
+试用数据回收后由 `evaluate-and-update` 工作流重新注入证据图谱并再裁决。