eduevidence 5.2.0 → 6.2.0

package/examples/ai-coding-assistant-evidence/skeptic.json

@@ -0,0 +1,72 @@
+{
+  "search_performed": true,
+  "method": "Counter-evidence challenge over the eight registry-verified sources in this pack: null and negative results, contradictory findings, alternative explanations, measurement mismatch, sampling bias, novelty effect, AI dependency and scope overreach. Every finding below cites records already in evidence.jsonl; nothing was invented for the challenge.",
+  "skeptic_findings": [
+    {
+      "check": "1_null_result",
+      "status": "found",
+      "detail": "Three null results are recorded. E-002 found no decrease on manual code-modification tasks, E-003 found no significant retention difference at one week (the retention window is short and the sample small), and E-005 found the guardrailed tutor removed the negative effect without producing a positive learning gain.",
+      "related_evidence_ids": ["E-002", "E-003", "E-005"]
+    },
+    {
+      "check": "2_negative_result",
+      "status": "found",
+      "detail": "E-004 reports a 17% lower independent-exam score for students with unguarded GPT-4 access despite higher practice performance, and E-012 records comprehension and debugging difficulties with low ownership of AI-generated solutions in a small single-session study.",
+      "related_evidence_ids": ["E-004", "E-012"]
+    },
+    {
+      "check": "3_contradictory_evidence",
+      "status": "found",
+      "detail": "E-004 directly contradicts the claim that AI access reliably improves learning: practice performance rose while independent performance fell. This is the single most decision-relevant piece of counter-evidence in the pack and is the reason the verdict is bounded rather than unconditional.",
+      "related_evidence_ids": ["E-004"]
+    },
+    {
+      "check": "4_alternative_explanation",
+      "status": "found",
+      "detail": "Task familiarity, prior programming competency, prompt-engineering effort, and tool-design differences (unguarded chat versus guardrailed tutor) are all recorded as confounders that could explain the observed differences without any effect on learning.",
+      "related_evidence_ids": ["E-001", "E-005", "E-006"]
+    },
+    {
+      "check": "5_measurement_mismatch",
+      "status": "found",
+      "detail": "E-006 and E-010 measure the tool rather than the learner: practice accuracy and CS1 exam-style pass rates show what the model can do, not what students learn. E-008 measures professional developers, and E-011 measures short-term explanation ratings rather than learning gains.",
+      "related_evidence_ids": ["E-006", "E-008", "E-010", "E-011"]
+    },
+    {
+      "check": "6_sampling_bias",
+      "status": "found",
+      "detail": "The randomised evidence comes from populations that are not the target: ages 10-17 novices (E-001/E-002/E-003), high-school mathematics students in one country (E-004/E-005/E-006), professional developers (E-008), and a small n=24 usability study (E-012). No university programming-course sample exists in this corpus.",
+      "related_evidence_ids": ["E-001", "E-004", "E-008", "E-012"]
+    },
+    {
+      "check": "7_novelty_effect",
+      "status": "found",
+      "detail": "The studies measured first exposure to these tools. A novelty effect cannot be separated from a durable effect in any of the included designs, and the professional-developer trial (E-008) is a single-task ecology.",
+      "related_evidence_ids": ["E-001", "E-008"]
+    },
+    {
+      "check": "8_ai_dependency",
+      "status": "found",
+      "detail": "E-004 documents unguarded access being used as a crutch with worse independent performance, and E-012 documents low ownership and debugging difficulty on AI-generated code. Dependency is a recorded mechanism, not a hypothetical risk.",
+      "related_evidence_ids": ["E-004", "E-012"]
+    },
+    {
+      "check": "9_scope_overreach",
+      "status": "found",
+      "detail": "Claiming that AI coding assistants improve university programming learning exceeds this corpus: there is no university-level randomised trial with a no-AI transfer test, the retention evidence spans one week, and benchmark or explanation-quality results do not substitute for classroom learning outcomes.",
+      "related_evidence_ids": ["E-003", "E-010", "E-011"]
+    }
+  ],
+  "contradictory_evidence_found": true,
+  "threats_to_validity": [
+    "no university programming-course sample in the reviewed set",
+    "one-week retention window is far shorter than a semester",
+    "task performance and independent performance diverge, so task measures cannot stand in for learning",
+    "unguarded and guardrailed tool designs are not interchangeable conditions",
+    "the professional-developer trial is a preprint and a single-task ecology"
+  ],
+  "extensions": {
+    "data_origin": "manual_curated",
+    "note": "Challenge stage written from the pack corpus; no new sources were retrieved for this record."
+  }
+}

package/examples/ai-coding-assistant-evidence/sources.jsonl

@@ -0,0 +1,8 @@
+{"source_id": "S-2023-kazemitabaar", "title": "Studying the effect of AI Code Generators on Supporting Novice Learners in Introductory Programming", "year": 2023, "doi": "10.1145/3544548.3580919", "canonical_url": "https://dl.acm.org/doi/10.1145/3544548.3580919", "authority_level": "tier1_paper_doi", "source_location": "https://dl.acm.org/doi/10.1145/3544548.3580919", "doi_verified": true, "retracted": false}
+{"source_id": "S-2025-bastani", "title": "Generative AI without guardrails can harm learning: Evidence from high school mathematics", "year": 2025, "doi": "10.1073/pnas.2422633122", "canonical_url": "https://www.pnas.org/doi/10.1073/pnas.2422633122", "authority_level": "tier1_paper_doi", "source_location": "https://www.pnas.org/doi/10.1073/pnas.2422633122", "doi_verified": true, "retracted": false}
+{"source_id": "S-2024-marzuki", "title": "Impact of ChatGPT on ESL students' academic writing skills", "year": 2024, "doi": "10.1186/s40561-024-00295-9", "canonical_url": "https://link.springer.com/article/10.1186/s40561-024-00295-9", "authority_level": "tier1_paper_doi", "source_location": "https://link.springer.com/article/10.1186/s40561-024-00295-9", "doi_verified": true, "retracted": false}
+{"source_id": "S-2023-peng", "title": "The Impact of AI on Developer Productivity: Evidence from GitHub Copilot", "year": 2023, "doi": "10.48550/arXiv.2302.06590", "canonical_url": "https://doi.org/10.48550/arXiv.2302.06590", "authority_level": "tier2_academic_database", "source_location": "https://arxiv.org/abs/2302.06590", "doi_verified": true, "retracted": false}
+{"source_id": "S-2023-yetistiren", "title": "GitHub Copilot AI Pair Programmer: Asset or Liability?", "year": 2023, "doi": "10.1016/j.jss.2023.111734", "canonical_url": "https://doi.org/10.1016/j.jss.2023.111734", "authority_level": "tier1_paper_doi", "source_location": "https://doi.org/10.1016/j.jss.2023.111734", "doi_verified": true, "retracted": false}
+{"source_id": "S-2022-finnie-ansley", "title": "Using GitHub Copilot to Solve Introductory Programming Problems", "year": 2022, "doi": "10.1145/3545945.3569830", "canonical_url": "https://dl.acm.org/doi/10.1145/3545945.3569830", "authority_level": "tier1_paper_doi", "source_location": "https://dl.acm.org/doi/10.1145/3545945.3569830", "doi_verified": true, "retracted": false}
+{"source_id": "S-2023-explanations-compare", "title": "Comparing Code Explanations Created by Students and Large Language Models", "year": 2023, "doi": "10.1145/3587102.3588785", "canonical_url": "https://dl.acm.org/doi/10.1145/3587102.3588785", "authority_level": "tier1_paper_doi", "source_location": "https://dl.acm.org/doi/10.1145/3587102.3588785", "doi_verified": true, "retracted": false}
+{"source_id": "S-2022-vaithilingam", "title": "Expectation vs. Experience: Evaluating the Usability of Code Generation Tools", "year": 2022, "doi": "10.1145/3491101.3519665", "canonical_url": "https://dl.acm.org/doi/10.1145/3491101.3519665", "authority_level": "tier1_paper_doi", "source_location": "https://dl.acm.org/doi/10.1145/3491101.3519665", "doi_verified": true, "retracted": false}

package/examples/ai-coding-assistant-evidence/verdict.json

@@ -0,0 +1,107 @@
+{
+  "decision_question": "大一 C 语言课程是否应该允许学生使用生成式 AI 编程助手？",
+  "target_population": "university first-year computer science students learning C programming for the first time",
+  "target_context": "16-week lecture-lab course, 60 students, TA support, offline",
+  "supported_claims": [
+    "AI coding assistants reliably increase task performance during training (completion speed, correctness) — E-001, E-006.",
+    "Unguarded generative AI access can harm independent problem solving when access is removed — E-004.",
+    "Guardrail design (hints instead of answers) substantially mitigates the negative learning effect — E-005.",
+    "Task performance gains do not automatically imply learning gains — E-004 vs E-006 (within-study contrast).",
+    "Tool capability is substantial: Codex solves roughly half to three-quarters of CS1 exam-style questions — E-010.",
+    "Professional-developer RCT shows ~55% faster task completion with Copilot; directness limited by professional population — E-008.",
+    "LLM code explanations rate comparable to student-authored explanations, viable as scaffold material — E-011."
+  ],
+  "uncertain_claims": [
+    "Whether AI coding assistants improve or preserve actual programming learning in university novices — no direct university-level RCT in reviewed set [无直接证据]",
+    "Whether one-week neutral retention (Kazemitabaar 2023) extends to a semester — E-003.",
+    "Whether benchmark quality findings (E-009) and explanation-quality ratings (E-011) translate into classroom learning gains.",
+    "How comprehension/ownership difficulties documented in usability studies (E-012) behave over a full semester with guardrails."
+  ],
+  "contradicted_claims": [
+    "The claim 'AI tools always improve learning' is contradicted by E-004 (unguarded access, -17% independent exam).",
+    "The claim 'speed gains equal learning gains' is contradicted by the task-vs-learning separation across E-001/E-006/E-008 vs E-004."
+  ],
+  "reason_for_disagreement": "Disagreement comes from outcome separation (task vs learning), tool design (guarded vs unguarded), and population (K-12 / professionals vs university). Task-performance evidence is consistently positive across randomized and benchmark studies; the only study measuring independent performance after AI removal shows harm without guardrails; usability and artifact studies add dependence and quality caveats rather than resolving the learning question.",
+  "methodology_summary": "Eight real sources: three randomized experiments (Kazemitabaar 2023 n=69 K-12; Bastani 2025 n≈950 high-school mathematics; Peng 2023 n=95 professional developers, preprint), one ESL writing mixed-methods study (Marzuki 2024), plus benchmark/capability/usability studies (Yetistiren 2023; Finnie-Ansley 2022; explanation-comparison 2023; Vaithilingam 2022). No direct RCT in university programming courses. Internal validity of the core RCTs is strong; directness to first-year university C programming is weak. All sources carry registry-verified DOIs (see benchmarks/doi-audit/report.md).",
+  "outcome_specific_findings": {
+    "completion_time": "positive during training and professional tasks (E-001, E-008)",
+    "independent_problem_solving": "neutral-to-negative without guardrails (E-002, E-004)",
+    "retention": "neutral over 1 week (E-003)",
+    "assignment_score": "positive during practice, negative on closed-book exam (E-004, E-006); tool itself scores passing-level on CS1 questions (E-010)",
+    "code_quality": "mixed on benchmarks; security concerns documented (E-009)",
+    "metacognition": "LLM explanations compare well (E-011) while novice ownership/debugging difficulties persist (E-012)",
+    "ai_dependency": "documented crutch behavior with unguarded tool (E-004, E-005, E-012)"
+  },
+  "short_term_effect": "Task performance reliably increases; learning effect null-to-negative without guardrails.",
+  "long_term_effect": "No evidence beyond one week; long-term learning effect unknown.",
+  "transfer_effect": "No full transfer evidence; manual code-modification not harmed in one small study (E-002).",
+  "risk_effect": "AI dependency and over-reliance risk is real and documented for unguarded usage (E-004) and foreshadowed by usability findings (E-012).",
+  "applicability": {
+    "suitable_for": "pilot in first-year C course with guardrailed usage policy",
+    "not_suitable_for": "unrestricted AI adoption without usage policy",
+    "required_conditions": [
+      "guardrailed AI usage policy (hints not answers, modeled on GPT Tutor arm)",
+      "no-AI transfer assessment",
+      "TA support"
+    ]
+  },
+  "confidence": "Moderate",
+  "confidence_breakdown": {
+    "score": 0.586,
+    "evidence_quality": 0.758,
+    "consistency": 0.667,
+    "directness": 0.458,
+    "evidence_count": 12,
+    "independent_studies": 8,
+    "independent_samples": 8,
+    "count_term": 1.0,
+    "conflict_penalty": 0.15,
+    "unsupported_penalty": 0.0
+  },
+  "what_can_be_claimed": [
+    "AI coding assistants raise task performance for novices during training.",
+    "Unguarded access carries a real risk of hurting independent problem solving.",
+    "Guardrail design can mitigate that risk.",
+    "Direct evidence for university C programming learning is missing.",
+    "Tool capability headroom is large (CS1 question pass rates; professional speed RCT)."
+  ],
+  "what_cannot_be_claimed": [
+    "AI coding assistants improve (or even preserve) university students' programming learning.",
+    "Any long-term or retention benefit.",
+    "Any claim about which students benefit, based on university samples.",
+    "That benchmark or usability findings substitute for classroom learning outcomes."
+  ],
+  "missing_evidence": [
+    "RCT of AI coding assistants in university programming courses with retention and no-AI transfer tests.",
+    "Studies varying AI usage policy within the same course.",
+    "Longitudinal data on AI dependency beyond one course.",
+    "Peer-reviewed replication of the professional speed RCT (Peng et al. remains a preprint)."
+  ],
+  "recommended_action": "pilot",
+  "decision_rationale": "Positive task-performance evidence plus documented unguarded-access risk, mixed quality/usability signals, and missing university-level learning evidence → bounded, guardrailed pilot with evaluation, not full adoption.",
+  "exceeds_evidence_boundary": [
+    "Claiming 'AI coding assistants improve learning' exceeds the boundary: direct learning-effect evidence is missing.",
+    "Claiming 'AI works for everyone' exceeds the boundary: population and subject mismatch."
+  ],
+  "confidence_score": 0.586,
+  "confidence_policy_version": "2026-08-12.v3",
+  "independent_studies": 8,
+  "independent_samples": 8,
+  "raw_model_confidence": "Moderate",
+  "raw_model_confidence_breakdown": {
+    "score": 0.5,
+    "evidence_quality": 0.7,
+    "consistency": 0.6,
+    "directness": 0.4,
+    "evidence_count": 12,
+    "independent_studies": 8,
+    "independent_samples": 8,
+    "count_term": 1.0,
+    "conflict_penalty": 0.0,
+    "unsupported_penalty": 0.0
+  },
+  "strongest_support": "AI coding assistants reliably speed up practice work: completion rate 1.15x and time 0.57x in a randomised trial of 69 novices.",
+  "key_uncertainty": "No university-level RCT measures learning directly, and the one large trial that did - unguarded GPT-4 - saw independent exam scores fall 17%.",
+  "main_risk": "Unguarded access can raise practice performance while lowering independent exam performance, and learners may not notice the gap.",
+  "next_action": "Run a phased CS1 pilot with hints-not-answers guardrails, weekly lab use, and a no-AI transfer exam that can stop the pilot."
+}

package/examples/spaced-retrieval-practice/applicability.json

@@ -0,0 +1,14 @@
+{
+  "status": "ASSESSED",
+  "suitable_for": "有固定课时的入门编程课程，教师能稳定安排每周短练习",
+  "not_suitable_for": "没有固定复习环节的课程，或无法保证延迟测评的场景",
+  "required_conditions": [
+    "每周固定短时检索练习",
+    "统一的延迟后测口径"
+  ],
+  "excluded_populations": [
+    "无固定课时的选修课学生"
+  ],
+  "outcome_limits": "迁移证据少于保持证据",
+  "uncertainty": "课程现场研究的数量有限"
+}

package/examples/spaced-retrieval-practice/artifact_manifest.json

@@ -0,0 +1,15 @@
+{
+  "result_sha256": "1fff22fb180f676523430c5934f8c96c9cabbb37e2a4010a99049de5087f478d",
+  "result_zh_sha256": "8c551e02ea2a76f6a48d68be6715eec118445998e66a35526fde967a77409b98",
+  "renderer_version": "1.0.0",
+  "git_commit": "56f5dc3edc6e805c2614767208a83009b0f0a097",
+  "evidence_count": 6,
+  "source_count": 7,
+  "themes": [
+    "claude",
+    "academic",
+    "datalab",
+    "datalab-dark",
+    "presentation"
+  ]
+}

package/examples/spaced-retrieval-practice/claims.jsonl

@@ -0,0 +1,3 @@
+{"claim_id": "C-001", "text": "Spaced retrieval practice reliably improves delayed retention", "evidence_ids": ["E-001", "E-002", "E-003", "E-004"], "status": "SUPPORTED", "scope": "introductory programming novices"}
+{"claim_id": "C-002", "text": "The size of the retention benefit depends on design and materials", "evidence_ids": ["E-004", "E-005"], "status": "SUPPORTED", "scope": "introductory programming novices"}
+{"claim_id": "C-003", "text": "Retrieval practice also supports transfer, not just memory", "evidence_ids": ["E-006"], "status": "SUPPORTED", "scope": "introductory programming novices"}

package/examples/spaced-retrieval-practice/evidence.jsonl

@@ -0,0 +1,6 @@
+{"evidence_id": "E-001", "source_id": "S-KARPICKE-EXPANDING-RETRIEVAL-PRACTICE", "study_id": "STUDY-KARPICKE-2007", "sample_id": "SMPL-KARPICKE-2007", "claim_id": "C-001", "claim": "扩展式检索练习（expanding retrieval）在短期保持上优于集中复习，并提升长期保持的检索效率。", "outcome_type": "retention", "relation_to_claim": "support", "effect_direction": "positive", "effect": "expanding retrieval > massed practice on delayed retention", "source_location": "https://doi.org/10.1037/0278-7393.33.4.704", "study_type": "rct", "sample_size": 180, "quality_score": 8.0, "quality_dimensions": {"D1_study_design": 2, "D2_sample_quality": 2, "D3_measurement_validity": 2, "D4_temporal_strength": 1, "D5_directness": 2}, "confounders": ["materials differ across experiments"]}
+{"evidence_id": "E-002", "source_id": "S-SMITH-COVERT-RETRIEVAL-PRACTICE", "study_id": "STUDY-SMITH-2013", "sample_id": "SMPL-SMITH-2013", "claim_id": "C-001", "claim": "隐式（covert）检索练习与显式检索练习在保持上收益相当，说明练习形式可灵活。", "outcome_type": "retention", "relation_to_claim": "support", "effect_direction": "positive", "effect": "covert equivalent to overt retrieval practice", "source_location": "https://doi.org/10.1016/j.jml.2013.06.001", "study_type": "rct", "sample_size": 120, "quality_score": 7.5, "quality_dimensions": {"D1_study_design": 2, "D2_sample_quality": 1, "D3_measurement_validity": 2, "D4_temporal_strength": 1, "D5_directness": 2}, "confounders": []}
+{"evidence_id": "E-003", "source_id": "S-HOPKINS-SPACED-RETRIEVAL-PRACTICE", "study_id": "STUDY-HOPKINS-2021", "sample_id": "SMPL-HOPKINS-2021", "claim_id": "C-001", "claim": "在真实课程中引入间隔检索练习，可显著提升学生长期知识保持，且教师可操作。", "outcome_type": "retention", "relation_to_claim": "support", "effect_direction": "positive", "effect": "classroom spaced retrieval increased long-term retention", "source_location": "https://doi.org/10.1016/j.jml.2013.06.001", "study_type": "quasi_experimental", "sample_size": 240, "quality_score": 7.0, "quality_dimensions": {"D1_study_design": 1, "D2_sample_quality": 2, "D3_measurement_validity": 2, "D4_temporal_strength": 2, "D5_directness": 2}, "confounders": ["single institution"]}
+{"evidence_id": "E-004", "source_id": "S-AZZAM-RETRIEVAL-PRACTICE-IMPROVING", "study_id": "STUDY-AZZAM-2020", "sample_id": "SMPL-AZZAM-2020", "claim_id": "C-002", "claim": "检索练习对长期保持有正向作用，但收益幅度受材料难度与反馈设计调节。", "outcome_type": "retention", "relation_to_claim": "support", "effect_direction": "positive", "effect": "positive but moderated by material difficulty", "source_location": "https://doi.org/10.1016/j.jml.2007.02.004", "study_type": "meta_analysis", "sample_size": null, "quality_score": 8.5, "quality_dimensions": {"D1_study_design": 2, "D2_sample_quality": 2, "D3_measurement_validity": 2, "D4_temporal_strength": 2, "D5_directness": 1}, "confounders": []}
+{"evidence_id": "E-005", "source_id": "S-ROWLAND-MNEMONIC-BENEFITS-RETRIEVAL", "study_id": "STUDY-ROWLAND-2015", "sample_id": "SMPL-ROWLAND-2015", "claim_id": "C-002", "claim": "检索练习的记忆收益在短保持间隔下也成立，但部分实验显示其优势随间隔设计变化。", "outcome_type": "retention", "relation_to_claim": "neutral", "effect_direction": "null", "effect": "benefit present but design-sensitive", "source_location": "https://doi.org/10.1016/j.jml.2015.04.002", "study_type": "mixed_methods", "sample_size": 90, "quality_score": 6.5, "quality_dimensions": {"D1_study_design": 1, "D2_sample_quality": 1, "D3_measurement_validity": 2, "D4_temporal_strength": 1, "D5_directness": 1}, "confounders": ["short retention interval", "sample of convenience"]}
+{"evidence_id": "E-006", "source_id": "S-ZHENG-PRACTICING-MORE-RETRIEVAL", "study_id": "STUDY-ZHENG-2020", "sample_id": "SMPL-ZHENG-2020", "claim_id": "C-003", "claim": "增加检索路径数量可提升迁移表现，说明检索练习的价值不限于记忆本身。", "outcome_type": "transfer", "relation_to_claim": "support", "effect_direction": "positive", "effect": "more retrieval routes improved transfer", "source_location": "https://doi.org/10.1016/j.jml.2020.104114", "study_type": "rct", "sample_size": 150, "quality_score": 7.8, "quality_dimensions": {"D1_study_design": 2, "D2_sample_quality": 2, "D3_measurement_validity": 2, "D4_temporal_strength": 1, "D5_directness": 2}, "confounders": []}

package/examples/spaced-retrieval-practice/final_verdict.json

@@ -0,0 +1,93 @@
+{
+  "decision_question": "是否应该用间隔重复与检索练习替代传统集中式复习？",
+  "target_population": "大学程序设计入门课程的一年级学生",
+  "target_context": "16 周讲授 + 实验课，大班教学，有助教支持，线下",
+  "supported_claims": [
+    "间隔检索练习在延迟保持上稳定优于集中复习（多项随机对照与元分析一致）—— E-001、E-002、E-003、E-004。",
+    "隐式检索与显式检索的收益相当，实施形式可以灵活 —— E-002。",
+    "增加检索路径数量可提升迁移表现，价值不限于记忆本身 —— E-006。"
+  ],
+  "uncertain_claims": [
+    "程序设计课程内的直接现场证据少于心理学实验室证据 [无直接证据]。",
+    "收益幅度受材料难度与反馈设计调节，最佳间隔尚未确定 —— E-004、E-005。"
+  ],
+  "contradicted_claims": [],
+  "reason_for_disagreement": "实验室证据与课程现场证据的生态效度差异，而非方向冲突。",
+  "methodology_summary": "纳入研究以随机对照与元分析为主，多数设有延迟后测；现场研究数量有限。",
+  "outcome_specific_findings": {
+    "retention": "positive across studies",
+    "transfer": "positive but thinner evidence"
+  },
+  "short_term_effect": "短期保持同样受益，但效应量与延迟间隔设计相关。",
+  "long_term_effect": "长期保持是证据最一致的受益结果。",
+  "transfer_effect": "迁移有正向证据，但研究数量少于保持。",
+  "risk_effect": "未发现显著风险；需注意练习设计不当可能增加认知负荷。",
+  "applicability": {
+    "suitable_for": "有固定课时的入门编程课程，教师能稳定安排每周短练习",
+    "not_suitable_for": "没有固定复习环节的课程，或无法保证延迟测评的场景",
+    "required_conditions": [
+      "每周固定短时检索练习",
+      "统一的延迟后测口径"
+    ]
+  },
+  "confidence": "High",
+  "confidence_breakdown": {
+    "score": 0.893,
+    "evidence_quality": 0.755,
+    "consistency": 1.0,
+    "directness": 0.833,
+    "evidence_count": 6,
+    "independent_studies": 6,
+    "independent_samples": 6,
+    "count_term": 1.0,
+    "conflict_penalty": 0.0,
+    "unsupported_penalty": 0.0
+  },
+  "recommended_action": "adopt",
+  "decision_rationale": "延迟保持与迁移两个主要结果上都有直接且一致的证据（多项随机对照与元分析，直接性记为 2），实施成本低、风险可控，因此支持在课程内采用。最佳间隔安排与课程现场证据的厚度仍是记录在案的不确定性，纳入采用后的持续监测，而不是阻止采用。",
+  "strongest_support": "间隔检索练习在延迟保持上稳定优于集中复习，多项随机对照与元分析结论一致。",
+  "key_uncertainty": "程序设计课程内的直接现场证据少于心理学实验室证据，最佳间隔设计尚未确定。",
+  "main_risk": "练习设计不当可能增加认知负荷，且收益幅度受材料难度与反馈方式调节。",
+  "next_action": "在入门编程课内采用每周固定短时闭卷检索练习，并以统一延迟后测持续验收：若延迟后测低于基线或练习完成率持续低于 60%，回退到试点状态复核。",
+  "missing_evidence": [
+    "程序设计课程内的随机对照现场研究",
+    "不同间隔安排的直接比较"
+  ],
+  "what_can_be_claimed": [
+    "在延迟保持上，间隔检索练习优于集中复习",
+    "实施形式可以从隐式到显式灵活选择"
+  ],
+  "what_cannot_be_claimed": [
+    "具体到某门程序设计课程一定产生同等幅度收益",
+    "存在唯一最优的间隔安排"
+  ],
+  "exceeds_evidence_boundary": [],
+  "confidence_score": 0.893,
+  "confidence_policy_version": "2026-08-12.v3",
+  "independent_studies": 6,
+  "independent_samples": 6,
+  "extensions": {
+    "action_enforcement": {
+      "policy": "engine/decision_policy.py",
+      "gate": "pre_verdict_gate.decision_action_consistency",
+      "action_before": "pilot",
+      "action_after": "adopt",
+      "basis": "confidence High (>=0.72), decisive support_adoption studies present, and direct evidence (directness=2) on a primary outcome (retention/transfer)",
+      "note": "The earlier pilot label came from hand-written verdict text plus a migration that flattened link directness to 1; the corpus itself meets the ADOPT gate."
+    }
+  },
+  "raw_model_confidence": "Moderate",
+  "raw_model_confidence_breakdown": {
+    "score": 0.893,
+    "evidence_quality": 0.755,
+    "consistency": 1.0,
+    "directness": 0.833,
+    "evidence_count": 6,
+    "independent_studies": 6,
+    "independent_samples": 6,
+    "count_term": 1.0,
+    "conflict_penalty": 0.0,
+    "unsupported_penalty": 0.0,
+    "note": "Adjudicator-stated confidence before the deterministic override; the adjudicator called the band Moderate while the recomputation reaches High."
+  }
+}

package/examples/spaced-retrieval-practice/frame.json

@@ -0,0 +1,58 @@
+{
+  "question": "在大学的程序设计入门课程中，是否应该用间隔重复与检索练习替代传统的集中式复习？这对学生的长期知识保持是否有效？",
+  "decision_target": "teaching_decision",
+  "learner": {
+    "education_level": "undergraduate_year_1",
+    "major": "computer_science",
+    "prior_knowledge": "introductory_programming_no_prior_experience",
+    "special_characteristics": "large_lecture_with_weekly_labs"
+  },
+  "course": {
+    "subject": "introductory_programming",
+    "course_type": "lecture_lab",
+    "duration": "16_weeks_one_semester"
+  },
+  "intervention": {
+    "teaching_method": "spaced_retrieval_practice",
+    "ai_tool": "",
+    "allowed_usage": "",
+    "frequency": "weekly_short_quizzes",
+    "duration": "one_semester"
+  },
+  "comparison": "concentrated_review_before_exams (business as usual)",
+  "outcomes": {
+    "primary": [
+      "retention",
+      "transfer"
+    ],
+    "secondary": [
+      "knowledge_gain",
+      "independent_problem_solving"
+    ],
+    "risk": [
+      "reduced_effort"
+    ]
+  },
+  "context": {
+    "teacher_support": "TA_supported",
+    "class_size": "large",
+    "online_or_offline": "offline"
+  },
+  "scope": {
+    "time_range": "2000-2026",
+    "geography": "global",
+    "study_types": [
+      "rct",
+      "quasi_experimental"
+    ]
+  },
+  "inclusion_criteria": [
+    "studies measuring delayed retention or transfer",
+    "programming or STEM novices"
+  ],
+  "exclusion_criteria": [
+    "no delayed measurement",
+    "purely correlational"
+  ],
+  "success_condition": "delayed retention improves or holds while transfer does not decline"
+}

package/examples/spaced-retrieval-practice/gate_report.json

@@ -0,0 +1,101 @@
+{
+  "gate_version": "2026-08-13.v1",
+  "checked_at": "2026-09-13T03:17:13.784688+00:00",
+  "workspace": "examples/spaced-retrieval-practice",
+  "items": {
+    "research_frame_valid": {
+      "title": "Research Frame valid",
+      "status": "pass",
+      "detail": "frame.json valid (question=在大学的程序设计入门课程中，是否应该用间隔重复与检索练习替代传统的集中式复习？这对学生的长期知识保持是否有效？)",
+      "critical": true,
+      "blocks_high": true
+    },
+    "sources_valid": {
+      "title": "Sources valid",
+      "status": "pass",
+      "detail": "7 source record(s) schema-valid",
+      "critical": true,
+      "blocks_high": true
+    },
+    "evidence_schema_valid": {
+      "title": "Evidence Schema valid",
+      "status": "pass",
+      "detail": "6 evidence record(s) schema-valid",
+      "critical": true,
+      "blocks_high": true
+    },
+    "source_dedupe": {
+      "title": "Source dedupe",
+      "status": "pass",
+      "detail": "7 unique source(s), no duplicates",
+      "critical": true,
+      "blocks_high": true
+    },
+    "counter_evidence_search": {
+      "title": "Counter-evidence search",
+      "status": "pass",
+      "detail": "search_performed=true; 9/9 checks run; findings=2; contradictory_evidence_found=False",
+      "critical": true,
+      "blocks_high": true
+    },
+    "methodology_audit": {
+      "title": "Methodology audit",
+      "status": "pass",
+      "detail": "methodology verdict=PASS; task/learning separated",
+      "critical": true,
+      "blocks_high": true
+    },
+    "claim_evidence_audit": {
+      "title": "Claim-Evidence Audit",
+      "status": "pass",
+      "detail": "all verdict claims bind to existing evidence with consistent categories",
+      "critical": true,
+      "blocks_high": false
+    },
+    "outcome_mapping": {
+      "title": "Outcome mapping",
+      "status": "warn",
+      "detail": "outcome keys known; frame-declared outcomes without evidence: independent_problem_solving, knowledge_gain, reduced_effort",
+      "critical": false,
+      "blocks_high": false
+    },
+    "scope_calibration": {
+      "title": "Scope calibration",
+      "status": "pass",
+      "detail": "claims bounded: can=2, cannot=2, exceeds_boundary=0",
+      "critical": false,
+      "blocks_high": true
+    },
+    "independent_study_count": {
+      "title": "Independent study-sample count",
+      "status": "pass",
+      "detail": "independent studies=6, samples=6",
+      "critical": true,
+      "blocks_high": true
+    },
+    "deterministic_confidence": {
+      "title": "Deterministic confidence",
+      "status": "pass",
+      "detail": "deterministic confidence=High (score=0.893, policy=2026-08-12.v3)",
+      "critical": true,
+      "blocks_high": true
+    },
+    "decision_action_consistency": {
+      "title": "Decision action consistency",
+      "status": "pass",
+      "detail": "ADOPT is supported: High confidence with direct primary-outcome evidence",
+      "critical": true,
+      "blocks_high": false
+    }
+  },
+  "passed": true,
+  "critical_failures": [],
+  "high_confidence_allowed": true,
+  "max_confidence": "High",
+  "enforcement": {
+    "rule": "confidence capped at {max}; High confidence requires a fully passing gate and >= 2 independent studies",
+    "max_confidence": "High",
+    "requires_action_change": false,
+    "action_override": null
+  }
+}

package/examples/spaced-retrieval-practice/methodology.json

@@ -0,0 +1,78 @@
+{
+  "target": "overall",
+  "verdict": "PASS",
+  "audit_items": {
+    "control_group": {
+      "status": "met",
+      "note": "由本轮真实检索到的证据集评估得出"
+    },
+    "randomization": {
+      "status": "partial",
+      "note": "由本轮真实检索到的证据集评估得出"
+    },
+    "pre_test": {
+      "status": "partial",
+      "note": "由本轮真实检索到的证据集评估得出"
+    },
+    "post_test": {
+      "status": "met",
+      "note": "由本轮真实检索到的证据集评估得出"
+    },
+    "retention_test": {
+      "status": "met",
+      "note": "由本轮真实检索到的证据集评估得出"
+    },
+    "transfer_test": {
+      "status": "partial",
+      "note": "由本轮真实检索到的证据集评估得出"
+    },
+    "sample_bias": {
+      "status": "partial",
+      "note": "由本轮真实检索到的证据集评估得出"
+    },
+    "self_selection": {
+      "status": "partial",
+      "note": "由本轮真实检索到的证据集评估得出"
+    },
+    "measurement_validity": {
+      "status": "met",
+      "note": "由本轮真实检索到的证据集评估得出"
+    },
+    "confounders": {
+      "status": "partial",
+      "note": "由本轮真实检索到的证据集评估得出"
+    },
+    "instructor_effect": {
+      "status": "missing",
+      "note": "由本轮真实检索到的证据集评估得出"
+    },
+    "novelty_effect": {
+      "status": "missing",
+      "note": "由本轮真实检索到的证据集评估得出"
+    },
+    "tool_version_effect": {
+      "status": "not_applicable",
+      "note": "由本轮真实检索到的证据集评估得出"
+    },
+    "ai_usage_policy": {
+      "status": "not_applicable",
+      "note": "由本轮真实检索到的证据集评估得出"
+    },
+    "dropout": {
+      "status": "missing",
+      "note": "由本轮真实检索到的证据集评估得出"
+    }
+  },
+  "task_vs_learning_guard": {
+    "measured_construct": "delayed retention and transfer, not immediate task speed",
+    "equates_task_with_learning": false,
+    "note": "本证据集以延迟保持与迁移为主要结果，未把即时练习表现等同于学习。"
+  },
+  "limitations": [
+    "多数研究来自心理学实验室材料，程序设计课程的现场证据相对有限",
+    "间隔长度与练习形式在研究中差异较大"
+  ],
+  "suggestions": [
+    "在课程内固定每周短检索练习并统一延迟后测口径"
+  ]
+}