Files
gh-lyndonkl-claude/skills/research-claim-map/resources/evaluators/rubric_research_claim_map.json
2025-11-30 08:38:26 +08:00

160 lines
10 KiB
JSON

{
"name": "Research Claim Map Evaluator",
"description": "Evaluates claim verification for precise claim definition, evidence triangulation, source credibility assessment, and confidence calibration",
"criteria": [
{
"name": "Claim Precision and Testability",
"weight": 1.4,
"scale": {
"1": "Vague claim with undefined terms, no specific metrics, untestable",
"2": "Somewhat specific but missing key details (timeframe, scope, or metrics unclear)",
"3": "Specific claim with most terms defined, testable with some clarification needed",
"4": "Precise claim with numbers/dates/scope, clear terms, fully testable",
"5": "Exemplary: Reformulated from vague to specific, key terms defined explicitly, potential ambiguities addressed, claim decomposed into testable sub-claims if complex"
}
},
{
"name": "Evidence Triangulation Quality",
"weight": 1.5,
"scale": {
"1": "Single source or only sources agreeing with claim (no contradicting evidence sought)",
"2": "Multiple sources but all similar type/origin (not truly independent)",
"3": "2-3 independent sources for and against, basic triangulation",
"4": "3+ independent sources with different methodologies, both supporting and contradicting evidence gathered",
"5": "Exemplary: Systematic triangulation across source types (primary + secondary), methodologies (quantitative + qualitative), perspectives (proponent + skeptic), active search for disconfirming evidence, circular citations identified and excluded"
}
},
{
"name": "Source Credibility Assessment",
"weight": 1.4,
"scale": {
"1": "No credibility evaluation or accepted sources at face value",
"2": "Basic credibility check (author identified) but no depth (expertise, bias, track record ignored)",
"3": "Credibility assessed on 2-3 factors (e.g., expertise and independence) with brief reasoning",
"4": "All four factors assessed (expertise, independence, track record, methodology) for each major source with clear reasoning",
"5": "Exemplary: Systematic CRAAP test or equivalent, conflicts of interest identified, track record researched (retractions, corrections), methodology transparency evaluated, source bias explicitly noted and accounted for in weighting"
}
},
{
"name": "Evidence Quality Rating",
"weight": 1.3,
"scale": {
"1": "No distinction between evidence types (treats social media post = peer-reviewed study)",
"2": "Vague quality labels ('good source', 'reliable') without systematic rating",
"3": "Evidence categorized (primary/secondary/tertiary) but inconsistently or without justification",
"4": "Systematic rating using evidence hierarchy, each source classified with brief rationale",
"5": "Exemplary: Evidence rated on multiple dimensions (type, methodology, sample size, recency), quality justification detailed, limitations of even high-quality evidence acknowledged"
}
},
{
"name": "Limitations and Gaps Documentation",
"weight": 1.3,
"scale": {
"1": "No limitations noted or only minor caveats mentioned",
"2": "Generic limitations ('more research needed') without specifics",
"3": "Some limitations noted (missing data or assumptions stated) but incomplete",
"4": "Comprehensive limitations: gaps identified, assumptions stated, quality concerns noted, unknowns distinguished from known contradictions",
"5": "Exemplary: Systematic gap analysis (what evidence expected but not found), assumptions explicitly tested for sensitivity, distinction made between 'no evidence found' vs 'evidence of absence', suggestions for what would increase confidence"
}
},
{
"name": "Confidence Calibration and Reasoning",
"weight": 1.4,
"scale": {
"1": "No confidence level stated or vague ('probably', 'likely')",
"2": "Confidence stated but not justified or clearly miscalibrated (100% on weak evidence or 10% on strong)",
"3": "Numeric confidence (0-100%) stated with basic reasoning",
"4": "Well-calibrated confidence with clear reasoning linking to evidence quality, source credibility, and limitations",
"5": "Exemplary: Confidence range provided (not point estimate), reasoning traces from base rates through evidence to posterior, sensitivity analysis shown (confidence under different assumptions), explicit acknowledgment of uncertainty"
}
},
{
"name": "Bias Detection and Mitigation",
"weight": 1.2,
"scale": {
"1": "Clear bias (cherry-picked evidence, ignored contradictions, one-sided presentation)",
"2": "Unintentional bias (confirmation bias evident, didn't actively seek contradicting evidence)",
"3": "Some bias mitigation (acknowledged contradicting evidence, noted potential biases)",
"4": "Active bias mitigation (sought disconfirming evidence, considered alternative explanations, acknowledged own potential biases)",
"5": "Exemplary: Systematic bias checks (CRAAP test, conflict of interest disclosure), actively argued against own hypothesis, considered base rates to avoid availability/anchoring bias, source framing bias identified and corrected"
}
},
{
"name": "Actionable Recommendation",
"weight": 1.2,
"scale": {
"1": "No recommendation or action unclear ('interesting finding', 'needs more study')",
"2": "Vague action ('be cautious', 'consider this') without decision guidance",
"3": "Basic recommendation (believe/reject/uncertain) but not tied to decision context",
"4": "Clear actionable recommendation linked to confidence level and decision context (what to do given uncertainty)",
"5": "Exemplary: Recommendation with decision thresholds ('if you need 80%+ confidence to act, don't proceed; if 60% sufficient, proceed with mitigation X'), contingency plans for uncertainty, clear next steps to increase confidence if needed"
}
}
],
"guidance": {
"by_claim_type": {
"vendor_claims": {
"recommended_approach": "Seek independent verification (analysts, customer references, trials), avoid relying solely on vendor sources",
"evidence_priority": "Primary (customer data, trials) > Secondary (analyst reports) > Tertiary (vendor press releases)",
"red_flags": ["Only vendor sources", "Vague metrics ('up to X%')", "Cherry-picked case studies", "No independent verification"]
},
"news_claims": {
"recommended_approach": "Trace to primary source, check multiple outlets, verify context and framing",
"evidence_priority": "Primary sources (official statements, documents) > Secondary (news reports citing primary) > Tertiary (opinion pieces)",
"red_flags": ["Single source", "Anonymous claims without corroboration", "Circular reporting (outlets citing each other)", "Out-of-context quotes"]
},
"research_claims": {
"recommended_approach": "Check replication studies, meta-analyses, assess methodology rigor, look for conflicts of interest",
"evidence_priority": "RCTs/meta-analyses > Observational studies > Expert opinion",
"red_flags": ["Single study", "Small sample", "Conflicts of interest undisclosed", "P-hacking indicators", "Correlation claimed as causation"]
},
"statistical_claims": {
"recommended_approach": "Verify methodology, sample size, confidence intervals, base rates, check for denominators",
"evidence_priority": "Transparent methodology with raw data > Summary statistics > Infographics without sources",
"red_flags": ["Denominator unclear", "Cherry-picked timeframes", "Correlation ≠ causation", "No confidence intervals", "Misleading visualizations"]
}
}
},
"common_failure_modes": {
"confirmation_bias": {
"symptom": "Only evidence supporting claim found, contradictions ignored or dismissed",
"root_cause": "Want claim to be true (motivated reasoning), didn't actively search for disconfirmation",
"fix": "Actively search 'why this might be wrong', assign someone to argue against, seek base rates for skepticism"
},
"authority_bias": {
"symptom": "Accepted claim because prestigious source (Nobel Prize, Harvard, Fortune 500) without evaluating evidence",
"root_cause": "Heuristic: prestigious source = truth (often valid but not always)",
"fix": "Evaluate evidence quality independently, check if expert in this specific domain, verify claim not opinion"
},
"single_source_overconfidence": {
"symptom": "High confidence based on one source, even if high quality",
"root_cause": "Didn't triangulate, assumed quality source = truth",
"fix": "Require 2-3 independent sources for high confidence, check for replication/corroboration"
},
"vague_confidence": {
"symptom": "Used words ('probably', 'likely', 'seems') instead of numbers (60%, 75%)",
"root_cause": "Uncomfortable quantifying uncertainty, didn't calibrate",
"fix": "Force numeric confidence (0-100%), use calibration guides, test against base rates"
},
"missing_limitations": {
"symptom": "Conclusion presented without caveats, gaps, or unknowns acknowledged",
"root_cause": "Focused on what's known, didn't systematically check for what's unknown",
"fix": "Template section for limitations forces documentation, ask 'what would make me more confident?'"
}
},
"excellence_indicators": [
"Claim reformulated from vague to specific, testable assertion",
"Evidence triangulated: 3+ independent sources, multiple methodologies",
"Source credibility systematically assessed: expertise, independence, track record, methodology",
"Evidence rated using clear hierarchy: primary > secondary > tertiary",
"Contradicting evidence actively sought and fairly presented",
"Limitations and gaps comprehensively documented",
"Assumptions stated explicitly and tested for sensitivity",
"Confidence calibrated numerically (0-100%) with reasoning",
"Recommendation actionable and tied to decision context",
"Bias mitigation demonstrated (sought disconfirming evidence, checked conflicts of interest)",
"Distinction made between 'no evidence found' and 'evidence of absence'",
"Sources properly cited with links/references for verification"
]
}