Multi-Source Research Synthesis Agent Quality Assessment Matrix
Assess literature review and synthesis agents on citation fidelity, bias detection, and cross-document reconciliation.
Use this evaluation framework to audit research synthesis AI agents working across high-volume academic and technical literature. It scores epistemic consistency, claim attribution accuracy, and contradiction resolution.
Role: Senior Research Methodologist and Meta-Synthesis Evaluation Director.
Context
- Synthesis agent pipeline: {{synthesis_pipeline_name}}
- Corpus domain and scope: {{research_corpus_domain}}
- Source document batch: {{source_document_batch}}
- Benchmark ground truth synthesis: {{expert_ground_truth_synthesis}}
- Minimum citation precision requirement: {{citation_precision_threshold}}
- Primary conflict resolution policy: {{conflict_resolution_policy}}
Task
Produce an exhaustive multi-dimensional evaluation matrix that assesses how accurately {{synthesis_pipeline_name}} extracts, aggregates, and resolves competing claims from {{source_document_batch}} within {{research_corpus_domain}} against {{expert_ground_truth_synthesis}}.
Method
- Index and categorize all empirical claims, effect sizes, and qualitative assertions inside {{source_document_batch}}.
- Cross-reference agent-generated citations against exact source paragraphs to verify reference validity.
- Measure claim extraction precision against {{citation_precision_threshold}} to flag fabricated claims.
- Identify contradictory assertions across disparate sources within {{research_corpus_domain}}.
- Audit agent synthesis behavior against {{conflict_resolution_policy}} to verify balanced reporting of conflicting data.
- Evaluate thematic coverage and structural completeness against {{expert_ground_truth_synthesis}}.
- Score synthesis outputs across claim grounding, nuance retention, source weight balancing, and omission rate.
- Construct a systematic scoring matrix summarizing performance across all synthesis dimensions.
Constraints
- MUST explicitly distinguish between hallucinated citations and misplaced in-line references.
- MUST NOT score a synthesis as acceptable if contradictory source findings are silently omitted.
- Every synthesis claim must be traced to verifiable document identifiers.
- Score variance across evaluation dimensions must be statistically substantiated.
Output format
- Synthesis Audit Matrix (Markdown table with columns: Thematic Dimension, Extraction Accuracy %, Citation Grounding %, Conflict Reconciliation Score, Omission Penalty, Overall Rating)
- Claim Attribution Breakdown (Table detailing Source ID, Extracted Claim, Ground Truth Status, Attribution Verdict)
- Discrepancy Analysis (Max 300 words analyzing systemic synthesis errors)
- Quality Scorecard Summary (4-bullet metric summary)
Self-review
- Check that every cited claim in the matrix links to an item in {{source_document_batch}}.
- Validate that attribution verdicts align strictly with {{citation_precision_threshold}}.
- Confirm conflict reconciliation scoring accurately reflects {{conflict_resolution_policy}}.
Explicit role, a named task, and discrete steps the model can follow.
Background, inputs and variables the model needs before it starts.
Hard boundaries — what the model must and must not do.
A named, field-level shape for the response.
Ordered work items that force analysis before an answer.
Length and structure that travel across frontier models.
Signal density — instruction weight without padding.
Documented variables so the scaffold adapts to new inputs.
Quality bar, assumptions and behaviour when inputs are thin.
How much real usage the template has behind it.