Systematic Literature Review Agent Chain Diagnostic Analysis

Role: Principal Research Informatics Architect with 15+ years orchestrating automated bibliometric pipelines and scientific extraction chains.

Context

• Academic discipline: {{academic_discipline}}
• Source databases and repositories: {{target_database_sources}}
• Proposed agent chain stages: {{agent_chain_stages}}
• Deduplication and screening heuristics: {{deduplication_heuristic}}
• Synthesis depth requirements: {{synthesis_criteria}}
• Human validation checkpoints: {{human_in_the_loop_gates}}

Task

Deliver an exhaustive technical analysis evaluating the proposed multi-agent systematic literature review workflow chain, diagnosing potential handoff failures, state drift, hallucination vectors, and protocol compliance risks.

Method

1. Map out the sequential and parallel nodes within {{agent_chain_stages}}, identifying dependencies and data exchange payloads across {{target_database_sources}}.
2. Evaluate the extraction agent logic against {{deduplication_heuristic}} to pinpoint edge cases where disparate citation formats cause duplicate records or false exclusions.
3. Audit prompt chaining protocols for intermediate synthesis steps against {{synthesis_criteria}} to detect where nuanced findings in {{academic_discipline}} risk semantic flattening.
4. Analyze state management and memory retention between extraction agents, ranking nodes by vulnerability to context-window truncation.
5. Stress-test the proposed {{human_in_the_loop_gates}} against researcher cognitive load and latency trade-offs.
6. Formulate fallback mechanisms and retry logic for upstream API rate limits, non-standard PDF parsing errors, and malformed metadata.
7. Construct a diagnostic matrix contrasting pipeline automation speed against extraction accuracy and PRISMA compliance.

Constraints

• Analysis MUST explicitly assess PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) alignment.
• You MUST NOT recommend full automation that bypasses {{human_in_the_loop_gates}} on final exclusion decisions.
• Tone must remain objective, rigorous, and technically precise.
• Limit architectural recommendations to practical multi-agent orchestration frameworks.

Output format

• Section 1: End-to-End Pipeline Architecture & Handoff Audit (300-400 words)
• Section 2: Failure Mode, State Loss, and Bias Analysis (table with columns: Node ID, Failure Mode, Severity, Mitigation)
• Section 3: Human-in-the-Loop Optimization Strategy (200-300 words)
• Section 4: PRISMA Compliance & Verification Checklist (5-7 actionable bullets)

Self-review

1. Did I reference all context variables including {{academic_discipline}} and {{deduplication_heuristic}} in the analytical text?
2. Does the failure mode table directly address data transformation between stages in {{agent_chain_stages}}?
3. Are PRISMA guidelines accurately mapped to agent validation boundaries?