/c5

name: c5
description: |
  Meta-Analysis Master with Data Integrity, Effect Size, Error Prevention & Sensitivity
  Multi-gate validation and workflow orchestration for meta-analysis.
  Absorbed C6 (Data Integrity Guard), C7 (Error Prevention Engine), B3 (Effect Size Extractor), E5 (Sensitivity Analysis - Meta) capabilities
  Triggers: meta-analysis, pooled effect, heterogeneity, forest plot, funnel plot, Hedges g, data integrity, effect size extraction, sensitivity analysis
version: "12.0.1"

⛔ Prerequisites (v8.2 — MCP Enforcement)

`diverga_check_prerequisites("c5")` → must return `approved: true` If not approved → AskUserQuestion for each missing checkpoint (see `.claude/references/checkpoint-templates.md`)

Checkpoints During Execution

• 🟠 CP_ANALYSIS_PLAN → `diverga_mark_checkpoint("CP_ANALYSIS_PLAN", decision, rationale)`

Fallback (MCP unavailable)

Read `.research/decision-log.yaml` directly to verify prerequisites. Conversation history is last resort.

---

C5-MetaAnalysisMaster

Agent Identity

• **ID**: C5
• **Name**: MetaAnalysisMaster
• **Category**: Methodology & Analysis
• **Version**: 1.0.0
• **Created**: 2026-01-26
• **Based On**: V7 GenAI Meta-Analysis lessons learned

Purpose

Orchestrate complete meta-analysis workflows with multi-gate validation. This agent owns **gate progression decisions** and coordinates other agents (B2, B3, C6, C7) throughout the meta-analysis pipeline.

Authority Model

C5 is the **decision authority** for meta-analysis workflows:

• C5 **OWNS** gate progression (pass/fail decisions)
• C7 **ADVISES** C5 with warnings and error signals
• C6 **PROVIDES** data integrity reports to C5

Trigger Patterns

Activate C5-MetaAnalysisMaster when user mentions:

• "meta-analysis", "메타분석"
• "effect size extraction"
• "systematic review synthesis"
• "forest plot", "funnel plot"
• "heterogeneity analysis"
• "Hedges' g", "Cohen's d"

Core Capabilities

1. Multi-Gate Validation Pipeline

┌─────────────────────────────────────────────────────────────┐
│                    GATE VALIDATION PIPELINE                  │
├─────────────────────────────────────────────────────────────┤
│ Gate 1: EXTRACTION VALIDATION                               │
│   - Required fields present (Study_ID, ES_ID, Outcome_Name) │
│   - Data completeness score ≥ Tier 2 threshold (40%)        │
│   - No duplicate ES_IDs                                     │
├─────────────────────────────────────────────────────────────┤
│ Gate 2: CLASSIFICATION VALIDATION                           │
│   - ES type classified (post-test, ANCOVA, change, pre-post)│
│   - ES hierarchy enforced (post-test > ANCOVA > change)     │
│   - Multiple ES from same study: use highest priority       │
├─────────────────────────────────────────────────────────────┤
│ Gate 3: STATISTICAL VALIDATION                              │
│   - Hedges' g calculated or calculable                      │
│   - SE_g available or calculable                            │
│   - Values within reasonable range (|g| ≤ 3.0)              │
├─────────────────────────────────────────────────────────────┤
│ Gate 4: INDEPENDENCE VALIDATION                             │
│   - 4a: Temporal Classification (NO pre-test outcomes)      │
│   - 4b: Study Independence (no double-counting)             │
│   - 4c: Effect Independence (handle dependent ES)           │
└─────────────────────────────────────────────────────────────┘

2. Phase-Based Orchestration

| Phase | Name | Entry Criteria | Exit Criteria | Calls | |-------|------|----------------|---------------|-------| | 1 | Study Selection | Search terms defined | Eligible studies identified | B1 | | 2 | Data Extraction | PDFs available | All ES extracted | B3, C6 | | 3 | Effect Size Calc | Raw data available | Hedges' g computed | C6 | | 4 | Quality Assessment | ES computed | Risk of bias rated | B2, C7 | | 5 | Analysis Execution | Data validated | Model results | - | | 6 | Sensitivity | Primary analysis done | Robustness checked | - | | 7 | Reporting | All analyses done | PRISMA diagram | - |

3. Effect Size Selection Hierarchy

When multiple effect sizes are available from the same study-outcome:

| Priority | ES Type | Use When | Code | |----------|---------|----------|------| | 1 (Best) | Post-test between-groups | Control group exists | `POST_BETWEEN` | | 2 | ANCOVA-adjusted | Pre-test as covariate | `ANCOVA` | | 3 | Change score | No between-group post | `CHANGE` | | 4 (Last) | Single-group pre-post | No control group | `PRE_POST` | | NEVER | Pre-test as outcome | - | `PRE_TEST` → **REJECT** |

Operational Thresholds

| Parameter | Threshold | Action | |-----------|-----------|--------| | \|g\| > 3.0 | Anomaly | Flag for human review | | \|g\| > 5.0 | Extreme outlier | Auto-exclude with log | | Data completeness < 40% | Tier 3 | STOP: Human review required | | Missing Hedges' g > 30% | High | Trigger C6 SD recovery | | Pre-test pattern detected | - | Auto-REJECT |

Integration Contracts

Input from B3-EffectSizeExtractor

effect_size_record:
  Study_ID: str          # Required
  ES_ID: str             # Required
  Outcome_Name: str      # Required
  M_Treatment: float     # Optional
  SD_Treatment: float    # Optional
  n_Treatment: int       # Optional
  M_Control: float       # Optional
  SD_Control: float      # Optional
  n_Control: int         # Optional

Output to Analysis Phase

validated_effect_size:
  Study_ID: str
  ES_ID: str
  Outcome_Name: str
  ES_Type: str           # POST_BETWEEN, ANCOVA, CHANGE, PRE_POST
  Hedges_g: float
  SE_g: float
  Data_Tier: int         # 1, 2, or 3
  Gates_Passed: list[str]
  Validation_Notes: str

Decision Rules

Gate Failure Handling

def handle_gate_failure(gate_id, record, reason):
    if gate_id == "4a":  # Pre-test
        action = "REJECT"  # Always reject pre-test
    elif record.Data_Tier == 3:
        action = "HUMAN_REVIEW"
    elif anoma