Audit Trails & Traceability

Every decision in EvidAI—whether made by AI or human—is captured in an immutable audit trail designed for regulatory scrutiny and complete reproducibility.

---

Why Audit Trails Matter

In pharmaceutical evidence synthesis, the ability to defend every decision is not optional—it's required. Traditional tools leave gaps; EvidAI provides complete traceability.

---

What We Capture

Every Event Type

AI Decision Documentation

When AI participates in decisions, we capture complete reasoning:

AI SCREENING DECISION RECORD
════════════════════════════════════════════════════════════════

Study: "Randomized trial of empagliflozin in heart failure"
PMID: 32847529
Decision: INCLUDE
Confidence: 96.8%

MODEL VOTING BREAKDOWN:
┌──────────────────┬──────────┬────────────┬─────────────────────┐
│ Model            │ Decision │ Confidence │ Key Reasoning       │
├──────────────────┼──────────┼────────────┼─────────────────────┤
│ GPT-4o           │ Include  │ 97%        │ RCT design matches  │
│ Claude-3.5       │ Include  │ 96%        │ Population eligible │
│ Gemini-1.5       │ Include  │ 98%        │ Outcome relevant    │
│ EvidAI-Custom    │ Include  │ 96%        │ All criteria met    │
└──────────────────┴──────────┴────────────┴─────────────────────┘

CRITERIA EVALUATION:
├── Population (Adults with HFrEF): MATCH
│   └── Evidence: "patients aged ≥18 years with heart failure"
├── Intervention (SGLT2 inhibitor): MATCH
│   └── Evidence: "empagliflozin 10 mg once daily"
├── Comparator (Placebo/SOC): MATCH
│   └── Evidence: "compared with placebo"
├── Outcome (Hospitalization): MATCH
│   └── Evidence: "primary outcome of cardiovascular death or 
│       hospitalization for heart failure"
└── Study Design (RCT): MATCH
    └── Evidence: "randomized, double-blind, placebo-controlled"

AUTOMATIC DISPOSITION: Included (confidence >95%, unanimous)
HUMAN REVIEW: Not required

════════════════════════════════════════════════════════════════

---

Audit Trail Features

Immutability

Once logged, audit records cannot be modified or deleted:

Searchability

Find any decision instantly:

AUDIT SEARCH

Query: Screening decisions for study "DAPA-HF"

Results:
┌─────────────────────────────────────────────────────────────┐
│ 2024-03-15 09:23:14 │ AI Screening │ INCLUDE (97.2%)       │
│ 2024-03-15 09:23:17 │ Human Review │ CONFIRMED by j.smith  │
│ 2024-03-18 14:45:33 │ Data Extract │ 12 fields by m.jones  │
│ 2024-03-18 15:12:08 │ Verification │ VERIFIED by s.chen    │
│ 2024-03-22 11:30:45 │ RoB Assess   │ LOW RISK by j.smith   │
│ 2024-03-22 11:35:22 │ RoB Review   │ AGREED by m.jones     │
└─────────────────────────────────────────────────────────────┘

[View Full Details] [Export Selection] [Generate Report]

Reconstruction

Rebuild any point-in-time view:

Time Travel: Auditors can see exactly what the review looked like at any moment in time—useful for investigating decisions or demonstrating the review evolution.

---

Regulatory Export Formats

FDA Submission Package

FDA AUDIT EXPORT
Generated: 2024-12-22 16:45:00 UTC
Review: SLR-2024-001
Exporter: Dr. Sarah Chen (authorized)

Package Contents:
├── 00_Executive_Summary.pdf (2 pages)
├── 01_Protocol_History/
│   ├── Protocol_v1.0.pdf
│   ├── Protocol_v1.1.pdf (with amendment rationale)
│   └── Protocol_v2.0.pdf (final approved)
├── 02_Search_Documentation/
│   ├── Search_Strategies_All_Databases.pdf
│   ├── Search_Execution_Log.pdf
│   └── Database_Screenshots/ (timestamped)
├── 03_Screening_Audit/
│   ├── Screening_Decision_Log.pdf (all 4,287 decisions)
│   ├── AI_Voting_Records.xml
│   ├── Human_Review_Decisions.pdf
│   └── Disagreement_Resolution.pdf
├── 04_Extraction_Verification/
│   ├── Extraction_Forms_Completed.pdf
│   ├── Source_Attribution_Report.pdf
│   └── Discrepancy_Resolution.pdf
├── 05_Quality_Assessment/
│   ├── RoB_Assessments_All_Studies.pdf
│   ├── RoB_Supporting_Quotes.pdf
│   └── GRADE_Certainty_Assessment.pdf
├── 06_Analysis_Documentation/
│   ├── Statistical_Analysis_Plan.pdf
│   ├── Analysis_Code_Audit.txt
│   └── Sensitivity_Analyses.pdf
├── 07_Signature_Records/
│   ├── Electronic_Signatures.pdf
│   └── Signature_Certificates.xml
└── 08_System_Validation/
    ├── Validation_Summary.pdf
    └── System_Configuration.pdf

Total Files: 147
Total Size: 234 MB
Integrity Hash: sha256:a7b3c9d4e5f6...

[Download Package] [Send to FDA Gateway]

---

Best Practices for Audit Trail Management

During the Review

1. Complete All Fields: Missing data = audit gaps
2. Document Rationale: Especially for judgment calls
3. Use Structured Notes: Free text is harder to analyze
4. Review Regularly: Catch issues early

For Regulatory Submission

Tip: Run the "Submission Readiness Check" tool before finalizing—it identifies common audit gaps and missing signatures.