Date Shift Audit Report

[AndyRae]

Is this the right issue type?

• Yes, I'm planning work for this project team.

Summary

For studies, we need formalised auditability around our date shifting process.

We should generate a structured Date Shifting Audit Report automatically whenever the date shifting pipeline is executed.

The goal is to demonstrate:

• Determinism and reproducibility
• Consistent application
• Data integrity preservation
• Traceability to code version and run context

---

Content

1. Run Metadata

Include:

• Study ID
• Dataset version
• Extraction date/time (UTC)
• Git commit hash
• Environment (dev/test/prod)
• Operator or service account
• Source snapshot reference if applicable

Purpose: anchor the run in time and code version.

---

2. Date Shift Configuration

Document the exact configuration used:

• Per-patient random offset
• Per-study fixed offset
• Offset range (e.g. ±180 days)
• Random seed used
• Unit of shift (days)
• Direction symmetry

Purpose: ensure reproducibility and clarity of implementation.

---

3. Scope of Application

Explicitly list:

• Date fields shifted
• Date fields excluded
• Any preserved anchor/index dates
• Handling of partial dates (e.g. year-only)

Purpose: show intentional application of shifting rules.

---

4. Statistical Summary

Generate summary statistics:

• Number of patients shifted
• Number of records shifted
• Min offset
• Max offset
• Mean offset
• Distribution of offsets (bucketed counts)
• Null or skipped date counts

Purpose: demonstrate controlled and expected distribution.

---

5. Integrity Checks

Automated validation checks with pass/fail status:

• Chronological order preserved within patient
• No negative durations introduced
• No dates shifted outside allowed study window
• No system boundary violations
• Referential integrity unchanged

Purpose: confirm no unintended data distortion.

---

6. Reproducibility Statement

Include statement:

This dataset can be regenerated by executing commit with seed against source snapshot .

Purpose: explicit reproducibility assurance.

---

7. Output Fingerprint

Include:

• Row count
• Column count
• SHA256 hash of final dataset

Purpose: detect post-generation alteration.

---

Deliverables

• JSON audit artifact stored with study outputs
• Human-readable summary (markdown or PDF)

---

Non-Goals

• Do not include per-patient offset mapping in report
• Do not expose identifiable data
• Do not store sensitive seeds outside secure context

Acceptance Criteria

• When the date shifting process runs, generate a structured audit report (JSON + human-readable summary).
• Audit report generated automatically during date shift run
• Report includes all required metadata and validation checks
• Output dataset hash reproducible across reruns with same seed
• Tests pass.
• Documentation added.

Confirm creation

• Ready to create this feature?