1. Purpose
This SOP governs the correct use of the HCM-AO AI Talent Pool Simulator & Stress-Tester. The Simulator is a pre-deployment validation tool that generates synthetic candidate CVs, runs them through the HCM-AO AI parsing and ranking pipeline, and measures the pipeline's reliability under controlled, statistically modelled conditions.
2. Scope
This procedure applies to:
| Role | Use Case |
|---|---|
| QA Engineers | Validating a new deployment of HCM-AO before go-live |
| HR Technology Teams | Benchmarking AI recruitment tools |
| Data Scientists | Studying the impact of talent pool variance on algorithmic outcomes |
| Compliance Officers | Seeking formal certification prior to operational use |
3. Prerequisites
| Requirement | Detail |
|---|---|
| Access | HCM-AO application running (local or Cloud Run URL) |
| Job Description File | PDF or TXT of the role being simulated. Must be uploaded before generating the batch. |
| Browser | Any modern browser. Hard-refresh (Ctrl+Shift+R) after any system update.
|
4. Simulator Parameters
4.1 Mean (μ) — Years of Experience Target
Controls the centre of the talent distribution. The majority of generated candidates will have a YOE close to this value.
| μ Value | Meaning |
|---|---|
| 1–3 | Entry-level talent pool (junior roles) |
| 4–7 | Mid-level pool (experienced professionals) |
| 8–15 | Senior / specialist pool |
| 15–30 | Executive / highly specialised pool |
4.2 Standard Deviation (σ) — Diversity & Stress Level
σ controls both the spread of candidate profiles AND the intensity of adversarial trap injection. This is the most critical parameter.
| σ Range | Label | Trap Probability / CV | Certificate Eligible? |
|---|---|---|---|
| 0 | Uniform Pool | 0% | ❌ |
| 1–3 | Low Entropy | 7%–20% | ❌ |
| 4–6 | Moderate Entropy | 27%–40% | ❌ |
| 7 | High Entropy | ~47% (boundary) | ❌ |
| > 7 (8–12) | Stress Test Mode 🔴 | 53%–80% | ✅ Yes |
| > 12 | Extreme / Pathological ⚠️ | >80% | ✅ Yes |
4.3 Batch Size
Number of CVs to generate (1–200). Recommended minimum for certification: 50 CVs. Standard operational test: 20–30 CVs.
5. Trap Types Injected by the System
| Trap Type | Description | Metric Tested |
|---|---|---|
| 🕵️ White Text / Hidden Keywords | Invisible keyword stuffing injected into CV text | SEO Resilience |
| ⚠️ Date Conflict | Employment dates deliberately reversed (start > end) | Data Integrity |
| 🔡 Encoding Noise | Zero-width Unicode characters injected into the name field | Parsing Accuracy |
Trap probability per CV = min(σ / 15.0, 0.80) — mirrors the backend logic exactly.
6. Procedure
-
Navigate to the AI Simulator tab
Click "🧪 AI Simulator & Stress-Test" in the top navigation. -
Configure Bell Curve Parameters
Set μ and σ using the sliders. Observe the live stress-level indicator below the σ slider update in real-time. -
Upload the Target Job Description
Click the upload zone and select a PDF or TXT file. A green toast "✅ JD file ready: [filename]" confirms the upload. -
Set Core Skills Count and Batch Size
Choose Core Skills Count from the dropdown and set the Count input. -
Generate the Batch
Click "Generate Mock CV Batch". A progress bar will appear. -
Review Generated CVs
Click any candidate row to view their full simulated CV. -
Optional: Run in S&R Mode
Click "🚀 Run in Selection & Recruiting Mode" to feed the batch directly into the ranking engine and see impact results. -
Export
Click "📦 Export All as ZIP" to download all CVs as individual .txt files. -
Issue Certificate (for formal stress tests only, σ > 7)
Click "Issue Robustness Certificate".
7. Interpreting Audit Metrics
| Metric | Formula | Pass | Warning | Fail |
|---|---|---|---|---|
| SEO Resilience | (detected / injected) × 100% | ≥ 90% | 70–89% | < 70% |
| Data Integrity | (conflicts caught / injected) × 100% | ≥ 90% | 70–89% | < 70% |
| Parsing Accuracy | avg(1 − |parsed_YOE − truth_YOE| / truth_YOE) × 100% | ≥ 90% | 70–89% | < 70% |
8. Certification Criteria
| Criterion | Required Value | If Not Met |
|---|---|---|
| σ (Standard Deviation) | > 7 | Certificate rejected |
| Average of all 3 audit metrics | > 95% | Certificate rejected |
| Job Description | Uploaded and parsed | Certificate rejected |
9. Document Control
| Action | Owner | Frequency |
|---|---|---|
| Review SOP | AI QA Lead | Before each major release |
| Archive certificate | HR Records | Per simulation run |
| Update σ thresholds | Data Science Lead | When model retraining occurs |