← Back to App
MAN-SIM-001 · v1.0 · March 2026

User Manual — AI Talent Pool Simulator

HCM-AO v2.0 · TICNES AI · Angola Digital

What is the Simulator?

The AI Talent Pool Simulator is a scientific validation tool built into HCM-AO that lets HR teams and AI engineers test the system's reliability before using it with real candidates. It generates realistic synthetic CVs, runs them through the same AI parsing pipeline that processes real CVs, and measures how accurately the AI performs.

✈️
AnalogyThink of it as a flight simulator for your recruitment engine — you fly with synthetic passengers first, under realistic and extreme conditions, before going live.

The Interface At A Glance

┌─────────────────────────────────────────────────┐
  🧪 AI Talent Pool Simulator & Stress-Tester     
├─────────────────────────────────────────────────┤
  [Bell Curve Visualisation]                      
                                                  
  μ ──────────────────── 10  ← mean YOE target   
  σ ──────────────────── 3   ← diversity level   
  ⚠️  [Stress Level Indicator]                    
                                                  
  [Core Skills Count: 5 ▾]    [Count: 20]        
                                                  
  📄 Target Job Description (Required)            
     [ Drop or click to upload ]                  
                                                  
  [Generate Mock CV Batch]                       
                                                  
  [📦 Export ZIP]  [🚀 Run in S&R Mode]          
└─────────────────────────────────────────────────┘

Step 1 — Upload Your Target Job Description

Before generating CVs, upload the job offer file (PDF or TXT). The Simulator uses this to extract required skills, calibrate skill matching per candidate, and provide job context to the Certificate.

Success indicatorWhen you upload, a green notification "✅ JD file ready: [filename]" appears and the upload zone turns green.

Step 2 — Set the Mean (μ) — "What kind of candidates?"

Move the μ slider to set the average years of experience you expect in this talent pool.

μ Candidate Profile
2–4 Fresh graduates / entry-level
5–8 Mid-career professionals
9–15 Senior / specialist hires
16–30 Executive / director-level

Example: For an Accounting Assistant role requiring 2–5 years, set μ = 3 or 4.

Step 3 — Set the Diversity (σ)

σ is the most powerful control. It has two simultaneous effects: talent pool diversity (spread of YOE values) and stress-test trap injection intensity.

⚠️
Certification requires σ > 7At σ ≤ 7, the system may inject some traps but the run does NOT qualify for formal certification. Increase σ above 7 to enter full Stress Test Mode.

σ Reference Table

σ Setting Mode Trap Prob. / CV Certificate?
0 Uniform Pool 0%
1–3 Low Entropy 7%–20%
4–6 Moderate Entropy 27%–40%
7 Near Threshold ~47%
≥ 8 🔴 Stress Test Mode 53%–80% ✅ Yes
> 12 ⚠️ Extreme / Pathological >80% ✅ Yes

The Stress Level Indicator bar below the σ slider updates in real-time showing exactly the current trap probability and mode label.

The Three Trap Types

Trap What it is Audit Metric
🕵️ Hidden Keywords Invisible white text keyword stuffing SEO Resilience
⚠️ Date Conflicts Employment dates reversed (start > end) Data Integrity
🔡 Encoding Noise Invisible Unicode zero-width characters in names Parsing Accuracy

Step 4 — Batch Size and Core Skills

  • Count (default 20) — how many CVs to generate. Use ≥ 50 for a statistically meaningful stress test.
  • Core Skills Count — how many JD-specific skills each candidate carries. Use 5 for most roles; 8–10 for highly technical roles.

Step 5 — Generate the Batch

Click "Generate Mock CV Batch". The system will:

  1. Sample YOE values from the Gaussian distribution N(μ, σ²)

  2. Build a full CV per candidate — name, experience, education, skills, projects

  3. Inject adversarial traps probabilistically based on σ

  4. Run the NLP parsing pipeline on every CV

  5. Compare parsed results to known ground truths

  6. Compute the three Real-Time Audit Metrics

Step 6 — Review Each CV

After generation, the Generated CVs Batch panel appears. Click any candidate row to open their full profile: name, email, phone, skills (showing JD skill matches), education, and experience.

Step 7 — Run in Selection & Recruiting Mode

Click "🚀 Run in Selection & Recruiting Mode" to feed the entire batch directly into the AI ranking engine — just as if they were real candidates applying for the job. This provides a full end-to-end impact view of what the algorithm would do with this talent pool distribution.

🔬
What the bridge doesThe backend converts generated CVs to full parsed format, runs rank_candidates(), and populates the S&R Results Dashboard — including Assessment Reports and AI-drafted Response Emails for each candidate.

Step 8 — Export as ZIP

Click "📦 Export All as ZIP" to download a .zip file with all generated CVs as individual .txt files. This works entirely in your browser — no server round-trip needed.

Step 9 — Audit Metrics Explained

🕵️
SEO Resilience
≥ 90% = PASS
"Did the AI detect hidden keyword tricks?" 0% with no traps injected = NOT a failure
📅
Data Integrity
≥ 90% = PASS
"Did the AI catch impossible career timeline conflicts?"
🎯
Parsing Accuracy
≥ 90% = PASS
"How accurately did the AI read years of experience vs. ground truth?"
ℹ️
Why SEO and Integrity show 0% at low σAt σ ≤ 3, the trap injection probability is only 7%–20% per CV. In a batch of 5–10 CVs, it is entirely normal to have zero traps injected. These metrics are only meaningful at σ ≥ 5.

Step 10 — Certificate of Algorithmic Robustness

Click "Issue Robustness Certificate" after a simulation to generate a formal certification document.

Condition Required Value
σ (stress level) > 7
Average of all 3 audit metrics > 95%

If either condition is not met, the certificate displays "⚠ NOT CERTIFIED" with the specific unmet criteria listed. Both conditions must be true simultaneously.

Frequently Asked Questions

Why is SEO Resilience showing 0%?
At low σ (e.g. σ=3), the probability of injecting a SEO trap per CV is only ~20%. In a batch of 5 CVs, it is likely that zero traps were planted. Increase σ to ≥ 8 to guarantee meaningful trap injection. "0%" with "0 traps injected" is not a failure — it is an expected result at low entropy.
Why does the system say "NOT CERTIFIED" even though accuracy is 86%?
Two conditions must both be met: σ > 7 AND average accuracy > 95%. Your σ was too low (e.g. 3), which does not qualify as a formal stress test regardless of the accuracy score.
Can I use the same JD as my real recruitment process?
Yes — this is the recommended practice. Upload your real JD, generate a batch, check the audit metrics, and only proceed to actual candidate processing once the system performs well.
How many CVs do I need for a reliable stress test?
Minimum 50 CVs at σ ≥ 8. With fewer CVs, the probabilistic trap injection may not generate enough traps to draw reliable statistical conclusions.
Why does the Export ZIP button download immediately but "Run in S&R Mode" needs a server call?
Export is 100% client-side — it uses browser memory (JSZip). The S&R bridge calls the server to run the full AI ranking pipeline and generate detailed Assessment Reports, which cannot be done in the browser.

Quick Reference Card

# WARM-UP — casual validation
μ     = role target YOE
σ     = 2–4
count = 10–20
→ Confirms parser works on clean data

# STANDARD TEST — pre-operational check
μ     = role target YOE
σ     = 5–7
count = 20–50
→ Moderate traps. Gets meaningful Parsing Accuracy scores.

# FULL STRESS TEST — certification eligible
μ     = role target YOE
σ     = 8–12
count = ≥ 50
→ All trap types active. Certificate issued if avg accuracy > 95%.