It's Not Your Engine. It's Your List.

Clean, context-rich PEP control lists are the only scalable way to cut false positives.

---

Real talk: your name isn't unique

Unless you're literally named "X Æ A-Xii" (Elon Musk's child's name), your name isn't unique. Screening engines see millions of name look-alikes; that's why you need other identifiers—DOB, jurisdiction, role dates, addresses, and well-structured aliases—to triangulate who you actually are and cut false positives.

TL;DR

Upgrade to clean, context-rich, time-bounded PEP data and watch false positives drop—without torturing thresholds.

---

Engines are smart—lists must be smarter

Modern screening stacks blend fuzzy matching, phonetics, transliteration, nickname mapping, and name-order logic to estimate match probability. What they can't do is invent context your list doesn't supply.

• No DOB? Larger candidate sets and more noise.
• No jurisdictions? More cross-border overlaps.
• No role dates? Yesterday's official pollutes today's matches.

Control list quality > endless retuning.

False Positive Rate vs PEP Control List Richness

Richer PEP records (DOB, role dates, jurisdictions, aliases) shrink the candidate set before scoring.

---

Reality check: when lists are garbage, FPR can hit 100%

We've seen programs where the upstream PEP source was so poor that false positives hit 100%—every single alert was junk. That's not an engine problem; that's a list-quality problem.

Key point: Name-only, undated roles, missing jurisdictions, and absent DOBs create candidate sets so broad that precision collapses. No amount of threshold tweaking fixes missing context.

---

Regulatory reality: you can't turn off PEP screening—so do this

1. Upgrade the control list: full DOBs, role start/end dates, role jurisdictions, structured aliases, RCAs, and source lineage.
2. Time-bound roles: explicitly retire signals when a term ends to avoid "forever PEP" matches.
3. Boost precise matches: give higher weight to role-dated + DOB-confirmed candidates; de-emphasize name-only fuzz.
4. Delta updates: ingest new appointments quickly and capture exits—yesterday's data shouldn't pollute today.
5. Analyst feedback loop: capture dispositions (TP/FP) to refine aliasing and edge-case heuristics in the list.

Bottom line: compliance stays on; noise goes down; analysts focus on real risk.

---

"100 Men vs 1 Gorilla" simulation (100,000 clients with full CIP)

We compared Pergamon Enriched Lists (DOB, role dates, jurisdictions, aliases, sources) to a Legacy name-heavy list (DOB 50%, address 50%, sources 15%). Assumptions: true PEP prevalence 0.1%, non-PEP name-collision baseline 4%, and stricter alerting when context is present.

Data note: Name-collision baselines calibrated using Statistics Canada's 2021 Census first-name frequencies. Higher concentration → higher collision risk when DOB & role dates are missing.

Visual Comparison

The difference is stark when visualized:

---

Math & Validation Appendix

When you have to show the proof and the numbers actually add up.

Simulation math

TPₛ = N · π · rₛ
FPₛ = N · (1 - π) · (p_name-coll · p(alert|collision)ₛ)
Alertsₛ = TPₛ + FPₛ
Precisionₛ = TPₛ / Alertsₛ
FP%ₛ = FPₛ / Alertsₛ
Alerts per 1k = Alertsₛ / (N / 1000)
FP per 1k = FPₛ / (N / 1000)

Sensitivity (name-collision = 2%, 4%, 6%)

---

The Bottom Line

"It's not your screening model. It's your control list." — change my mind.

---

© 2025 Pergamon Labs — Context-first PEP control lists.