PropClear Risk Assessment Methodology v1.0

How a PropClear screening produces its score

Every client screening runs through three components, each with a distinct role and a distinct level of trust.

Component 1 — Data gathering. Claude (the AI) performs adverse media web search, structured PEP and sanctions narrative analysis, and country risk assessment. It returns structured findings — a list of yes/no facts and short narratives. It does not calculate the risk score.

Component 2 — Deterministic scoring engine. A weighted scoring engine encoded in JavaScript reads Component 1's findings, the NameScan API result, the agent's wizard inputs, and any Form B flags. It applies the five-category weighted scoring set out below to produce a deterministic risk score from 0 to 100. The same inputs always produce the same outputs. This is the audit defence.

Component 3 — AI escalation reviewer. A separate, narrowly-scoped AI call reviews the deterministic score for factors the engine could not capture mechanically — combinations of behavioural flags suggesting structuring, contextual mismatches between profile and transaction, patterns resembling known typologies. It can add 0 to 25 points of escalation with written reasoning. It cannot reduce the deterministic score under any circumstances.

The five risk categories

Each category is sized based on its individual regulatory severity, not in proportion to a fixed total. The maximum theoretical sum of category caps is 125 points; the final score is capped at 100 to fit the rating bands. This preserves each category's standalone weight without artificial compression — genuine high-risk clients typically present multiple concurrent risk factors, and the methodology cannot under-rate them just to make the math add to 100.

Hard stops — mandatory transaction halts

Hard stops are regulatory bright lines. When triggered, the transaction must not proceed regardless of any other factors. The agent must not facilitate the transaction and must consider filing a Suspicious Transaction Report through SONAR.

• Exact sanctions match (NameScan or other authoritative source) — score saturates at 100, STR required, transaction halt.
• Named in active UN Security Council Resolution — score saturates at 100, STR required, transaction halt.
• Designated under TSOFA 2002 or UN Act 2001 without exemption order — transaction halt, STR required.

Score floors — deterministic minimums

Score floors elevate the deterministic risk score to a defined minimum when specific factors are present. Unlike hard stops, floors don't prevent the transaction — they just ensure the rating reflects severity that the category math alone might under-state.

• FATF blacklist nationality — minimum score 40 (Medium)
• Confirmed foreign PEP — minimum score 70 (High)
• Confirmed domestic PEP (Singapore) — minimum score 40 (Medium)
• Three or more Group C behavioural flags ticked on Form B — minimum score 50

How the final score is calculated

The deterministic engine combines category scores, hard stops, score floors, and AI escalation in a strictly defined sequence:

The 125-point maximum subtotal capped at 100 is intentional. Each category is sized on its individual regulatory severity, not forced to compete for share of a fixed total. The 100-cap step handles the rare case where multiple severe categories saturate at once. This is what allows confirmed foreign PEPs to be weighted at 50 points without artificially compressing the other categories.

Rating bands and required forms

For unrepresented counterparties, the equivalent forms in Annex G of the PMLPFTF Guide apply (U1, U2, U5, U6). For ongoing client relationships, Form D (Ongoing Due Diligence) applies in addition.

Audit trail — what's logged for every screening

For every screening, PropClear records and retains for the regulatory minimum of 5 years per Regulation 12 of the PMLPFTF Regulations 2021:

• All inputs to Component 1 (CDD wizard data, NameScan response)
• Component 1 structured findings (PEP determinations, country assessment, adverse media)
• Component 2 deterministic score breakdown (each category's points and the specific factors triggering them)
• Hard stops triggered, with rule identification
• Score floors applied, with rule identification
• Component 3 AI escalation amount and full written reasoning
• Final score and rating
• Methodology version applied (e.g. "v1.0")
• Timestamp and screening reference identifier

This audit trail is rendered in the screening result page, in Form B Section 3, and on the Form B PDF that the agent retains as compliance evidence. A CEA inspector reviewing any past screening can reproduce the exact reasoning that led to the rating.

Methodology maintenance

This methodology is reviewed annually, with the next scheduled review on 1 May 2027. It is also reviewed on an event-driven basis when:

• Material changes occur to the PMLPFTF Regulations 2021 or the CEA PMLPFTF Guide
• Material changes occur to FATF Recommendations 10 or 12
• FATF updates its blacklist or greylist (handled operationally on a quarterly cycle)
• MAS issues new guidance or notices materially affecting AML/CFT requirements for property transactions
• Methodology gaps are identified through CEA inspection feedback or operational experience