Methodology

01 — Framework

Scope and authorship

Adrien Pesa developed this free research diagnostic within Governed Autonomy, his independent, self-funded and non-commercial research programme. It supports structured reflection on AI governance practices in private banks in Singapore.

The register is an authored interpretation of the selected corpus, not a complete inventory of Singapore law or a regulator-approved assessment. No criterion in the current register is classified as statutory.

The short diagnostic covers 35 selected priority criteria and takes approximately 25–30 minutes; the extended diagnostic covers all 193 and takes approximately 90–120 minutes. Both are free and use the same scoring rules. A short diagnostic leaves most of the corpus unassessed, so its results describe that limited scope.

Results are calculated in your browser. If you request a PDF, your answers and optional institution label are sent to the report server. Reports use authored analysis by default. An optional AI summary sends selected findings to Anthropic only if you choose it. Anthropic normally retains API inputs and outputs for 30 days, with legal and policy enforcement exceptions. The privacy policy explains processing and retention.

02 — Authority

Source classes and corpus

The programme distinguishes five normative tiers—statutory, supervisory, consultation, methodology and assurance—and a separate observed-practice class. These 6 source classes describe different kinds of authority and evidence; they are not six levels of legal obligation. Expand a class to see the selected instruments, editions and criterion counts. Source class matters for interpretation and the specific maturity override below; it does not weight the readiness percentage.

Counts sum to 193 across the source classes. Each criterion is counted once under its primary source; some also cite supporting instruments. Edition labels describe the selected research corpus. Publisher links may open an initiative page or directory. Check the publishers for later revisions; the corpus is not an exhaustive or continuously updated regulatory inventory.

03 — Coverage

The seven domains

Every criterion is assigned to one of 7 thematic domains. Results report the fully implemented percentage and findings within each domain. The overall maturity label uses all applicable answered criteria; domain percentages are not averaged to produce it.

Domain 1 carries 60% of the register, reflecting the depth of P017 and the MAS AI MRM Information Paper around the model lifecycle. The remaining 40% is distributed across the other 6 domains. Because each applicable answered criterion has equal weight, domains with more criteria contribute more to the overall percentage. This distribution reflects the selected corpus, not an independently established ranking of institutional risks.

04 — Logic

Response model & scoring

Select the response that best matches the evidence available to you. The diagnostic does not inspect that evidence or independently verify your answers. Not applicable is available only where the criterion offers option D.

If none of the available answers accurately describes your institution, leave the criterion unanswered. You can review partial results in the browser; a PDF requires an answer to every criterion in the selected scope.

Implementation percentage and severity

The implementation percentage is A ÷ (A + B + C) × 100, rounded to one decimal place for display. A counts in the numerator and denominator; B and C count in the denominator, with no partial credit for B. D and unanswered criteria are excluded. Domain percentages use the same calculation. Every applicable answered criterion has equal weight; neither source class nor severity weights the percentage.

HIGH, MEDIUM and LOW are authored severity classifications for the findings. Severity determines their presentation order and the HIGH-gap overrides described below. Only C answers count as gaps for those overrides; partial answers remain separate findings. These are research conventions, not regulator-issued severity scores.

05 — Output

Maturity rating

Zero applicable answers produce “Insufficient Data”. Otherwise, any STATUTORY HIGH-severity C answer, or three or more HIGH-severity C answers across any source class, produces “Critical Gaps”. If neither override applies, the unrounded overall percentage determines the band below. These labels and thresholds are research conventions; they are not regulatory grades or validated benchmarks of institutional performance.

The statutory override remains part of the scoring framework, although the current register contains no statutory criteria. Fewer than half of the register’s criteria answered A, B or C adds a coverage warning; it does not otherwise change the label. The short diagnostic will therefore normally carry that warning. Rule-based consistency checks flag specified answer combinations for review without changing answers. Scoring, labels and gap identification are deterministic; AI does not calculate them.