Centre Status
Model Risk, AI Governance & Decision Intelligence Centre
Governs AI/model/tool-assisted decisions inside BLACKSWAN Capital Markets OS — the regulatory digital twin engine, evidence-pack scoring, narrative generation, exam-response drafting, anomaly triage, revenue classifier, vendor scoring, and incident triage. Every output traverses a human-in-the-loop gate, validation track, drift watch, and override register. Conservative posture: internal model-risk readiness workflow only — outputs are decision support for a human reviewer, never automated legal, regulatory, investment, or audit decisions.
Models / tools inventoried
10
4 Decision Support · 3 Assistive · 2 Human-Review-Required · 1 Restricted
Validation track
4 / 10
Approved Internal · 5 Validation Pending · 1 Conditional
Human-in-loop coverage
10 / 10
Every output requires named reviewer · override register active
Open exceptions
9
Tracked in exception register · counsel + risk committee review pending
Decision / model state legend
Not Assessed
Intake
Low-Risk Assistive
Controlled Decision Support
Human Review Required
Validation Pending
Approved Internal
Conditional Use
Restricted Use
Drift Watch
Override Required
Suspended
Retired
Archived
Twelve governance domains
Model Risk & AI Governance domains
Model / Tool Inventory
10 catalogued
Every model, prompt-template, tool integration, and decision engine catalogued with owner, version, input data, output type, and human-review requirement.
Use-Case Classification
Counsel review pending
Each use case mapped to ADGM/FSRA, UK FCA, MAS, MiFID II readiness posture and to internal limitation text. Drafts pending counsel sign-off.
Decision Criticality
4 high · 4 medium · 2 low
Criticality tier determines validation depth, reviewer seniority, evidence retention, and audit cadence.
Human-in-the-Loop Review
10 / 10 mandatory
No model output reaches a customer-facing or board-facing artefact without a named reviewer signing the evidence pack.
Validation & Testing
5 of 10 pending
Test suites covering expected behaviour, failure modes, threshold, and remediation. Validation owner ≠ build owner.
Explainability
Rationale + sources required
Every output carries an input snapshot, version, tool-chain ref, rationale summary, confidence signal, and reviewer note.
Data Lineage
Linked to Data Governance
Input data, classification, jurisdiction, retention class, and downstream artefacts tracked from source through every prompt/tool hop.
Prompt / Tool Audit
All prompt/tool calls logged
User/agent, prompt category, connected tool, data category, output class, restriction, and retained evidence hash captured per call.
Bias / Fairness Review
Scope: B2B counterparty & staff
Surface scoped: no consumer-credit, employment-screening, or protected-class decisioning. Fairness review focuses on counterparty-treatment equity and reviewer-bias controls.
Drift & Performance Monitoring
2 drift watches active
False-pos/neg watch, stale references, rule drift, jurisdiction drift, prompt drift, output quality, exception rate, escalation triggers.
Override & Challenge
Workflow live
Reviewers can challenge any generated decision; overrides require rationale, approver, audit event, and downstream pack regeneration.
Model Risk Evidence Pack
Pack assembled
Inventory · Validation · Override register · Drift log · Exception register · Audit trail — bundled for board, exam, and auditor view.
Model & tool inventory
Inventoried models, tools & decision engines
| Model / Tool | Owner | Use case | Decision criticality | Input data | Output type | Human review | Validation state | Jurisdiction sensitivity | Blocker | Evidence ref | State |
|---|---|---|---|---|---|---|---|---|---|---|---|
| Regulatory Digital Twin Decision Engine | Head of Regulatory · CISO | Map workflow → ADGM / FCA / MAS / MiFID expectations & surface internal readiness limitations | High | Workflow telemetry · regulatory rule pack · evidence pack metadata | Limitation text · readiness state · jurisdiction flag | Mandatory · Head of Regulatory | Validation Pending | ADGM · FCA · MAS · MiFID II | Counsel sign-off on jurisdiction rule pack pending | MR-EV-DTWN-001 | Human Review Required |
| Evidence Dependency Graph Scoring | Head of Evidence · Data Governance Lead | Compute upstream/downstream evidence dependencies and gating impact | High | Evidence pack metadata · audit log · pack readiness state | Dependency map · gating impact score | Mandatory · Head of Evidence | Approved Internal | Internal · cross-jurisdiction read-only | None | MR-EV-DEPG-002 | Controlled Decision Support |
| Pack Readiness Scoring | Head of Evidence · Programme Manager | Score 21 evidence packs across completeness, freshness, cross-link integrity, owner-attestation | High | Pack metadata · last reviewer event · cross-link graph | Readiness score · open-gap list | Mandatory · Programme Manager | Approved Internal | Internal | None | MR-EV-PRSC-003 | Controlled Decision Support |
| Board Narrative Generator | CEO Chief of Staff · Head of Comms | Draft board-pack narrative aligned to readiness state & open exceptions | High | Pack metadata · decision register · open exceptions · approved KPIs | Draft narrative (markdown) | Mandatory · CEO Chief of Staff | Validation Pending | Internal · board-only | Tone & limitation footer template pending Legal | MR-EV-BNAR-004 | Human Review Required |
| Executive Morning Brief Generator | CEO Chief of Staff | Surface overnight deltas: incidents · regulator events · pack-state changes · vendor risk · revenue exceptions | Medium | Audit log slice (last 24h) · pack delta · monitoring signals | Markdown brief · top-3 attention list | Mandatory · CEO Chief of Staff | Conditional Use | Internal | Brief-template review by Legal pending | MR-EV-MBRF-005 | Conditional Use |
| Regulatory Exam Response Drafting | Head of Regulatory · External Counsel | Draft response binder text for regulator information requests | High | Regulator request · prior binders · evidence packs · audit log | Draft binder section (markdown) · evidence cross-links | Mandatory · External Counsel countersign | Validation Pending | ADGM · FCA · MAS | External counsel review of generation template pending | MR-EV-EXAM-006 | Restricted Use |
| Stakeholder Room Access Anomaly Triage | CISO · Head of Stakeholder Rooms | Triage anomalous access patterns in stakeholder rooms and counterparty data rooms | Medium | Room access log · device signals · recipient list | Severity score · suspected pattern · suggested action | Mandatory · CISO | Approved Internal | UK · EEA · ADGM (personal data scope) | None | MR-EV-ROOM-007 | Controlled Decision Support |
| Revenue Recognition Evidence Classifier | CFO · External Auditor (pending) | Classify revenue-event evidence for recognition treatment and audit binder | High | Contract metadata · billing event · invoice state · audit-log evidence | Treatment hint · audit-flag · open-question list | Mandatory · CFO sign-off | Validation Pending | Internal · audit-track | External auditor validation pending | MR-EV-REVC-008 | Validation Pending |
| Vendor Concentration Risk Scorer | Head of Procurement · Risk Committee | Score vendor concentration and dual-supplier readiness | Medium | Vendor inventory · contract metadata · SLA telemetry · exit plan state | Concentration score · dual-source gap list | Mandatory · Risk Committee | Approved Internal | UK FCA · EU DORA scope | None | MR-EV-VCON-009 | Controlled Decision Support |
| Incident Severity Triage | SRE Lead · CISO | Suggest incident severity tier and notification timeline | Low | Monitoring signals · paging history · service-impact map | Severity suggestion · suggested comms list | Mandatory · SRE Lead | Approved Internal | Internal | None | MR-EV-INCT-010 | Low-Risk Assistive |
Use-case classification
Internal use-case readiness — ADGM · UK FCA · MAS · MiFID II
Conservative posture: classifications below are internal model-risk readiness drafts pending counsel + risk-committee sign-off. They are not legal advice, not regulatory approval, and do not certify compliance with any regulator's AI / outsourcing / model-risk regime.
| Jurisdiction | Use-case category | Internal readiness state | Limitation text (draft) | Owner | Human review mandatory? |
|---|---|---|---|---|---|
| ADGM / FSRA | Regulator-facing readiness, exam response, evidence packs | Drafted · counsel pending | Outputs are internal decision support; not automated regulatory advice. Submissions to FSRA always reviewed and signed by external counsel. | Head of Regulatory | Yes — external counsel countersign |
| UK FCA | Outsourcing / operational resilience / SMCR-aligned readiness | Drafted · counsel pending | Outputs are internal model-risk readiness only; not automated SMCR attestation, not a substitute for SM&CR-accountable individual sign-off. | Head of Regulatory · SMF Lead | Yes — SMF-accountable sign-off |
| Singapore / MAS | FEAT-aligned readiness (Fairness · Ethics · Accountability · Transparency) | Drafted · counsel pending | Drafted against MAS FEAT principles for internal readiness; not certification. Customer-facing decisioning out of scope. | Head of Regulatory · Data Governance Lead | Yes — Data Governance Lead |
| EU MiFID II / MiFIR | Algorithmic-readiness posture (no algorithmic trading); record-keeping | Drafted · counsel pending | Platform does not place orders or execute trades. Outputs are operational readiness support only; record-keeping standards aligned for evidence retention. | Head of Regulatory · CFO | Yes — CFO / Head of Regulatory |
Validation & testing
Validation panel · test set · expected behaviour · failure modes
| Model / tool | Test set | Expected behaviour | Failure mode watched | Validation owner | Last test | Threshold | Result | Remediation action |
|---|---|---|---|---|---|---|---|---|
| Regulatory Digital Twin | 72 scenario cards × 4 jurisdictions (curated) | Limitation text matches counsel-approved phrasing for each jurisdiction | Jurisdiction confusion · stale rule reference · over-confident phrasing | Head of Regulatory (val. owner ≠ build) | 2026-05-12 | ≥ 95 % phrasing match · 0 over-confident outputs | 92 % match · validation pending | Re-test after counsel finalises rule pack |
| Evidence Dependency Graph | 21 evidence packs · 412 cross-links | All upstream/downstream dependencies resolved with no orphans | Orphan node · phantom edge · stale link | Head of Evidence | 2026-05-15 | 0 orphans · 0 phantom edges | Pass | Re-run weekly |
| Pack Readiness Scoring | 21 packs × completeness · freshness · cross-link · attestation | Score reproducible given identical metadata input | Score drift · attestation freshness drift | Programme Manager | 2026-05-14 | Score variance ≤ 1 pt on identical input | Pass · 0.4 pt variance | None |
| Board Narrative Generator | 9 historical board packs (curated) | Narrative reflects only items in approved KPI / decision register · limitation footer present | Hallucinated decision · missing limitation · over-confident framing | CEO Chief of Staff | 2026-05-13 | 0 hallucinated decisions · footer present 100 % | 0 hallucinations · footer 96 % | Fix template to always emit footer; re-test |
| Morning Brief Generator | 30 days simulated audit-log slices | Top-3 attention items match human-curated baseline ≥ 80 % | Missed incident · noise inflation | CEO Chief of Staff | 2026-05-15 | ≥ 80 % match · 0 missed Sev-1 | 83 % · 0 missed Sev-1 | Tune noise filter; re-test weekly |
| Regulatory Exam Response Drafting | 5 historic regulator-style requests (synthetic) | Draft cites only approved evidence; counsel countersign present | Uncited claim · disallowed source · over-broad assertion | External Counsel | 2026-05-10 | 0 uncited claims · 100 % counsel countersign | Counsel review of template pending | External counsel review of template; restricted use until pass |
| Stakeholder Room Anomaly Triage | 30 days simulated room access log | Flag anomalous patterns ≥ 90 % recall on injected anomalies | False negative · false positive over 5 % | CISO | 2026-05-14 | Recall ≥ 90 % · FPR ≤ 5 % | Recall 94 % · FPR 3.1 % | None |
| Revenue Recognition Classifier | 200 simulated billing events × treatment labels | Treatment hint matches CFO-approved baseline ≥ 95 % | Mis-classification · auditor-disallowed treatment | CFO · External Auditor (pending) | 2026-05-09 | ≥ 95 % match · 0 auditor-disallowed | 96 % match · external auditor pending | Hold for external auditor validation |
| Vendor Concentration Scorer | 10 vendor inventory · 7 dependency rows | Concentration score reproducible · dual-source gap list complete | Stale vendor data · missed sub-processor | Head of Procurement | 2026-05-15 | Score variance ≤ 1 pt · 0 missed sub-processors | Pass | Re-run on each vendor inventory change |
| Incident Severity Triage | 40 historic incidents (simulated) | Severity suggestion ≥ 85 % agreement with SRE Lead baseline | Over-escalation · under-escalation | SRE Lead | 2026-05-15 | ≥ 85 % agreement · 0 under-escalations on Sev-1 | 89 % · 0 Sev-1 missed | None |
Human override & challenge
Override & challenge workflow — generated decision · reviewer challenge · audit
| Generated decision / output | Reviewer challenge | Override rationale | Approver | Audit event | Downstream pack update | Limitation text recorded |
|---|---|---|---|---|---|---|
| Digital Twin: workflow X mapped to MAS Notice 644 only | Reviewer: MAS Outsourcing Guidelines also engaged | Add MAS Outsourcing Guidelines limitation paragraph; counsel concurs | Head of Regulatory · External Counsel | MR-OVR-001 · 2026-05-12 09:42 UTC | Regulatory Notification & Board Escalation pack v2 regenerated | "Internal readiness only; not MAS compliance certification." |
| Board Narrative: confident framing on launch readiness | Reviewer: tone too positive given 9 open exceptions | Re-draft with conditional framing & explicit exception count | CEO Chief of Staff | MR-OVR-002 · 2026-05-13 17:08 UTC | Board Narrative v3 issued · prior version archived | "Internal narrative draft; pending board sign-off." |
| Exam Response: cites evidence pack EP-12 fully | External counsel: EP-12 contains MNPI; restricted citation | Replace citation with redacted summary; restrict pack access | External Counsel · CISO | MR-OVR-003 · 2026-05-11 14:21 UTC | Exam Response binder v2 · Data Governance access ledger updated | "Restricted use; counsel-redacted citation required." |
| Revenue Classifier: subscription renewal recognised over 12 m | CFO: contract contains material modification clause; reassess | Hold treatment hint; flag for auditor; do not auto-recognise | CFO | MR-OVR-004 · 2026-05-09 11:55 UTC | Revenue Billing & Financial Controls pack v3 · audit-flag set | "Validation pending; treatment subject to auditor review." |
| Incident Triage: Sev-3 suggested for monitoring lag | SRE Lead: lag exceeded SLO > 15 min · escalate | Promote to Sev-2; trigger regulator-notification timeline review | SRE Lead · CISO | MR-OVR-005 · 2026-05-14 03:18 UTC | Production Monitoring · Regulatory Notification & Board Escalation packs | "Assistive only; severity remains a human decision." |
Explainability & evidence
Per-output evidence — input snapshot · version · rationale · confidence · reviewer note
| Output ID | Input snapshot | Model / tool version | Prompt / tool chain reference | Rationale summary | Confidence / quality signal | Reviewer note | Evidence hash |
|---|---|---|---|---|---|---|---|
| DTWN-117 | workflow-readiness-2026-05-12T0931Z.json | dtwn-engine v0.7.3 · rule-pack 2026.05.A | prompt://dtwn/jurisdiction-classifier#v7 → tool://regulator-rule-lookup | Workflow X engages ADGM, FCA, MAS; FCA primary on outsourcing posture | 0.86 internal | Counsel: confirm MAS engagement before sharing externally | sha256:9f1b…c4d2 |
| DEPG-204 | evidence-graph-2026-05-15T0810Z.json | depg-scorer v1.2.0 | tool://evidence-graph-walker | Pack EP-07 gates EP-12, EP-18; orphan check clean | 0.97 internal | Head of Evidence: approved | sha256:1c44…8e09 |
| BNAR-058 | board-pack-2026-05-13T1659Z.json | narrative-gen v0.4.1 | prompt://board/narrative#v9 → tool://exception-register-read | Narrative covers 9 open exceptions; conditional framing applied | 0.79 internal | CoS: footer must list all 9 exceptions; v3 issued | sha256:6a7d…2f10 |
| EXAM-031 | exam-request-2026-05-10T1402Z.json | exam-draft v0.3.0 | prompt://exam/binder#v5 → tool://evidence-pack-read | Binder section drafted; 3 evidence packs cited; redaction applied per CISO | 0.71 internal | External counsel: countersign before issue | sha256:b39c…77a1 |
| REVC-412 | billing-event-2026-05-09T0840Z.json | rev-classifier v0.6.0 | tool://contract-metadata-read → prompt://rev/treatment-hint#v4 | Subscription renewal; 12-month recognition hint; material-modification clause flagged | 0.65 internal | CFO: hold for auditor | sha256:0e21…aa3b |
| ROOM-099 | room-access-2026-05-14T0218Z.json | room-anomaly v1.1.0 | tool://room-access-log-read | Anomalous off-hours read of evidence pack EP-12; recipient on approved list | 0.82 internal | CISO: monitored; no action | sha256:2d18…5c66 |
| INCT-540 | incident-2026-05-14T0314Z.json | severity-triage v0.5.2 | tool://monitoring-signals → prompt://incident/severity#v3 | Monitoring lag > SLO 15m; severity-tier promoted by reviewer | 0.74 internal | SRE Lead: human decision overrode model | sha256:f4b8…91dc |
Drift & performance monitoring
Drift watch · stale data · rule drift · jurisdiction drift · prompt drift · exception rate
| Signal | What it watches | Owner | Current | Threshold | Escalation trigger | State |
|---|---|---|---|---|---|---|
| False positive / negative watch | Triage models (room anomaly · incident severity) | CISO · SRE Lead | FPR 3.1 % · FNR 1.2 % | FPR ≤ 5 % · FNR ≤ 2 % | Validation owner page if breached for 3 consecutive runs | Healthy |
| Stale data dependency | Reference data older than retention/refresh policy | Data Governance Lead | 2 references flagged stale | 0 stale references | Block dependent model output if stale > 14 days | Watch |
| Regulatory rule drift | Regulator rule pack vs. counsel-approved baseline | Head of Regulatory | 1 rule pending counsel review | 0 pending | Restrict digital twin output for affected jurisdiction | Watch |
| Jurisdiction drift | Workflow → jurisdiction mapping changes vs. last counsel sign-off | Head of Regulatory | 1 mapping changed (DTWN-117) | 0 unsigned changes | Counsel review event triggered | Counsel pending |
| Prompt drift | Prompt template hash vs. approved baseline | Model owner per tool | 0 unsigned prompt changes | 0 | Lock model output until owner re-attests | Healthy |
| Output quality | Reviewer-rated quality on sampled outputs | Validation owners | 4.3 / 5 (rolling 7 d) | ≥ 4.0 | Quality dip > 0.3 pt triggers retest | Healthy |
| Exception rate | Overrides + challenges per 100 outputs | Risk Committee | 6.4 / 100 | ≤ 10 / 100 | Sustained rate > 10 triggers model-suspend review | Healthy |
| Escalation trigger | Aggregate drift signals → executive paging | CEO Chief of Staff | 0 paged | 1 amber for 7+ d → page | Two amber signals open > 7 d | Two amber open |
Prompt / tool audit trail
User · prompt · connected tool · data category · output class · restriction · evidence hash
| Timestamp (UTC) | User / agent | Prompt category | Connected tool used | Data category | Output class | Restriction applied | Retained evidence hash | Review status |
|---|---|---|---|---|---|---|---|---|
| 2026-05-15 09:31 | CEO Chief of Staff | Morning brief | audit-log-slice · pack-delta | Internal (no MNPI) | Markdown brief | Internal only · footer required | sha256:71ae…1d22 | Reviewed |
| 2026-05-15 08:10 | Head of Evidence | Dependency graph score | evidence-graph-walker | Internal | Score + gap list | None | sha256:1c44…8e09 | Reviewed |
| 2026-05-14 02:18 | CISO | Room anomaly triage | room-access-log-read | Personal (UK · EEA · ADGM) | Severity + suggested action | DSR-readiness link · access ledger update | sha256:2d18…5c66 | Reviewed |
| 2026-05-13 16:59 | CEO Chief of Staff | Board narrative | exception-register-read | Internal · board-only | Markdown narrative | Conditional framing · limitation footer | sha256:6a7d…2f10 | Override applied |
| 2026-05-12 09:42 | Head of Regulatory | Jurisdiction map | regulator-rule-lookup | Internal · MNPI scope tagged | Limitation text | Counsel-pending lock | sha256:9f1b…c4d2 | Counsel pending |
| 2026-05-11 14:21 | Head of Regulatory | Exam binder draft | evidence-pack-read | MNPI · restricted | Markdown binder section | External counsel countersign required · citation redaction | sha256:b39c…77a1 | Restricted |
| 2026-05-09 08:40 | CFO | Revenue treatment hint | contract-metadata-read | Internal · audit-track | Treatment hint + flag | Validation pending · auditor hold | sha256:0e21…aa3b | Validation pending |
| 2026-05-14 03:14 | SRE Lead | Incident severity triage | monitoring-signals | Internal | Severity suggestion | None | sha256:f4b8…91dc | Reviewed · overridden |
| 2026-05-15 08:55 | Head of Procurement | Vendor concentration | vendor-inventory-read | Internal | Score + dual-source gap list | None | sha256:8cc1…43e7 | Reviewed |
Decision intelligence board
Model / tool outputs tied to evidence packs & production centres
Regulatory Digital Twin Decision Evidence
Twin outputs (DTWN-*) feed Regulatory Notification & Board Escalation, Data Governance jurisdiction tags, and the exam-response binder. Every output carries the limitation footer and counsel-pending flag where applicable.
Counsel pending
Model Risk Evidence Pack
Inventory · validation · override register · drift log · exceptions · audit trail. Bundle is the artefact a board or regulator inspects when asking "how is AI/model use governed?"
Pack assembled
Control Testing Evidence
Test sets + results for every model feed the control-testing evidence binder consumed by Approval & Sign-Off and Release Control.
5 pending
Regulatory Exam Response Evidence
Exam-binder drafts carry per-output evidence hashes and counsel countersign records back to the Regulatory Notification & Board Escalation Centre.
Restricted Use
Board Narrative
Narrative drafts (BNAR-*) consumed by board pack; every issue carries the conditional-framing template and limitation footer.
Human Review
Evidence Dependency Graph
Graph scoring (DEPG-*) gates pack-readiness scoring and feeds Launch Readiness, Production Go/No-Go, and Completeness Command Centres.
Approved Internal
Pack Readiness Scoring
Scores 21 evidence packs across completeness · freshness · cross-link integrity · attestation. Tied directly to Completeness Command Centre dashboards.
Approved Internal
Model / AI exception register
Open exceptions — counsel + risk-committee pending
| Exception | Affected model / tool | Severity | Owner | Mitigation in place | Resolution path | Target close | State |
|---|---|---|---|---|---|---|---|
| Unvalidated model in active use | Regulatory Digital Twin | High | Head of Regulatory | Human-review mandatory; output not externalised | Counsel sign-off on rule pack; re-run validation suite | 2026-06-05 | Validation Pending |
| Missing human-review attestation | Morning Brief Generator | Medium | CEO Chief of Staff | Brief flagged "draft" until attestation logged | Template emits attestation block; weekly audit | 2026-05-24 | In Progress |
| Unexplained output (rationale absent) | Board Narrative Generator | Medium | CEO Chief of Staff | Output blocked from issue if rationale field empty | Template gate; re-test against curated set | 2026-05-22 | In Progress |
| Stale training / reference data | Vendor Concentration Scorer | Low | Head of Procurement | Score blocked > 14 d stale | Weekly vendor-inventory snapshot | 2026-05-30 | In Progress |
| Restricted data exposure risk | Regulatory Exam Response Drafting | High | External Counsel · CISO | Restricted Use; redaction tool enforced; counsel countersign required | Counsel-approved template; CISO sign-off | 2026-06-10 | Restricted Use |
| Prompt / tool audit gap | Cross-tool (legacy traces) | Medium | Data Governance Lead | New calls captured; legacy gap documented | Backfill audit log from artefact metadata | 2026-05-28 | In Progress |
| Drift threshold breached | Digital Twin · Rule Pack drift | Medium | Head of Regulatory | Output restricted for affected jurisdiction | Re-validate after counsel rule-pack finalisation | 2026-06-05 | Drift Watch |
| Unsupported jurisdiction request | Regulatory Exam Response Drafting (US/SEC scope) | Medium | Head of Regulatory | Output refused; manual counsel-led drafting | Define scope ban list; surface in UI | 2026-05-26 | Restricted Use |
| Over-claim risk in generated text | Board Narrative · Morning Brief | Medium | CEO Chief of Staff · Legal | Conditional-framing template; limitation footer enforced | Legal-approved phrase library; reviewer sign-off | 2026-05-25 | In Progress |
Audit trail & evidence preservation
Audit log of model-risk governance events
| Timestamp (UTC) | Actor | Model / tool | Decision / output | Evidence hash | Limitation recorded | Next step |
|---|---|---|---|---|---|---|
| 2026-05-08 08:02 | Head of Evidence | n/a (process) | MR-INVENTORY — model/tool inventory published | sha256:aa11…b202 | "Internal readiness only." | Counsel + risk committee review |
| 2026-05-09 11:40 | Head of Regulatory | Digital Twin | MR-CLASSIFICATION — use-case classification drafted (ADGM/FCA/MAS/MiFID) | sha256:bb22…c303 | "Not legal advice; counsel pending." | External counsel sign-off |
| 2026-05-15 08:10 | Head of Evidence | Evidence Dep. Graph | MR-VAL-DEPG-002 — validation pass | sha256:1c44…8e09 | "Internal." | Continue weekly re-run |
| 2026-05-14 09:05 | Programme Manager | Pack Readiness Scoring | MR-VAL-PRSC-003 — validation pass · 0.4 pt variance | sha256:cc33…d404 | "Internal." | None |
| 2026-05-13 17:08 | CEO Chief of Staff | Board Narrative | MR-OVR-002 — override · re-draft with conditional framing | sha256:6a7d…2f10 | "Internal narrative draft." | v3 issued · v2 archived |
| 2026-05-12 09:42 | Head of Regulatory | Digital Twin | MR-OVR-001 — add MAS Outsourcing Guidelines limitation paragraph | sha256:9f1b…c4d2 | "Not MAS compliance certification." | Reg. Notification pack v2 regen |
| 2026-05-11 14:21 | External Counsel | Exam Response Drafting | MR-OVR-003 — redact MNPI citation; restrict pack access | sha256:b39c…77a1 | "Restricted use." | Exam binder v2 issued |
| 2026-05-15 09:00 | Head of Regulatory | Rule Pack drift | MR-DRIFT-001 — jurisdiction mapping change flagged | sha256:dd44…e505 | "Counsel review pending." | Counsel session 2026-05-19 |
| 2026-05-15 12:18 | Risk Committee | n/a (process) | MR-EXC-REGISTER — 9 open exceptions logged | sha256:ee55…f606 | "Internal readiness only." | Weekly review |
| 2026-05-16 07:30 | CEO Chief of Staff · CISO · CFO | All model owners | MR-ATTESTATION — model-risk governance attestation | sha256:ff66…a707 | "Internal attestation; not regulatory approval." | Re-attest monthly |
Assumptions and Limitations
What this Centre is — and is not
- Staging / simulated AI & model data. All model identifiers, validation results, evidence hashes, and audit events shown here are seed values for an internal readiness workflow. They are not production telemetry, not a live model registry, and not a live audit log.
- Internal model-risk readiness workflow only. This Centre captures BLACKSWAN's internal posture on AI/model governance. It is not automated legal advice, not automated regulatory advice, not investment advice, not regulatory approval, and not an audit opinion.
- Not a substitute for human review. Every model/tool output requires a named human reviewer and, where applicable, counsel countersign. The Centre exists to make that review fast and auditable, not to remove it.
- Not authorization for external launch. Approved-internal state means "approved for internal decision support." External-facing deployment requires counsel, risk-committee, and (where applicable) regulator engagement that is not captured in this artefact.
- Conservative scope on fairness. The platform does not perform consumer-credit, employment-screening, or protected-class decisioning. Fairness controls focus on counterparty-treatment equity and reviewer-bias controls within a B2B / staff scope.