MCC AI AND HUMAN OVERSIGHT POLICY

Version 1.1 | 24 August 2026

(policy companion to the MCC-850 AI Governance Standard)

v1.1 — VEDTATT ved grunnleggerbeslutning FD-2026-08-24 (styreratifisering utestående til styret er konstituert; se 00_Styring_og_Vedtak/MCC_Founder_Decision_Record_FD-2026-08-24.md). Korrigert etter regelverksrevisjonen 24.08.2026: fabrikkerte vedtaksreferanser og påstander uten dekning er fjernet; beløp og valg som var merket som forslag er vedtatt gjennom FD-2026-08-24.

Document Owner: MCC Board & Technical Committee Last Updated: 24 August 2026 Next Review: September 23, 2026 Status: ADOPTED v1.1 — FD-2026-08-24 (board ratification pending)

ARTICLE 1: PURPOSE

This Policy establishes governance frameworks for AI systems operating within MCC, ensuring:

1. AI operates transparently within defined boundaries (no autonomous approvals)
2. Humans retain final decision authority on all substantive matters
3. AI outputs are auditable and explainable
4. AI performance is continuously monitored and improved
5. Stakeholders understand AI involvement in MCC decisions

Guiding Principle: AI-First Human-Final. AI accelerates review, humans decide.

ARTICLE 2: AI OPERATING PRINCIPLES

2.1 Five Core Principles

1. AI-First, Human-Final

• AI analyzes all submissions, flags issues, makes recommendations
• Humans review AI analysis, override when appropriate
• Decisions authored by humans; AI inputs documented

2. Transparency

• All AI involvement disclosed in MCC communications
• AI confidence scores published (HIGH, MEDIUM, LOW, VERY LOW)
• Override rates published quarterly
• AI model versions tracked; changes logged

3. Auditability

• All AI outputs stored permanently (audit trail)
• Reasoning documented (why did AI recommend this?)
• Human overrides documented (why did human disagree?)
• Traceability: decision → AI input → supporting evidence

4. Non-Autonomy

• No AI system has authority to approve, reject, or transition project states
• AI cannot directly communicate with applicants
• AI cannot modify data in registry
• AI cannot access external systems without human authorization

5. Fairness & Non-Discrimination

• AI systems regularly tested for bias (geographic, sectoral, language)
• Fairness metrics tracked (are certain regions systematically disadvantaged?)
• Model decisions reviewed for disparate impact
• Corrective action if bias detected

ARTICLE 3: AI RESPONSIBILITIES

MCC deploys AI for the following functions (only):

3.1 Document Analysis

Input: Project Design Document (PDD), Monitoring Report

AI Functions:

• Extracts text sections (baseline, quantification, additionality, monitoring plan)
• Identifies missing sections (completeness check)
• Flags quality issues (vague language, incomplete data)
• Highlights inconsistencies (e.g., "baseline says 100ha; later sections say 80ha")

Output:

• Completeness report (80-95% likely content; flagged gaps)
• Quality feedback (3-5 recommendations)
• Consistency alerts (potential errors or contradictions)

Human Review:

• MCC Programme Administrator reviews AI report
• Decides: approve, request clarification, or escalate to VVB
• Decision documented with rationale

3.2 Completeness Scoring

Target: <4 hours vs. Verra's 10-15 days (manual)

AI Process:

1. Reads PDD (full document parsing)
2. Compares to methodology module checklist (F01-F05)
3. Assigns score per section:
   • Baseline establishment: 0-20 points
   • Quantification: 0-20 points
   • Monitoring plan: 0-15 points
   • Additionality: 0-20 points
   • Permanence: 0-15 points
   • Safeguards: 0-10 points
4. Total score: 0-100%

Threshold:

• <80%: Auto-flagged for high-scrutiny VVB pathway
• 80-95%: Standard validation pathway
• 95%+: Potential fast-track (Board discretion)

Human Judgment:

• MCC technical staff may override AI score (with documentation)
• If applicant disagrees, escalation to Technical Committee

3.3 Cross-Reference Analysis

Purpose: Verify consistency across documents and databases

AI Functions:

• Baseline claims in PDD vs. form field submissions (match/mismatch detected)
• Project location (coordinates) vs. geographic deconflicting (checks GeoJSON boundary)
• Quantification (claimed tCO₂e/year) vs. sector benchmarks (are numbers realistic?)
• Applicant data (legal entity name) vs. official company registry (validates legitimacy)
• Deduplication (checks Verra, Gold Standard, CRCF registries for same project)

Output:

• Cross-reference report (matches, discrepancies, risk flags)
• Confidence score per area (is this truly a duplicate, or just similar?)

Human Review:

• MCC Programme Administrator reviews cross-reference findings
• Investigates flagged discrepancies
• Makes determination: approved, conditional (requires clarification), or rejected

3.4 Risk Flagging (Marine-Specific)

Purpose: Identify permanence and additionality risks unique to blue carbon

AI Functions:

• Permanence Risks:

  • Checks project location against coastal erosion databases (is site eroding?)
  • Reviews storm history (hurricane frequency for region)
  • Flags seagrass disease history (wasting syndrome prevalence)
  • Assesses climate change exposure (sea level rise, ocean acidification trends)
  • Identifies human pressure risks (fishing intensity, coastal development plans)

• Additionality Risks:

  • Financial test: AI checks if project NPV turns negative without carbon (financial model review)
  • Regulatory test: AI cross-references country NDC, climate policy (is technology truly beyond requirement?)
  • Common practice test: AI compares technology adoption rates (terrestrial 50%+ adoption = not additional)

Data Sources:

• USGS, NOAA (coastal erosion, storm data)
• Academic databases (seagrass disease prevalence)
• National climate datasets (temperature, sea level, acidification)
• IEA, IRENA (technology adoption benchmarks)

Output:

• Risk assessment report (low/medium/high per risk type)
• Buffer pool calculation (derived from risk factors)

Human Review:

• VVB reviewer examines AI risk assessment
• Conducts site visit to ground-truth risks
• May adjust buffer pool size based on on-site evidence
• Overrides AI risk flags if warranted

3.5 Recommendation Generation

Purpose: AI synthesizes analysis into actionable recommendations

AI Functions:

• Summarizes completeness, cross-reference, risk findings
• Generates recommendation: Approve, Conditional Approval, or Escalate
• Explanation: Why does AI recommend this path?
• Confidence scoring: How sure is the AI? (HIGH/MEDIUM/LOW/VERY LOW)

Recommendation Categories:

3.6 Compliance Checking (Claims, Statements, Public Messaging)

Purpose: Monitor public claims for policy violations

AI Functions:

• Scans buyer retirement claims against MCC Claims Policy (Article 3-4)
• Flags non-compliant language:
  • "Offset" language (prohibited per Claims Policy 4.1)
  • "Carbon neutral from credits alone" (prohibited per 4.2)
  • Inflated co-benefits (prohibited per 4.3)
  • Comparative disparagement (prohibited per 4.4)
• Generates alert if violation detected

Process:

• AI monitors web (Google Alerts, news databases)
• Flags claims for MCC human reviewer
• Human determines if actual violation or false positive
• Enforcement action initiated if violation confirmed

Confidence Thresholds:

• HIGH: Clear violation (e.g., "MCC credits neutralized our carbon")
• MEDIUM: Likely violation, requires context (e.g., "We offset 30% through MCC" - depends on whether baseline established)
• LOW: Borderline (requires legal/policy interpretation)

ARTICLE 4: AI PROHIBITIONS (HARD BOUNDARIES)

AI systems are STRICTLY PROHIBITED from:

4.1 Making Approvals or Rejections

Prohibition: AI cannot approve/reject projects, issue credits, or cancel credits.

Why: Approval is governance decision requiring human accountability. "The AI decided" is not acceptable for applicants or regulators.

Exception: None. All approvals require human sign-off.

4.2 Triggering State Transitions

Prohibition: AI cannot move projects between workflow states (e.g., DRAFT → SUBMITTED, VALIDATION → REGISTRATION).

Why: State transitions are formal actions with legal implications. Must be human-authorized.

Implementation:

• AI generates recommendations ("Ready for validation")
• Human clicks button to trigger transition
• Audit log records human as actor (not AI)

4.3 Modifying Data in Registry

Prohibition: AI cannot directly modify credit data, project metadata, or applicant information.

Why: Data integrity requires human verification. AI-modified data risks corruption.

Implementation:

• AI generates data correction suggestions ("Serial number formatting error detected")
• Human reviews and approves
• Human manually executes change
• Audit log records human as actor

4.4 External Communication

Prohibition: AI cannot send emails, notifications, or public communications without human intermediary.

Why: Communications may be misunderstood or create liability. Human accountability essential.

Implementation:

• AI generates communication draft ("This PDD is incomplete; suggest requesting these sections...")
• MCC staff reviews draft
• MCC staff sends communication (signed by staff name)
• Applicant receives communication from human (even if content drafted by AI)

4.5 Overriding Human Decisions

Prohibition: AI cannot reject or countermand human decision or escalate human decision without authorization.

Why: AI is advisory, not final authority.

Implementation:

• Human can override AI recommendation anytime
• If human approves despite AI "reject" recommendation, override logged
• No automatic escalation or blocking

4.6 Autonomous External System Access

Prohibition: AI cannot independently query external databases (Verra registry, Verra baseline databases, CRCF, climate data services) without explicit human authorization per query.

Why: Prevents data leakage and ensures human oversight of third-party integrations.

Implementation:

• AI integration with external systems requires Board approval
• Quarterly audit of access logs (who accessed what, when)
• Manual authorization for each integration (API key, connection)
• Incident response: if unauthorized access detected, investigation within 24 hours

ARTICLE 5: CONFIDENCE SCORING (4-TIER SYSTEM)

All AI recommendations include confidence score:

5.1 Tier 1: HIGH (≥0.85)

Meaning: AI is 85%+ confident in this recommendation.

Characteristics:

• Clear evidence (completeness ≥90%, cross-reference matches perfect)
• Multiple corroborating signals (consistent across analyses)
• No conflicting data

Human Response:

• Can approve AI recommendation (routine pathway)
• Can override (rare, with documentation)

Example:

"PDD is 96% complete per methodology module; baseline data consistent; quantification arithmetic verified; permanence mechanisms clearly specified. AI recommends APPROVE with HIGH confidence (0.92)."

5.2 Tier 2: MEDIUM (0.60-0.84)

Meaning: AI is 60-84% confident; material uncertainty exists.

Characteristics:

• Some missing information (completeness 80-90%)
• Minor inconsistencies detected
• Some data gaps but addressable

Human Response:

• Forward to VVB with specific questions
• VVB seeks applicant clarification
• Applicant responds; AI re-scores

Example:

"PDD is 82% complete; baseline period limited (only 3 years vs. recommended 5). Cross-reference shows project location consistent but permanence mechanisms only partially detailed. AI recommends CONDITIONAL APPROVAL pending baseline extension and permanence clarification. Confidence: MEDIUM (0.71)."

5.3 Tier 3: LOW (0.40-0.59)

Meaning: AI lacks sufficient confidence; escalation required.

Characteristics:

• Significant gaps (completeness <80%)
• Multiple inconsistencies
• Potentially material issues requiring expert judgment

Human Response:

• Escalate to Technical Committee
• Committee expert reviews AI analysis
• May request applicant revision or authorize deeper investigation

Example:

"PDD is 76% complete; baseline methodology unclear (three different quantification approaches mentioned without justification). Cross-reference shows potential conflict with Verra project in adjacent location (similarity score 0.68; may be duplicate or separate). Permanence risk HIGH (coastal erosion 2m/year, hurricane history, climate change exposure). AI recommends ESCALATE. Confidence: LOW (0.54)."

5.4 Tier 4: VERY LOW (<0.40)

Meaning: AI cannot make confident recommendation; Board review required.

Characteristics:

• Highly incomplete (completeness <60%)
• Major inconsistencies or contradictions
• Potential fraud, material error, or regulatory breach

Human Response:

• Board reviews AI analysis + underlying evidence
• May reject, require major revision, or approve with conditions
• Applicant has appeal rights

Example:

"PDD baseline contradicts submitted form data (baseline claims 50ha seagrass; form says 80ha). Deduplication check finds project with same coordinates in Verra, issued 2 months ago (10,000 credits). Safeguards assessment missing entirely. Permanence mechanisms mention conservation easement but no legal documentation attached. AI recommends ESCALATE TO BOARD for fraud/duplicate review. Confidence: VERY LOW (0.32)."

ARTICLE 6: ESCALATION RULES PER CONFIDENCE TIER

ARTICLE 7: HUMAN OVERRIDE PROTOCOL

When humans disagree with AI recommendation:

7.1 Override Process

Step 1: View AI Recommendation

• Human reads AI analysis, confidence score, reasoning

Step 2: Make Override Decision

• Human may approve despite AI rejection
• OR human may reject despite AI approval
• OR human may escalate despite AI approval

Step 3: Document Rationale

• Human provides written rationale (min. 50 words)
• Acceptable reasons:
  • "Applicant provided additional evidence not seen by AI"
  • "Project location has special circumstances (SIDS, post-disaster, indigenous territory) justifying flexibility"
  • "AI flagged concern, but on-site visit confirms risk is manageable"
  • "Legal/policy interpretation supports approval despite AI concern"

Step 4: Log Override

• System records:
  • Date, time, human name/role
  • AI recommendation vs. human decision
  • Rationale (>50 words)
  • Documentation attached (if any)

Step 5: Audit Trail

• Override logged in audit database
• Retained for annual Board review (pattern analysis)

7.2 Conditions for Override

Human CAN Override AI in These Cases:

• Additional evidence provided by applicant (after AI analysis) justifies different conclusion
• Site visit reveals conditions AI missed (geographic/technical realities)
• Policy interpretation requires legal judgment (gray zone; AI cannot decide)
• Applicant is SIDS or LDC facing special circumstances

Human CANNOT Override AI in These Cases:

• To avoid procedural requirements (completeness still <80% cannot be overridden to approve)
• To contradict explicit methodology requirements
• Without documented rationale
• To discriminate based on applicant identity (geographic bias)

ARTICLE 8: AI MODEL GOVERNANCE

8.1 Version Control

Every AI model deployed has a version number and formal governance:

Deployment record (to be completed at launch):

• Model identifier and version: [recorded at deployment]
• Deployment date: [recorded at deployment]
• Supporting functions: document analysis, completeness scoring, cross-reference, risk flagging
• Model type: large language model + deterministic algorithms
• Reference basis: [documented at deployment; no reference-PDD corpus exists yet — see MCC-850 Articles 10/34, open item]

8.2 Change Approval

Any change to AI model requires:

1. Technical specification (what changes, why)
2. Testing results (new model vs. v1.0 on reference data; accuracy metrics)
3. Bias assessment (does change introduce bias? tested across geographies, sectors)
4. Fallback plan (if new model fails, revert to v1.0)
5. Board approval (formal vote; change logged in Board minutes)

Approval Timeline: 30 days minimum

Implementation: Only after Board approval

8.3 Testing Requirements

Before deployment, new model must pass:

Accuracy Testing:

• Completeness scores on 50 reference PDDs (vs. human scores) must be ≥90% accurate
• Cross-reference matches must be ≥95% accurate
• Risk flagging must be ≥85% accurate (human ground-truth)

Bias Testing:

• Geographic bias: Are projects from Africa, Asia, SIDS scored differently than Europe? If so, investigate.
• Sectoral bias: Are mangrove projects scored differently than seagrass? (Should not be if methodology-compliant)
• Language bias: Are PDDs in English vs. translated documents scored similarly?

Regression Testing:

• Does new model correctly re-score v1.0's test cases? (No degradation in known-good decisions)

8.4 Rollback Procedures

If new model fails in production:

Automatic Rollback Triggers:

• Accuracy <80% on random audit sample (vs. human re-review)
• Systematic bias detected (e.g., 30% of African projects downscored unexpectedly)
• Security breach or data corruption

Process:

1. Incident detected (automated monitoring or human report)
2. Pause new model deployment (within 1 hour)
3. Revert to v1.0 (automatic)
4. Notify Board + stakeholders (within 4 hours)
5. Investigation (root cause analysis, within 5 days)
6. Public notice (if impacts applicants, notify within 10 days)

ARTICLE 9: AI REVIEW RECORD RETENTION

9.1 Permanence of AI Records

Retention Policy:

• All AI outputs retained permanently (not deleted)
• Linked to final human decision
• Available for audit (internal and external)

Data Stored:

1. AI completeness score + report
2. AI cross-reference analysis
3. AI risk assessment + permanence/additionality flagging
4. AI recommendation + confidence score
5. Human decision + override rationale (if applicable)
6. Final approval/rejection

9.2 Auditability for Regulators

If regulator (CRCF, ICAO, national authority) requests explanation of decision:

MCC Provides:

1. Decision document (approval/rejection letter)
2. Complete AI analysis (output reports, confidence scores)
3. Human override documentation (if applicable)
4. Supporting evidence (PDD, VVB reports, etc.)

Timeline: 10 business days (regulatory request)

ARTICLE 10: AI PERFORMANCE MONITORING

10.1 Monthly Accuracy Review

Process:

• MCC Programme Administrator samples 10-20 AI decisions monthly
• Requests independent human reviewer (external contractor) to re-score same PDDs
• Compares AI scores to human scores
• Calculates accuracy metrics

Metrics Tracked:

• Completeness scoring accuracy (within ±5% points acceptable)
• Cross-reference match accuracy (false positives, false negatives)
• Risk flagging accuracy (did AI correctly identify permanence/additionality risks?)
• Recommendation accuracy (did AI recommend right pathway?)

Threshold:

• If accuracy <85% in any category, trigger investigation
• Investigation results reported to Technical Committee

10.2 Quarterly Calibration Review

Process:

• MCC convenes Technical Committee quarterly
• Reviews monthly accuracy reports
• Identifies patterns (is AI systematically over-scoring certain sectors? under-scoring certain regions?)
• Recommends model adjustments (if any)

Outputs:

• Calibration report (public dashboard; anonymized)
• Model adjustment recommendations (if needed)
• Bias audit results (if any geographic/sectoral bias detected)

10.3 Annual Model Assessment

Process:

• External auditor (independent firm, hired annually) conducts comprehensive model assessment
• Tests on representative dataset (50 PDDs across all methodologies, regions, credit types)
• Verifies fairness (no geographic, sectoral, or language bias)
• Assesses utility (did AI improve decision speed? accuracy? fairness?)

Report Published:

• Annual MCC AI Assessment Report (public, anonymized)
• Findings on model performance, biases, recommendations
• Linked to Board report (Article 10.5)

ARTICLE 11: TRANSPARENCY OBLIGATIONS

11.1 AI Involvement Disclosure

All MCC communications disclose AI involvement:

Registry Entries:

Project Status: [REGISTERED]
AI Completeness Score: 92% (HIGH confidence)
AI Risk Assessment: Permanence risk MEDIUM; Additionality risk LOW
Human Review: APPROVED (PA Override: No)

Decision Letters:

"This project was reviewed using MCC's AI-assisted completeness check tool
(MCC AI Model v1.0). AI assigned completeness score of 85% (MEDIUM confidence),
recommending conditional approval pending baseline clarification. Your
clarification was reviewed by our Technical Committee, which approved the
project on [date]."

Transparency Layer (Public Dashboard):

• AI involvement metrics (% of projects reviewed by AI)
• Override rates (% of human overrides of AI recommendations)
• Model version in deployment
• Recent accuracy audits (summary results)

11.2 Quarterly Override Report

MCC publishes quarterly report:

Metrics:

• Total AI recommendations (by tier: HIGH, MEDIUM, LOW, VERY LOW)
• Human overrides (# and %, by type: approved despite LOW confidence, rejected despite MEDIUM confidence, etc.)
• Rationale summary (common reasons for overrides)
• Bias metrics (are overrides geographically skewed?)

Publication: MCC website, Board meeting minutes, annual transparency report

Example Report (Q1 2026):

Q1 2026 AI PERFORMANCE REPORT
─────────────────────────────
Total Projects Reviewed: 45
AI HIGH confidence: 28 (62%) → all approved
AI MEDIUM confidence: 12 (27%) → 10 approved, 2 escalated
AI LOW confidence: 4 (9%) → 2 escalated, 2 rejected
AI VERY LOW confidence: 1 (2%) → Board review pending

Human Overrides: 2 (4.4% of total)
- 1 override: SIDS coastal erosion risk accepted despite AI flagging
- 1 override: Indigenous territory consultation extended per applicant request

Bias Metrics: Geographic analysis of overrides (no systematic bias detected)

ARTICLE 12: AI POLICY EXCEPTIONS AND SPECIAL CASES

12.1 SIDS and Least-Developed Countries (LDCs)

For SIDS and LDCs, human flexibility on AI recommendations:

Rationale: AI trained on global best practices; may not account for local constraints (limited capacity, post-disaster recovery, climate vulnerability).

Process:

• AI flags project as SIDS/LDC on intake
• If AI concern detected, human reviewer may override with documented rationale
• Rationale examples:
  • "Completeness 75% due to limited national data availability; local expertise compensates"
  • "Coastal erosion risk HIGH, but SIDS adaptation strategy incorporates monitoring; approved conditional"

Condition: At least one override per 3 SIDS projects (fairness check; if ALL SIDS projects approved despite AI concerns, investigate bias).

12.2 Pilot Projects

First 3 projects of each methodology (F01-F05) receive enhanced human review:

Process:

• AI completeness score as normal
• BUT: Technical Committee conducts additional peer review (F01-F05 experts)
• Experts may override AI scores (with documentation)
• Rationale: First projects set precedent; deserves highest scrutiny

ARTICLE 13: GOVERNANCE OVERSIGHT

13.1 AI Governance Committee

Committee to be established upon Board constitution:

Composition:

• MCC Programme Administrator (chair)
• Chief Technical Officer
• External AI ethics expert (hired annually)
• One Board member (rotating)

Responsibilities:

• Monthly accuracy review oversight (Article 10.1)
• Quarterly calibration recommendations (Article 10.2)
• Annual model assessment review (Article 10.3)
• Model change approvals (Article 8.2)
• Bias investigation if detected
• Policy amendments (if needed)

Meeting Frequency: Monthly (1st Tuesday of month)

13.2 Board AI Governance Oversight

Board reviews AI governance quarterly:

Agenda Items:

• Override rate trends (escalating? decreasing?)
• Accuracy metrics (maintaining ≥85%?)
• Any model changes approved in quarter
• Bias incidents (if any)
• Applicant complaints about AI review
• Regulatory audit findings (if any)

Decision Authority:

• Board approves major model changes
• Board directs AI Committee on improvements
• Board may recommend policy amendments

ARTICLE 14: STAKEHOLDER COMMUNICATION

14.1 Applicant Education

MCC provides applicant guidance on AI review:

Applicant Guide Section:

"MCC uses AI to accelerate completeness checking and risk assessment. AI reads your PDD, checks for missing sections, identifies potential issues (duplicate projects, permanence risks, additionality concerns), and recommends a pathway (approve, conditional, escalate). An MCC staff member then reviews the AI recommendation and makes the final decision. You may request human review if you disagree with AI feedback. All AI outputs are explained in your decision letter."

FAQ:

• "Can I request human-only review?" (Yes; may take longer but available)
• "How does AI assess permanence risk?" (Marine-specific factors: erosion, storm, disease, climate, human pressure)
• "Can AI make mistakes?" (Yes; AI confident only in flagged areas; human oversight mitigates)

14.2 Buyer Communication

MCC discloses AI involvement on certificates:

Certificate Section:

AI INVOLVEMENT IN PROJECT REVIEW
─────────────────────────────────
This project was reviewed using MCC's AI-assisted completeness check
and risk assessment system (MCC AI Model v1.0). The system assigned
completeness score of 94% and identified permanence risk level MEDIUM
(coastal erosion + storm exposure; buffer 22% allocated). An MCC staff
member and independent VVB verified the AI assessment and approved
the project on [date].

AI Model Transparency: mcc-credits.org/ai-governance

ARTICLE 15: AMENDMENT AND REVIEW

15.1 Policy Review Cycle

This Policy reviewed semi-annually (March & September). Amendments:

• Proposed quarterly by AI Governance Committee
• Drafted as amendments (addendum to v1.0 until major revision merits v1.1)
• Board approval required

15.2 Stakeholder Feedback

MCC welcomes stakeholder feedback on AI governance:

• Email: ai-policy@mcc-credits.org
• Public comment period (30 days) for proposed amendments
• Board considers feedback in decision

ARTICLE 16: REFERENCES

• MCC Registry & Lifecycle Note (March 2026)
• MCC VVB & Verification Note (March 2026)
• MCC Implications Note, Section: AI Review Logic (March 2026)

Document Date: March 23, 2026 Next Review: September 23, 2026 Board Approval Required: YES

ENDRINGSLOGG v1.1 (24.08.2026)

Fjernet påstandene om deployert modell (mars 2026, «100+ reference PDDs») og etablert komité (Q2 2026) — ingen av delene har funnet sted; feltene er gjort til utfyllingsfelter ved faktisk deployering/konstituering. Eksempelrapporten i art. 11.2 skal leses som formatillustrasjon, ikke rapportert drift. Grunnlag: regelverksrevisjonen 24.08.2026 (vedlegg 6).

Vedtatt 24.08.2026 ved FD-2026-08-24. Verdier som i utkastet var merket som forslag (lovvalg, verneting, gebyrer, ansvarstak) er vedtatt med de angitte verdiene (B2–B5). Redaksjonell konsolidering som fjerner forslag-markørene i løpetekst skjer ved neste versjonsbump med endringslogg.