---
layout: raw-data.njk
---
# Stage 5: Evaluation Against Generative AI Guidance

## Date and Agent
- Date: 2025-11-19
- Agent: Claude (general-purpose agent)

## Objective

Evaluate identification standards presentation against GCDO's generative AI guidance principles to identify alignment, gaps, and opportunities. Validate or challenge findings from Stages 1-4 against best practice guidance for government documentation and technical standards.

## Methodology

### Tools and Approaches Used

1. **MCP Server Familiarization**:
   - Used `get_document_stats` to understand AI guidance collection (1,393 nodes, 23 documents)
   - Used `get_schema` to understand structure (DocumentNode entities with CHILD_OF and SEMANTIC_SIMILARITY relationships)
   - Reviewed document composition to understand guidance scope

2. **Semantic Search Strategy**:
   Performed targeted semantic searches aligned with Stage 4 priority findings:
   - Active voice and plain language principles
   - User-centered content design and organization
   - Technical standards documentation requirements
   - Information hiding and progressive disclosure
   - Page structure and content hierarchy
   - Writing for diverse audiences (technical practitioners, specialists)
   - Accessibility and clear communication

3. **Evidence Collection**:
   - Extracted relevant guidance principles with DocRef citations
   - Mapped AI guidance to Stage 4 themes
   - Identified validation and challenges for each finding
   - Documented specific recommendations from AI guidance

4. **Critical Analysis**:
   - Evaluated Stage 4 findings against AI guidance evidence
   - Identified where AI guidance validates Stage 4 priorities
   - Identified any contradictions or alternative perspectives
   - Extracted actionable recommendations for Phase 2

## Key Findings

### CRITICAL DISCOVERY: Limited Direct Applicability

**Important Context**: The generative-ai-guidance-gcdo MCP server contains **AI/GenAI-specific guidance** for public service use of artificial intelligence systems, NOT general content design guidance for government documentation.

**Document Composition**:
- 23 documents focused on responsible AI use in government
- Topics: AI governance, transparency, security, privacy, bias, procurement, accessibility in AI systems
- Primary audience: Public servants developing or deploying AI/GenAI solutions
- Primary purpose: Responsible AI development and deployment, not general documentation standards

**Implication for Evaluation**:
The AI guidance MCP server does NOT contain the digital.govt.nz content design guidance that Stage 3 identified as directly addressing Stage 2 issues (active voice, plain language, user-centered organization, page structure, findable content). That guidance exists on digital.govt.nz website but is not in this MCP server.

**What This Server CAN Validate**:
- General principles about transparency, clarity, and user-centered design (from AI governance guidance)
- Accessibility principles that apply to all government information
- Documentation requirements for technical systems (AI systems parallel to identification standards)

**What This Server CANNOT Directly Address**:
- Specific content design principles for technical standards documentation
- Plain Language Act 2022 implementation guidance
- Page structure and findability best practices
- Active voice vs passive voice guidance for government writing

**Adjusted Evaluation Approach**:
This evaluation will extract **relevant principles** from AI guidance that can be **analogously applied** to identification standards documentation, while acknowledging the limitation that this is not the primary content design guidance Stage 3 identified.

## AI Guidance Collection Overview

### Database Statistics

**Collection Size**: 1,393 DocumentNode entities across 23 documents
- **Embeddings**: 1,027 nodes with 768-dimensional embeddings (73.7% coverage)
- **Relationships**:
  - CHILD_OF: 1,212 hierarchical links
  - SEMANTIC_SIMILARITY: 10,215 connections (K=10, threshold=0.7)
- **Virtual Nodes**: 150 (10.8%) - structural placeholders for missing parents

### Document Types and Focus

**Primary Documents**:
1. Public Service AI Framework (foundational principles)
2. Responsible AI Guidance for the Public Service: GenAI
3. Topic-specific guidance: Governance, Security, Privacy, Transparency, Bias/Discrimination, Accessibility
4. Implementation guidance: Procurement, Skills/Capabilities
5. Specialized topics: Misinformation/Hallucinations, Accountability/Responsibility
6. Supporting materials: Glossary of AI Terms, Cloud Jurisdictional Risk Guidance

**Content Categories**:
- Text: 66 nodes
- List: 49 nodes
- Metadata: 33 nodes
- Example: 19 nodes
- Structural: 13 nodes
- Figure: 7 nodes

**Key Observation**: This is a **specialized technical guidance collection** for AI systems, similar in nature to identification standards (specialized technical guidance for identification management). Parallel structure suggests analogous principles may apply.

## Validation of Stage 4 Priority Findings

### Finding 1: Conformance-Centered Organization

**Stage 4 Finding Summary**:
Tom identified conformance as "the whole point" of the identification standards. Stage 4 found conformance is semantically isolated (0 neighbors above 0.75), treated as peripheral despite being users' primary goal, and "tucked away" when it should be central organizing framework.

**Stage 4 Recommendation**:
Make conformance the primary organizing framework - structure entire resource around conformance workflow (Assess → Implement → Document → Get Assessed → Maintain).

#### AI Guidance Perspective

**Relevant AI Guidance Principles**:

AI guidance does NOT directly address conformance-centered organization for technical standards. However, it provides **analogous principles** for organizing technical documentation:

1. **Governance Framework Principle**: "Agencies should publish information about their development and use of AI, barring reasonable exceptions... This will help maintain transparency and trust in public service AI use. Agencies might consider publishing information about the type of AI they're using, what stage the project is at, the intent of use or the problem it's trying to solve, and an overview of how the system is being used and by whom." ([DocRef](https://docref.digital.govt.nz/nz/generative-ai-guidance-gcdo/governance-and-genai-in-the-public-service/2025/en/#part2-det5))

   **Analogous Application**: Just as AI governance recommends organizing around implementation stages and usage, identification standards could organize around conformance stages (assess, implement, document, get assessed).

2. **Accountability Principle**: "This should include the application of relevant regulatory and governance frameworks, reporting, auditing and/or independent reviews." ([DocRef](https://docref.digital.govt.nz/nz/generative-ai-guidance-gcdo/public-service-ai-framework/2025/en/#part3-det5-para2))

   **Analogous Application**: Conformance assessment is the "auditing and independent review" mechanism for identification standards - should be prominently featured.

3. **Transparency Framework**: AI guidance emphasizes transparency about systems, processes, and requirements. Conformance process transparency should be central.

#### Validation/Challenge

**VALIDATION (Indirect)**: AI guidance principles about transparency, governance frameworks, and accountability support the idea that compliance/assessment processes should be prominent and central to technical documentation. The parallel between "AI system lifecycle" and "identification conformance lifecycle" suggests similar organizational approaches are appropriate.

**LIMITATION**: AI guidance does not directly validate "conformance-centered organization" because AI guidance itself is not organized around conformance to standards - it's organized around responsible AI principles. This is a limitation of using this guidance collection for evaluation.

**STRENGTH OF STAGE 4 FINDING**: Stage 4's conformance-centered recommendation is grounded in user research (Tom's observation: "the whole point"), data evidence (semantic isolation), and practical user needs (practitioners seeking conformance). This finding stands on its own merits regardless of AI guidance validation.

#### Specific Recommendations from AI Guidance

**Process-Centered Documentation**: "We strongly recommend publishing an up-to-date register of all GenAI use in your agency. This is a commitment to transparency – and helps to connect with other government agencies on how GenAI is being used." ([DocRef](https://docref.digital.govt.nz/nz/generative-ai-guidance-gcdo/transparency-and-genai/2025/en/#part2-subpart3))

**Recommendation for Identification Standards**: Just as AI guidance recommends registers and transparency about AI usage, identification standards should feature conformance registers, assessment checklists, and evidence documentation prominently.

**Verdict on Finding 1**: **SUPPORTED BY ANALOGY** - AI guidance principles about transparency, process documentation, and accountability align with conformance-centered organization, though not directly validated by this guidance collection.

### Finding 2: Passive Voice Issues

**Stage 4 Finding Summary**:
Passive voice pervades guidance materials (15+ Tom annotations noting confusion). Stage 2 found passive voice makes content "very vague" while active voice sections are "much clearer." Stage 3 identified digital.govt.nz guidance: "Use the active voice, where possible" and "Use 'you' and 'your' when talking to the reader."

**Stage 4 Recommendation**:
Systematic active voice conversion in all guidance materials (not core standards). Apply digital.govt.nz Tone and Voice guidance throughout.

#### AI Guidance Perspective

**Relevant AI Guidance Principles**:

The AI guidance MCP server does NOT contain the digital.govt.nz Tone and Voice guidance that Stage 3 identified. However, I can observe AI guidance's own writing style:

1. **AI Guidance Uses Active Voice**: Throughout the guidance documents:
   - "Engage disabled people as stakeholders from conception of the GenAI solution to its deployment. Develop a vision statement and policy that prioritise inclusion and accessibility." ([DocRef](https://docref.digital.govt.nz/nz/generative-ai-guidance-gcdo/accessibility-and-genai/2025/en/#part3-subpart1-para1))
   - "Consider disabled people's experiences in training data, models and outputs to remove bias. Build GenAI to mandated accessibility standards and test it with disabled people." ([DocRef](https://docref.digital.govt.nz/nz/generative-ai-guidance-gcdo/accessibility-and-genai/2025/en/#part3-subpart2-para1))
   - "Check the quality of results produced by GenAI systems with trusted sources." ([DocRef](https://docref.digital.govt.nz/nz/generative-ai-guidance-gcdo/misinformation-hallucinations-and-genai/2025/en/#part3-subpart2-section2))

2. **Direct Address to Reader**: AI guidance consistently uses "you" and direct instructions:
   - "You use GenAI to summarise a long document... You should also ask a colleague to review the summary... You should not publish it until you've double-checked..." ([DocRef](https://docref.digital.govt.nz/nz/generative-ai-guidance-gcdo/accountability-responsibility-and-genai/2025/en/#part5-ex1))
   - "Before using the GenAI system, you check your agency's policies and record your use of GenAI in the public register." ([DocRef](https://docref.digital.govt.nz/nz/generative-ai-guidance-gcdo/transparency-and-genai/2025/en/#part2-subpart3))

3. **Clear, Direct Instructions**: AI guidance avoids passive constructions:
   - NOT: "Classification markings should be applied..."
   - BUT: "Clear classification marking of information allows for easy filtering techniques" (declarative) or implicit imperative ([DocRef](https://docref.digital.govt.nz/nz/generative-ai-guidance-gcdo/classify-information/2019/en/#para3))

#### Validation/Challenge

**VALIDATION (By Example)**: AI guidance **models the practice** of active voice and direct address. This is government technical guidance written in active voice, demonstrating that technical guidance can and should use active voice for clarity.

**CRITICAL INSIGHT**: If GCDO's AI guidance (highly technical, specialized domain, similar to identification standards) uses active voice and direct address throughout, this validates that identification standards guidance can and should do the same.

**GOVERNMENT STANDARD**: The fact that government AI guidance consistently uses active voice suggests this is government standard practice for technical guidance, supporting Stage 3's finding about digital.govt.nz requirements.

#### Specific Recommendations from AI Guidance

**By Modeling**: The AI guidance collection demonstrates effective active voice usage:
- "Ensure all GenAI systems used by your agency are certified and accredited before they're made available to users" ([DocRef](https://docref.digital.govt.nz/nz/generative-ai-guidance-gcdo/security-and-genai/2025/en/#part2-para1-3))
- "New software versions should be evaluated and tested before they're rolled out" ([DocRef](https://docref.digital.govt.nz/nz/generative-ai-guidance-gcdo/security-and-genai/2025/en/#part1-subpart1-det1-para2))

**Recommendation for Identification Standards**: Follow AI guidance's example - use active voice, direct address, clear imperative constructions for guidance materials.

**Verdict on Finding 2**: **STRONGLY VALIDATED BY EXAMPLE** - AI guidance models active voice usage in government technical guidance, demonstrating feasibility and appropriateness for identification standards.

### Finding 3: Content Fragmentation and "Tucking Away"

**Stage 4 Finding Summary**:
Detail expanders hide essential information (12+ Tom annotations: "no point to burying these"). Threshold information appears late when it should be upfront. Tom's directive: "Get rid of all detail expanders."

**Stage 4 Recommendation**:
Eliminate all detail expander syntax. Surface threshold information early in "Before You Start" sections. Use clear heading hierarchy - all information scannable.

#### AI Guidance Perspective

**Relevant AI Guidance Principles**:

AI guidance does not directly address progressive disclosure or detail expanders. However, it provides principles about information accessibility:

1. **Accessibility Principle**: "Accessibility is considering the needs of all potential users from the outset, engaging with individuals who have disabilities during the design process in order to create solutions that are genuinely usable by everyone. It also includes assistive technologies, screen readers, voice recognition software and alternative input devices." ([DocRef](https://docref.digital.govt.nz/nz/generative-ai-guidance-gcdo/glossary-of-ai-terms/2025/en/#part1-subpart1-para1))

   **Implication**: Detail expanders create accessibility barriers for screen readers and users who need to scan content. Hidden content is not "genuinely usable by everyone."

2. **Transparency Principle**: "Openness: Clearly communicating the purpose and capabilities of an AI system. This includes explaining what the system is designed to do and any limitations it may have." ([DocRef](https://docref.digital.govt.nz/nz/generative-ai-guidance-gcdo/glossary-of-ai-terms/2025/en/#part13-subpart1-para1-1))

   **Analogous Application**: Threshold information about conformance should be "clearly communicated" upfront, not hidden. Users need to understand scope and limitations before investing time.

3. **Clear Classification**: "Clear classification marking of information allows for easy filtering techniques such as outbound filtering and inspection by mail servers to lessen the risk of inadvertent information leakage via email." ([DocRef](https://docref.digital.govt.nz/nz/generative-ai-guidance-gcdo/classify-information/2019/en/#para3))

   **Implication**: Clear, visible marking and organization of information is valued throughout government guidance.

#### Validation/Challenge

**VALIDATION (Accessibility Basis)**: AI guidance's accessibility principles validate that all content should be visible and accessible. Hiding essential information contradicts accessibility requirements.

**VALIDATION (Transparency Basis)**: Transparency principles support surfacing important information upfront - users need to understand what's relevant before proceeding.

**AI GUIDANCE STRUCTURE**: Examining AI guidance documents themselves - they do NOT use detail expanders. All content is visible with clear heading hierarchy. This models the practice Stage 4 recommends.

#### Specific Recommendations from AI Guidance

**Accessibility Requirement**: "Accessibility means designing things to work for disabled people. The New Zealand Government has legal and ethical obligations to create accessible information and services, both for the public and for public servants." ([DocRef](https://docref.digital.govt.nz/nz/generative-ai-guidance-gcdo/accessibility-and-genai/2025/en/#h1-subtitle))

**Recommendation for Identification Standards**: Legal and ethical obligations to accessibility require visible content with clear hierarchy, not hidden detail expanders.

**Verdict on Finding 3**: **VALIDATED (Accessibility and Transparency Basis)** - AI guidance principles about accessibility and transparency support eliminating detail expanders and surfacing essential information.

### Finding 4: Standards-Guidance Integration

**Stage 4 Finding Summary**:
Navigation burden from separating standards (requirements) and guidance (implementation). Tom: "Does having separate pages really help? It's not for reading, it's for working through." 2,179 nodes across 8 documents for 4 topics. Implementation guides score higher in semantic searches than standards.

**Stage 4 Recommendation**:
Integrate standards and guidance into single cohesive documents with clear visual distinction between normative and explanatory content.

#### AI Guidance Perspective

**Relevant AI Guidance Principles**:

AI guidance does not directly address standards-guidance integration. However, I can observe the structure of AI guidance itself:

1. **AI Guidance Structure**: AI guidance documents integrate principles and implementation guidance:
   - "Inclusive, sustainable development: Public Service AI systems should contribute to inclusive growth and sustainable development... AI use should consider and address concerns about unequal access to technology." (Principle + implementation concern integrated) ([DocRef](https://docref.digital.govt.nz/nz/generative-ai-guidance-gcdo/public-service-ai-framework/2025/en/#part3-det1-para1))

2. **Principle + Practice Pattern**: Documents combine "what" (principles) with "how" (implementation):
   - Security guidance combines requirements ("Ensure all GenAI systems... are certified and accredited") with implementation details ("The certification process should validate that security controls... are in place") ([DocRef](https://docref.digital.govt.nz/nz/generative-ai-guidance-gcdo/security-and-genai/2025/en/#part2-para1-3))

3. **Example Scenarios**: AI guidance includes example scenarios that integrate principles with application:
   - Accountability example walks through principles applied in practice ([DocRef](https://docref.digital.govt.nz/nz/generative-ai-guidance-gcdo/accountability-responsibility-and-genai/2025/en/#part5-ex1))

#### Validation/Challenge

**VALIDATION (By Example)**: AI guidance **models integrated structure** - it does not separate "AI principles" from "AI implementation guidance" into different documents. Users get both together.

**DIFFERENT CONTEXT**: AI guidance is not a conformance framework with separate normative/non-normative distinction. It's all guidance. However, the integrated approach still supports reducing navigation burden.

**PRACTICAL INSIGHT**: The fact that AI guidance integrates principles and implementation in single documents suggests this approach works for technical government guidance.

#### Specific Recommendations from AI Guidance

**Integration Pattern**: AI guidance demonstrates integrated structure where principles, implementation guidance, and examples appear together in context, reducing navigation burden.

**Recommendation for Identification Standards**: Follow AI guidance pattern - integrate standards requirements with implementation guidance while maintaining visual distinction (formatting, headings) between normative and explanatory content.

**Verdict on Finding 4**: **SUPPORTED (By Example)** - AI guidance models integrated principle-implementation structure, demonstrating feasibility of reducing navigation burden while maintaining clear distinctions.

### Finding 5: User Journey and Entry Points

**Stage 4 Finding Summary**:
No clear "start here" or user pathways. 30 documents without guidance on which to choose. Tom: "When all pages are split into so many separated documents, you lose control over how people are approaching the information."

**Stage 4 Recommendation**:
Create explicit user journeys and entry points. "Start Here" section: Who are you? What do you need? Guided pathways for different roles.

#### AI Guidance Perspective

**Relevant AI Guidance Principles**:

AI guidance provides limited direct guidance on user pathways, but demonstrates some organizational approaches:

1. **Purpose-Driven Organization**: AI guidance organized by topic/concern rather than user type:
   - Security, Privacy, Governance, Transparency, Bias, Accessibility as separate pages
   - Users navigate by concern, not by role

2. **Compliance Documentation Concept**: "Many cloud providers will also have a range of documentation available for a range of issues. In Australia and New Zealand, this is also known as compliance documentation." ([DocRef](https://docref.digital.govt.nz/nz/generative-ai-guidance-gcdo/cloud-jurisdictional-risk-guidance/2024/en/#part10-para2))

   **Implication**: "Compliance documentation" is a recognized category - conformance documentation for identification standards fits this pattern.

3. **Diverse Stakeholder Engagement**: "Engage disabled people as stakeholders from conception of the GenAI solution to its deployment." ([DocRef](https://docref.digital.govt.nz/nz/generative-ai-guidance-gcdo/accessibility-and-genai/2025/en/#part3-subpart1-para1))

   **Implication**: Recognition of diverse stakeholder needs - suggests need for multiple entry points serving different audiences.

#### Validation/Challenge

**LIMITATION**: AI guidance itself does not model strong user pathway design - it's organized by topic rather than user journey. This limits its ability to validate Stage 4's user pathway recommendation.

**DIFFERENT CONTEXT**: AI guidance serves public servants generally (broad audience); identification standards serve specific practitioner roles (credential providers, assessors). More role-specific pathways may be appropriate for identification standards.

**PARTIAL SUPPORT**: AI guidance's recognition of diverse stakeholders supports the concept of serving multiple user types, even if execution differs.

#### Specific Recommendations from AI Guidance

**Stakeholder Diversity**: AI guidance principle: different stakeholders have different needs and should be engaged differently.

**Recommendation for Identification Standards**: Explicitly identify user types (credential providers, assessors, relying parties) and create pathways serving each.

**Verdict on Finding 5**: **PARTIALLY SUPPORTED** - AI guidance recognizes diverse stakeholder needs but doesn't model strong user pathway design. Stage 4 recommendation stands on user research, not AI guidance validation.

### Finding 6: Plain Language Act Compliance

**Stage 3/4 Finding Summary**:
Stage 3 identified Plain Language Act 2022 as mandatory legal requirement: content must be "clear, concise and well organised" and "appropriate to the audience." Stage 4 synthesized this with passive voice and structure issues.

**Stage 4 Recommendation**:
Phase 2 must comply with Plain Language Act 2022 through active voice, clear structure, appropriate language, and digital.govt.nz methodology.

#### AI Guidance Perspective

**Relevant AI Guidance Principles**:

AI guidance does not contain the Plain Language Act text or implementation guidance. However, I can observe principles about clarity:

1. **Clear Communication**: AI guidance emphasizes clarity throughout:
   - "Openness: Clearly communicating the purpose and capabilities of an AI system." ([DocRef](https://docref.digital.govt.nz/nz/generative-ai-guidance-gcdo/glossary-of-ai-terms/2025/en/#part13-subpart1-para1-1))
   - "Check the quality of results... This helps make sure the content generated is accurate." ([DocRef](https://docref.digital.govt.nz/nz/generative-ai-guidance-gcdo/misinformation-hallucinations-and-genai/2025/en/#part3-subpart2-section2))

2. **Appropriate to Audience**: AI guidance tailored to public servants:
   - "AI Fundamentals for Public Servants: Opportunities, Risks and Strategies" ([DocRef](https://docref.digital.govt.nz/nz/generative-ai-guidance-gcdo/skills-capabilities-and-genai/2025/en/#part3-para1-1))
   - Demonstrates audience-appropriate technical level

3. **Accessibility Requirement**: "Accessibility means designing things to work for disabled people. The New Zealand Government has legal and ethical obligations to create accessible information and services, both for the public and for public servants." ([DocRef](https://docref.digital.govt.nz/nz/generative-ai-guidance-gcdo/accessibility-and-genai/2025/en/#h1-subtitle))

   **Implication**: Plain language is part of accessibility - legal obligation extends to all government information.

#### Validation/Challenge

**VALIDATION (Legal Obligation)**: AI guidance confirms "legal and ethical obligations to create accessible information" - Plain Language Act is part of this obligation framework.

**VALIDATION (By Example)**: AI guidance models plain language practices - clear sentences, active voice, direct address - demonstrating government standard for technical guidance.

**GOVERNMENT-WIDE REQUIREMENT**: The fact that AI guidance (technical, specialized) uses plain language principles confirms these apply to all government documentation including identification standards.

#### Specific Recommendations from AI Guidance

**Legal Obligation**: "The New Zealand Government has legal and ethical obligations to create accessible information and services" ([DocRef](https://docref.digital.govt.nz/nz/generative-ai-guidance-gcdo/accessibility-and-genai/2025/en/#h1-subtitle))

**Recommendation for Identification Standards**: Plain Language Act compliance is mandatory legal requirement. Active voice, clear structure, appropriate language are not optional enhancements - they're legal obligations.

**Verdict on Finding 6**: **VALIDATED (Legal Basis)** - AI guidance confirms legal obligations for accessible, clear government information. Plain Language Act compliance is mandatory, not discretionary.

### Finding 7: Biometric Privacy Requirements (Gap)

**Stage 3/4 Finding Summary**:
Stage 3 identified Privacy Commissioner's Biometric Processing Privacy Code 2025 as mandatory law (effective 3 Nov 2025). Stage 4 noted identification standards have technical controls (AA9, AA10) but lack privacy requirements (necessity, consent, retention, purpose limitation).

**Stage 4 Recommendation**:
Augment Authentication Assurance guidance with biometric privacy requirements section. Extract 5 key Privacy Code rules, integrate with existing AA9/AA10 guidance.

#### AI Guidance Perspective

**Relevant AI Guidance Principles**:

AI guidance extensively addresses privacy requirements for AI systems - highly relevant parallel to biometric systems:

1. **Privacy as Core Principle**: "compliance with data-protection rules and legislation" ([DocRef](https://docref.digital.govt.nz/nz/generative-ai-guidance-gcdo/privacy-and-genai/2025/en/#part2-para1-1))

2. **Data Ethics Framework**: "Transparency: Being open about how the data is collected, used, and shared." ([DocRef](https://docref.digital.govt.nz/nz/generative-ai-guidance-gcdo/glossary-of-ai-terms/2025/en/#part3-subpart2-para2-2))

3. **Sensitive Data Protection**: "Use IN-CONFIDENCE for all personal information provided by users through online sites or services. The Privacy Act requires agencies to take reasonable steps to protect that information from unauthorised disclosure or access" ([DocRef](https://docref.digital.govt.nz/nz/generative-ai-guidance-gcdo/classify-information/2019/en/#part2-det2))

4. **Privacy-by-Design**: AI guidance integrates privacy considerations throughout system development guidance - not separate afterthought.

#### Validation/Challenge

**STRONG VALIDATION**: AI guidance **models the integration of privacy requirements with technical requirements**. AI systems guidance includes both technical capabilities AND privacy/data protection requirements together.

**PARALLEL STRUCTURE**: Just as AI guidance integrates privacy with AI system requirements, identification standards should integrate biometric privacy with authentication assurance requirements.

**GOVERNMENT STANDARD**: The integration of privacy in AI guidance demonstrates government expectation that privacy requirements appear alongside technical requirements, not separately.

#### Specific Recommendations from AI Guidance

**Integration Pattern**: AI guidance integrates privacy requirements throughout technical guidance rather than separating them.

**Compliance Requirement**: "compliance with data-protection rules and legislation" ([DocRef](https://docref.digital.govt.nz/nz/generative-ai-guidance-gcdo/privacy-and-genai/2025/en/#part2-para1-1))

**Recommendation for Identification Standards**: Follow AI guidance pattern - integrate Privacy Code requirements directly into Authentication Assurance guidance where biometric authentication is discussed. Don't separate privacy from technical controls.

**Verdict on Finding 7**: **STRONGLY VALIDATED** - AI guidance models privacy-technical integration, validating Stage 4's recommendation to augment identification standards with biometric privacy requirements.

## Where Identification Standards Already Align with AI Guidance

### Alignment 1: Technical Rigor and Specificity

**AI Guidance Approach**: Specific, actionable requirements:
- "Ensure all GenAI systems used by your agency are certified and accredited before they're made available to users" ([DocRef](https://docref.digital.govt.nz/nz/generative-ai-guidance-gcdo/security-and-genai/2025/en/#part2-para1-3))
- "New software versions should be evaluated and tested before they're rolled out" ([DocRef](https://docref.digital.govt.nz/nz/generative-ai-guidance-gcdo/security-and-genai/2025/en/#part1-subpart1-det1-para2))

**Identification Standards Alignment**: Standards provide specific, measurable controls (AA9.04, AA10.01, etc.) with clear technical requirements. This rigor should be preserved in Phase 2 restructuring.

### Alignment 2: Risk-Based Approach

**AI Guidance Approach**: "A security risk assessment should consider these factors to determine if the application is the right fit for your organisation." ([DocRef](https://docref.digital.govt.nz/nz/generative-ai-guidance-gcdo/security-and-genai/2025/en/#part1-para2))

**Identification Standards Alignment**: Authentication Assurance Standard uses levels of assurance (LoA1-LoA4) based on risk assessment. Risk-based approach is sound and aligns with government guidance patterns.

### Alignment 3: Example Scenarios

**AI Guidance Approach**: Includes example scenarios to illustrate principles in practice:
- "You use GenAI to summarise a long document..." ([DocRef](https://docref.digital.govt.nz/nz/generative-ai-guidance-gcdo/accountability-responsibility-and-genai/2025/en/#part5-ex1))
- "Your mother, who has a vision impairment... independently resolved her issues" ([DocRef](https://docref.digital.govt.nz/nz/generative-ai-guidance-gcdo/accessibility-and-genai/2025/en/#part4-ex1))

**Identification Standards Alignment**: Implementation guides include examples (driver's licence federation scenario, biometric authentication examples). Examples are valuable and should be preserved/expanded in Phase 2.

### Alignment 4: Reference to External Standards

**AI Guidance Approach**: References external standards (NZISM, OECD, PEAT frameworks) with specific links:
- "System Certification and Accreditation — NZISM Chapter 4" ([DocRef](https://docref.digital.govt.nz/nz/generative-ai-guidance-gcdo/security-and-genai/2025/en/#part4-para1-1))

**Identification Standards Alignment**: Standards reference ISO, NIST, Privacy Act appropriately. Cross-referencing is sound practice.

## Gaps Between Current Standards and Best Practices (AI Guidance Evidence)

### Gap 1: Passive Voice in Guidance (Not Standards)

**AI Guidance Evidence**: Government technical guidance uses active voice throughout, demonstrating feasibility and appropriateness.

**Identification Standards Gap**: Guidance materials use passive voice extensively despite being modifiable (unlike core standards).

**Severity**: HIGH - Affects clarity and usability; contradicts government writing standards modeled in AI guidance.

### Gap 2: Detail Expanders Hide Essential Content

**AI Guidance Evidence**: Accessibility principles require visible, scannable content. AI guidance documents don't use detail expanders.

**Identification Standards Gap**: Conforming and Assessing Risk guidance hide threshold information in detail expanders.

**Severity**: HIGH - Violates accessibility principles; users may miss critical information.

### Gap 3: Privacy-Technical Separation for Biometrics

**AI Guidance Evidence**: AI guidance integrates privacy requirements with technical requirements throughout.

**Identification Standards Gap**: Authentication Assurance has technical biometric controls but no privacy requirements despite mandatory Privacy Code.

**Severity**: HIGH - Users implementing biometric authentication could violate Privacy Code despite following standards.

### Gap 4: Conformance Process Visibility

**AI Guidance Evidence**: AI governance guidance emphasizes transparency about processes, frameworks, and compliance mechanisms.

**Identification Standards Gap**: Conformance process semantically isolated, separate document, not prominent despite being primary user goal.

**Severity**: CRITICAL - Misalignment between user goals and content organization.

### Gap 5: User Pathway Design

**AI Guidance Evidence**: Limited - AI guidance recognizes diverse stakeholders but doesn't model strong pathway design.

**Identification Standards Gap**: 30 documents without clear entry points or role-based pathways.

**Severity**: MEDIUM-HIGH - Creates navigation burden but AI guidance doesn't strongly validate solution.

## Extracted AI Guidance Recommendations

### Content Design Principles (From AI Guidance Modeling)

1. **Use Active Voice**: AI guidance consistently uses active voice for instructions and requirements. Government technical guidance can and should use active voice.

2. **Direct Address**: Use "you" and "your" when addressing the reader. AI guidance models this throughout.

3. **Clear, Specific Instructions**: "Ensure all GenAI systems... are certified" not "Systems should be certified" - active, specific, direct.

4. **Visible Content**: Don't hide essential information. All content should be scannable and accessible.

### Privacy Integration Principles

5. **Integrate Privacy with Technical Requirements**: Don't separate privacy requirements into different documents. AI guidance integrates privacy throughout technical guidance.

6. **Compliance as Core Concern**: "compliance with data-protection rules and legislation" is not optional enhancement - it's core requirement alongside technical requirements.

7. **Transparency About Data**: "Being open about how the data is collected, used, and shared" - applies to biometric data in identification systems.

### Accessibility Principles

8. **Legal and Ethical Obligations**: "The New Zealand Government has legal and ethical obligations to create accessible information and services" - applies to all government documentation including identification standards.

9. **Design for All Users**: "considering the needs of all potential users from the outset" - identification standards serve diverse practitioners with different needs and technical backgrounds.

10. **Assistive Technology Compatibility**: Content structure must work with screen readers, which requires visible content hierarchy (not hidden detail expanders).

### Transparency and Clarity Principles

11. **Openness**: "Clearly communicating the purpose and capabilities" - identification standards should clearly communicate scope, applicability, and conformance requirements upfront.

12. **Clear Classification**: "Clear classification marking of information allows for easy filtering" - clear organization and signposting helps users navigate.

13. **Verification and Quality**: "Check the quality... This helps make sure the content... is accurate" - verification principles apply to documentation quality as well as AI outputs.

### Process and Governance Principles

14. **Publish Process Information**: "publish information about... what stage the project is at, the intent of use" - analogous to surfacing conformance workflow stages and process information.

15. **Accountability Framework**: "This should include the application of relevant regulatory and governance frameworks, reporting, auditing and/or independent reviews" - conformance assessment is the auditing/review mechanism for identification standards.

## Supporting Evidence (Key Excerpts)

### Evidence of Active Voice as Government Standard

**AI Guidance Active Voice Examples**:
- "Engage disabled people as stakeholders from conception of the GenAI solution to its deployment." ([DocRef](https://docref.digital.govt.nz/nz/generative-ai-guidance-gcdo/accessibility-and-genai/2025/en/#part3-subpart1-para1))
- "Consider disabled people's experiences in training data, models and outputs to remove bias." ([DocRef](https://docref.digital.govt.nz/nz/generative-ai-guidance-gcdo/accessibility-and-genai/2025/en/#part3-subpart2-para1))
- "Check the quality of results produced by GenAI systems with trusted sources." ([DocRef](https://docref.digital.govt.nz/nz/generative-ai-guidance-gcdo/misinformation-hallucinations-and-genai/2025/en/#part3-subpart2-section2))
- "Ensure all GenAI systems used by your agency are certified and accredited before they're made available to users" ([DocRef](https://docref.digital.govt.nz/nz/generative-ai-guidance-gcdo/security-and-genai/2025/en/#part2-para1-3))

**Interpretation**: GCDO's AI guidance consistently uses active voice for government technical guidance. This models government standard practice and validates that technical guidance can use active voice effectively.

### Evidence of Privacy-Technical Integration

**AI Guidance Integration Pattern**:
- Security guidance combines technical requirements (certification, accreditation) with security controls and monitoring requirements ([DocRef](https://docref.digital.govt.nz/nz/generative-ai-guidance-gcdo/security-and-genai/2025/en/#part2-para1-3))
- Privacy requirements appear throughout AI system guidance, not isolated in separate privacy document
- "compliance with data-protection rules and legislation" presented as core requirement ([DocRef](https://docref.digital.govt.nz/nz/generative-ai-guidance-gcdo/privacy-and-genai/2025/en/#part2-para1-1))

**Interpretation**: Government practice integrates privacy with technical requirements. Identification standards should integrate Privacy Code requirements with Authentication Assurance guidance, not keep them separate.

### Evidence of Accessibility as Legal Obligation

**AI Guidance Statement**: "Accessibility means designing things to work for disabled people. The New Zealand Government has legal and ethical obligations to create accessible information and services, both for the public and for public servants." ([DocRef](https://docref.digital.govt.nz/nz/generative-ai-guidance-gcdo/accessibility-and-genai/2025/en/#h1-subtitle))

**Interpretation**: Accessibility is legal requirement, not optional. This includes:
- Content structure (visible, scannable, not hidden in detail expanders)
- Clear language (active voice, direct address, plain language)
- Appropriate organization (user-centered, findable)

These are legal obligations under broader accessibility framework (includes but not limited to Plain Language Act).

### Evidence of Transparency as Core Principle

**AI Guidance Transparency Examples**:
- "Openness: Clearly communicating the purpose and capabilities of an AI system. This includes explaining what the system is designed to do and any limitations it may have." ([DocRef](https://docref.digital.govt.nz/nz/generative-ai-guidance-gcdo/glossary-of-ai-terms/2025/en/#part13-subpart1-para1-1))
- "Agencies should publish information about their development and use of AI... This will help maintain transparency and trust" ([DocRef](https://docref.digital.govt.nz/nz/generative-ai-guidance-gcdo/governance-and-genai-in-the-public-service/2025/en/#part2-det5))
- "We strongly recommend publishing an up-to-date register" ([DocRef](https://docref.digital.govt.nz/nz/generative-ai-guidance-gcdo/transparency-and-genai/2025/en/#part2-subpart3))

**Interpretation**: Transparency about systems, processes, and requirements is government standard. Identification standards should be transparent about conformance process, assessment requirements, and how to achieve compliance.

## AI-Assisted Usage Considerations

### How AI Assistants Might Be Used with Identification Standards

**Potential Use Cases**:
1. **Question-Answering**: "What are the requirements for LoA3 authentication?" - AI assistant retrieves relevant controls
2. **Conformance Guidance**: "How do I demonstrate compliance with Binding Assurance Standard?" - AI assistant guides through process
3. **Risk Assessment**: "What authentication factors are appropriate for medium-risk application?" - AI assistant helps assess
4. **Cross-Reference**: "Which Privacy Code rules apply to biometric authentication?" - AI assistant connects related requirements

### Structural Improvements to Support AI-Assisted Usage

**What Makes Standards "AI-Friendly"**:

1. **Clear, Semantic Structure**:
   - Explicit headings describing content
   - Consistent terminology throughout
   - Well-defined relationships between concepts
   - Visible hierarchy (not hidden in detail expanders)

2. **Connected Concepts**:
   - Cross-references between related requirements
   - Integrated standards-guidance (not fragmented across documents)
   - Clear relationships (prerequisite, related, example of)
   - Semantic links via terminology

3. **Direct Language**:
   - Active voice: "You must perform authentication" - clearer for AI parsing than "Authentication is performed"
   - Explicit actors: "The credential provider validates..." - AI can understand roles and responsibilities
   - Specific requirements: "LoA3 requires multi-factor authentication" - AI can extract precise requirements

4. **Comprehensive Coverage**:
   - Privacy requirements alongside technical requirements (biometric example)
   - All related information in context (not requiring navigation to separate documents)
   - Examples and scenarios illustrating requirements

**What Hinders AI-Assisted Usage**:

1. **Content Fragmentation**: Standards spread across 8 documents (4 standards + 4 guides) makes it harder for AI to provide comprehensive answers
2. **Hidden Content**: Detail expanders hide information from AI extraction
3. **Passive Voice**: Obscures actor-action relationships that AI uses to understand requirements
4. **Semantic Isolation**: Conformance process with 0 semantic neighbors means AI won't connect conformance to related content
5. **Inconsistent Terminology**: Makes it harder for AI to map concepts across documents

### AI Guidance Recommendations

**AI Guidance Principle - Information Quality**: "Check the quality of results produced by GenAI systems with trusted sources. This helps make sure the content generated is accurate." ([DocRef](https://docref.digital.govt.nz/nz/generative-ai-guidance-gcdo/misinformation-hallucinations-and-genai/2025/en/#part3-subpart2-section2))

**Implication for Identification Standards**: Well-structured, clearly written standards enable more accurate AI-assisted guidance. Poor structure (fragmented, hidden content, unclear language) leads to incomplete or incorrect AI responses.

**AI Guidance Principle - Verification**: "Evaluate the references and citations provided in the system and check if the sources provided are legitimate and appropriate. Cross-check the information with credible sources and experts or relevant communities." ([DocRef](https://docref.digital.govt.nz/nz/generative-ai-guidance-gcdo/misinformation-hallucinations-and-genai/2025/en/#part3-subpart2-section3-para1-ex1-exl3))

**Implication for Identification Standards**: DocRef citations enable verification of AI-generated guidance against source documents. Phase 2 must maintain comprehensive DocRef citation coverage.

## Prioritized AI Guidance Recommendations

### Critical (Must Address) - Validated by AI Guidance

**1. Accessibility as Legal Requirement** (Privacy/Accessibility Guidance)
- **AI Guidance**: "legal and ethical obligations to create accessible information"
- **Application**: Plain language, active voice, visible content are legal obligations
- **Priority**: CRITICAL - mandatory compliance
- **Phase 2 Action**: Apply throughout content creation

**2. Privacy-Technical Integration** (Privacy/AI Systems Guidance)
- **AI Guidance**: Models integration of privacy with technical requirements
- **Application**: Augment biometric authentication guidance with Privacy Code requirements
- **Priority**: CRITICAL - mandatory Privacy Code compliance from 3 Nov 2025
- **Phase 2 Action**: Add biometric privacy section to Authentication Assurance guidance

**3. Active Voice as Government Standard** (Modeled Throughout AI Guidance)
- **AI Guidance**: Consistently uses active voice, direct address, clear instructions
- **Application**: Systematic active voice conversion in all guidance materials
- **Priority**: CRITICAL - government writing standard
- **Phase 2 Action**: Rewrite guidance using active voice and "you/your"

**4. Eliminate Hidden Content** (Accessibility Principles)
- **AI Guidance**: Accessibility requires visible, scannable content for all users
- **Application**: Remove all detail expanders, surface all content with clear hierarchy
- **Priority**: CRITICAL - accessibility compliance
- **Phase 2 Action**: Delete all detail expander syntax, restructure with headings

### Significant (Should Address) - Supported by AI Guidance

**5. Transparency About Processes** (Governance/Transparency Guidance)
- **AI Guidance**: Emphasizes transparency about processes, frameworks, compliance mechanisms
- **Application**: Make conformance process prominent and transparent
- **Priority**: HIGH - aligns with transparency principles
- **Phase 2 Action**: Conformance-centered organization, surface process information early

**6. Integrated Structure** (Modeled in AI Guidance)
- **AI Guidance**: Integrates principles and implementation in single documents
- **Application**: Integrate standards and guidance to reduce navigation burden
- **Priority**: HIGH - improves usability
- **Phase 2 Action**: Combine standards and guides with visual distinction

**7. Clear Purpose Communication** (Transparency/Openness Principles)
- **AI Guidance**: "Clearly communicating the purpose and capabilities... limitations"
- **Application**: Surface threshold information upfront - scope, applicability, prerequisites
- **Priority**: HIGH - helps users make informed decisions
- **Phase 2 Action**: "Before You Start" sections with threshold considerations

### Recommended (Could Address) - Limited AI Guidance Validation

**8. User Pathway Design** (Diverse Stakeholder Recognition)
- **AI Guidance**: Recognizes diverse stakeholders have different needs
- **Application**: Create role-based entry points and pathways
- **Priority**: MEDIUM-HIGH - improves navigation but AI guidance doesn't model this strongly
- **Phase 2 Action**: "Start Here" section with role-based pathways

**9. Example Scenarios** (Modeled in AI Guidance)
- **AI Guidance**: Includes example scenarios illustrating principles in practice
- **Application**: Expand examples in implementation guidance
- **Priority**: MEDIUM - improves comprehension
- **Phase 2 Action**: Add scenarios to complex requirement sections

## Validation Summary

### Stage 4 Findings VALIDATED by AI Guidance

1. **Active Voice (Finding 2)**: **STRONGLY VALIDATED** - AI guidance models active voice usage throughout government technical guidance
2. **Content Hiding/Detail Expanders (Finding 3)**: **VALIDATED** - Accessibility and transparency principles require visible content
3. **Privacy Integration (Finding 7)**: **STRONGLY VALIDATED** - AI guidance models privacy-technical integration pattern
4. **Plain Language Act Compliance (Finding 6)**: **VALIDATED** - Confirmed as legal obligation for government information
5. **Standards-Guidance Integration (Finding 4)**: **SUPPORTED** - AI guidance models integrated structure

### Stage 4 Findings with LIMITED AI Guidance Validation

6. **Conformance-Centered Organization (Finding 1)**: **SUPPORTED BY ANALOGY** - AI governance principles about transparency and accountability align with conformance-centered approach, but no direct validation. **Finding stands on user research evidence**.
7. **User Journey/Entry Points (Finding 5)**: **PARTIALLY SUPPORTED** - AI guidance recognizes diverse stakeholders but doesn't model strong pathway design. **Finding stands on usability principles**.

### Stage 4 Findings CHALLENGED by AI Guidance

**NONE** - No Stage 4 findings are contradicted by AI guidance principles.

### New Insights from AI Guidance

**1. Legal Framework for Accessibility**: AI guidance explicitly states government has "legal and ethical obligations" for accessible information - strengthens legal basis for Stage 4 recommendations (active voice, plain language, visible content).

**2. Privacy Integration is Government Standard**: AI guidance models integrated privacy-technical approach - strengthens Stage 4's biometric privacy recommendation by showing this is government standard practice.

**3. Active Voice is Government Technical Guidance Standard**: AI guidance demonstrates active voice works for highly technical government guidance - removes any concern that "standards must be passive."

**4. Transparency as Core Principle**: AI guidance emphasizes transparency about processes and frameworks - additional support for conformance process visibility beyond user research evidence.

## Limitation Acknowledgment

**Critical Limitation**: This evaluation used the generative-ai-guidance-gcdo MCP server, which contains **AI/GenAI-specific guidance**, NOT the digital.govt.nz content design guidance that Stage 3 identified as directly addressing Stage 2 issues.

**What This Means**:
- Validation is based on **principles extracted from AI guidance** and **examples modeled by AI guidance**
- Validation is **indirect/analogous** for some findings (conformance organization, user pathways)
- **Digital.govt.nz content design guidance** (Plain Language, Tone and Voice, Page Structure, Findable Content) would provide **more direct validation** but is not in this MCP server

**Strength of Stage 4 Findings Despite Limitation**:
Stage 4 findings are grounded in:
1. **User research**: Tom's direct observations and annotations (15+ on passive voice, 12+ on detail expanders, conformance is "the whole point")
2. **Data evidence**: Semantic isolation, navigation burden, fragmentation metrics
3. **Digital.govt.nz guidance**: Stage 3 identified relevant guidance even though not in this MCP server
4. **Plain Language Act 2022**: Mandatory legal requirement (Stage 3)

AI guidance evaluation **adds additional support** through government practice modeling and accessibility/transparency principles, but Stage 4 findings are **independently supported** by user research and legal requirements.

## Decisions Made

### Decision 1: Evaluate Despite MCP Server Limitation

**Decision**: Proceed with evaluation using AI guidance MCP server despite it not containing digital.govt.nz content design guidance.

**Rationale**:
- AI guidance **models government writing practices** (active voice, plain language, integrated structure)
- AI guidance **confirms legal obligations** (accessibility, transparency)
- **Analogous principles** (transparency, accessibility, privacy integration) can be extracted
- Stage 4 findings already well-supported by user research and legal requirements

**Alternative Considered**: Abort evaluation and request digital.govt.nz content design guidance in MCP server format - rejected as impractical for timeline.

### Decision 2: Focus on Modeling and Principles

**Decision**: Evaluate by examining how AI guidance models best practices and extracting applicable principles, rather than seeking direct guidance on standards documentation.

**Rationale**:
- AI guidance demonstrates **government standard practices** even if not explicitly prescribing them
- AI guidance **models active voice, integrated structure, privacy integration** - can learn from examples
- Principles about **transparency, accessibility, clarity** apply across documentation types

### Decision 3: Acknowledge Validation Strength Levels

**Decision**: Distinguish between "strongly validated" (direct evidence), "supported" (analogous principles), and "limited validation" (no strong evidence either way).

**Rationale**:
- Transparent about evaluation limitations
- Stronger validation for findings with direct evidence (active voice, privacy integration)
- Weaker validation for findings without direct parallels (user pathways)
- Preserves integrity of Stage 4 findings that stand on user research regardless of AI guidance

### Decision 4: Extract Government Practice Patterns

**Decision**: Treat AI guidance as **example of government technical guidance** and extract patterns about how government writes technical documentation.

**Rationale**:
- AI guidance is government technical guidance for specialized domain (AI)
- Identification standards are government technical guidance for specialized domain (identification)
- Parallel contexts suggest parallel best practices apply
- Patterns in AI guidance reflect government standards

## Questions and Uncertainties

### Question 1: Access to Digital.govt.nz Content Design Guidance

**Question**: Would evaluation be stronger with digital.govt.nz content design guidance in MCP server format?

**Answer**: YES - Stage 3 identified that guidance as directly addressing Stage 2 issues. It would provide direct validation for active voice, page structure, findability, plain language recommendations.

**Impact**: Current evaluation relies on indirect validation through AI guidance modeling. Digital.govt.nz guidance would provide explicit prescriptive validation.

**Resolution**: Stage 4 findings already supported by Stage 3 research and legal requirements. AI guidance adds supporting evidence but is not primary validation source.

### Question 2: Plain Language Act Implementation Details

**Question**: What specific Plain Language Act requirements apply to technical standards vs. general public information?

**Uncertainty**: AI guidance confirms "legal obligations" for accessible information but doesn't detail Plain Language Act implementation requirements for different content types.

**Resolution**: Stage 3 documented Plain Language Act requirements. Phase 2 should apply plain language principles (active voice, clear structure, appropriate language) while maintaining technical precision where necessary.

### Question 3: Conformance-Centered Organization Validation

**Question**: Does lack of direct AI guidance validation for conformance-centered organization weaken this recommendation?

**Answer**: NO - This finding is grounded in:
- User research (Tom: "the whole point")
- Data evidence (semantic isolation, 0 neighbors)
- Practical user needs (practitioners seeking conformance)

AI guidance limitation (not organized around conformance itself) doesn't invalidate user research findings.

### Question 4: Terminology Authority

**Question**: Does AI guidance provide insights on terminology authority approach (Stage 4 Theme 6)?

**Answer**: Limited - AI guidance includes glossary defining AI terms, but doesn't address authority question. Stage 4 terminology authority recommendation stands on DIA's role as standards maker.

## Next Steps for Stage 6

Based on Stage 5 AI guidance evaluation, Stage 6 (Final Recommendations and Structure Proposal) should:

### 1. Confirm Critical Priorities

AI guidance evaluation **strongly validates** these Stage 4 priorities:
- **Active voice systematic conversion** - government standard practice
- **Privacy-technical integration for biometrics** - government standard pattern
- **Eliminate detail expanders** - accessibility requirement
- **Plain Language Act compliance** - legal obligation

Stage 6 should confirm these as **non-negotiable critical priorities** for Phase 2.

### 2. Strengthen Legal Basis

AI guidance confirmed **legal and ethical obligations** for accessible information:
- Plain Language Act compliance is mandatory
- Accessibility requirements include content structure (visible, scannable)
- Privacy requirements must be integrated with technical requirements

Stage 6 should emphasize **legal compliance** as driver for Phase 2 changes, not just usability improvements.

### 3. Use AI Guidance as Precedent

AI guidance provides **government precedent** for Phase 2 approach:
- Active voice works for technical government guidance
- Privacy-technical integration is government practice
- Integrated structure (not fragmented) is modeled in government guidance

Stage 6 should reference AI guidance as **precedent** showing Phase 2 approach is consistent with government standards.

### 4. Address Limitation

Stage 6 should acknowledge:
- Evaluation used AI guidance (not digital.govt.nz content design guidance)
- Some validation is indirect/analogous
- Stage 4 findings remain well-supported by user research and legal requirements
- Digital.govt.nz guidance (identified in Stage 3) should be **primary methodology reference** for Phase 2

### 5. Develop Content Design Framework

Stage 6 should specify:
- **Active voice conversion methodology** (using digital.govt.nz Tone and Voice guidance from Stage 3)
- **Plain language principles** (using digital.govt.nz Plain Language guidance from Stage 3)
- **Privacy integration approach** (following AI guidance model)
- **Accessibility requirements** (legal obligations confirmed by AI guidance)
- **Structure principles** (integrated, visible, scannable)

### 6. Prepare for Phase 2 Implementation

Stage 6 should detail:
- **Which Stage 4 recommendations are mandatory** (legal compliance)
- **Which are high-priority improvements** (strong AI guidance validation)
- **Which need stakeholder validation** (user pathway design, terminology authority)
- **How to apply AI guidance principles** in Phase 2 content creation
- **Verification criteria** for Phase 2 Stage 12 (active voice %, visible content, privacy integration)

### 7. Document Government Standards Framework

Stage 6 should synthesize:
- **Legal requirements**: Plain Language Act, Privacy Act, Accessibility obligations
- **Government standards**: Active voice, plain language, integrated structure (evidenced by AI guidance)
- **Content design methodology**: Digital.govt.nz guidance (from Stage 3)
- **Precedents**: AI guidance models government technical documentation best practices

This provides comprehensive framework justifying Phase 2 approach.

## Conclusion

**Overall Validation Verdict**: AI guidance principles **STRONGLY SUPPORT** Stage 4 synthesis priorities, with some limitations due to MCP server scope.

**Top 3 AI Guidance Validations**:
1. **Active voice is government standard** - AI guidance models this throughout technical guidance
2. **Privacy-technical integration required** - AI guidance demonstrates government practice of integrating privacy with technical requirements
3. **Accessibility is legal obligation** - AI guidance confirms legal requirements for accessible, clear, well-organized government information

**Key Challenge**: AI guidance does not directly validate conformance-centered organization (Finding 1, Stage 4's top priority) because AI guidance itself is not conformance-focused. However, this finding stands on **user research evidence** (Tom's observations, semantic isolation data) regardless of AI guidance validation.

**Most Important New Insight**: Legal framework for accessibility (Plain Language Act, government obligations) provides **legal mandate** for Phase 2 changes. This is not optional enhancement - it's mandatory compliance with government legal obligations.

**Evaluation Confidence**:
- **HIGH** for findings with direct AI guidance modeling (active voice, privacy integration)
- **MEDIUM** for findings with analogous principles (conformance organization, user pathways)
- **INDEPENDENT** for findings grounded in user research and legal requirements regardless of AI guidance

**Phase 2 Readiness**: Stage 5 evaluation confirms Stage 4 synthesis priorities are well-founded and should proceed to implementation. AI guidance provides additional government precedent and legal basis strengthening the case for Phase 2 changes.