Structured Inputs, Outputs, and Multi-Stage Workflows

2. Learning Outcomes

By the end of this lesson, you should be able to:

1. Define structured input, structured output, workflow, and traceability.
2. Design structured inputs that reduce ambiguity.
3. Design structured, inspectable outputs.
4. Connect stages into a multi-stage workflow.
5. Establish verification and review points.
6. Define provisional outputs and distinguish them from final outputs.
7. Identify exception, escalation, and stop conditions.
8. Preserve traceability from source to output.
9. Define permission boundaries for outputs.
10. Apply the Produce, Review, and Permit element of SCOPE.

3. Before You Begin

• Assigned reading: Chapter 5: Advanced Prompting and Context Engineering (sections on structured inputs, structured outputs, workflow design, and permission boundaries).
• Prerequisite knowledge: You should be comfortable with PCAFE-V, context packages (Lesson 5.1), and task decomposition and prompt sequences (Lesson 5.2).
• Optional AI tool: You may use an institutionally approved generative AI system or the supplied-output alternatives provided in each activity. No paid service is required.
• Privacy reminder: Do not enter personal, confidential, proprietary, privileged, or security-sensitive information into any public or unapproved AI system.
• Academic integrity: All work you submit must be your own. Use AI as a tool, not as a substitute for your own critical thinking.
• Estimated preparation time: 15–20 minutes to review the relevant sections of Chapter 5.

4. Retrieval Practice: Concepts from Lessons 5.1 and 5.2

Before we begin, let's activate your knowledge from the previous lessons.

5. Opening Scenario: "The Polished but Unauditable Summary"

6. Structured Inputs

Structured inputs transform a messy collection of information into a clear, organised format that the AI can process more reliably.

Why structure inputs?

• Reduces ambiguity: Labelled sections tell the AI what each piece of information is.
• Supports consistency: A consistent structure helps the AI produce consistent outputs.
• Improves traceability: Source records can be linked to specific claims.
• Makes gaps visible: Missing information is obvious when a field is empty.
• Facilitates verification: Reviewers can check each section against sources.

Types of structured inputs

=== TASK INFORMATION ===
Purpose: Summarise the new waste-management guidelines for the public.
Audience: Community members (grade-8 reading level).
Format: 200-word summary with bullet points.

=== SOURCE RECORDS ===
Source 1 – Title: Council Meeting Minutes, 2026-06-15
Authority: City Council
Date: 2026-06-15
Content: [Extracted key points about the new guidelines]

Source 2 – Title: Policy Brief – Waste Management 2026
Authority: City Environmental Department
Date: 2026-06-20
Content: [Extracted key points]

Source 3 – Title: Email from Director – Implementation Timeline
Authority: Director of Environmental Services
Date: 2026-06-22
Content: [Extracted key points]

=== CONSTRAINTS ===
– Do not exceed 200 words.
– Do not use technical jargon.
– Do not include information not in the sources.

=== REQUIRED OUTPUT ===
A 200-word plain-text summary with bullet points.

7. Structured Outputs

Structured outputs transform a raw AI response into an auditable artefact that you can review, verify, and use with confidence.

Why structure outputs?

• Improves reviewability: Reviewers can check each claim against its source.
• Makes missing information visible: Empty fields or explicit "missing" notes show gaps.
• Supports traceability: Source links show where each claim came from.
• Reduces reliance on memory: All information is captured in a structured format.
• Facilitates hand-offs: Structured outputs are easier to pass to the next stage.

Types of structured outputs

=== SUMMARY ===
The city will introduce new waste-management guidelines effective 2026-09-01.
Key changes include:
– Bi-weekly collection for organic waste.
– New sorting requirements for recyclables.
– A fee structure for excess waste.

=== CLAIM INVENTORY ===
Claim 1: New guidelines effective 2026-09-01.
Source: Council Meeting Minutes, 2026-06-15, Section 3.1
Confidence: High

Claim 2: Bi-weekly collection for organic waste.
Source: Policy Brief – Waste Management 2026, page 4
Confidence: High

Claim 3: New sorting requirements for recyclables.
Source: Policy Brief – Waste Management 2026, page 5
Confidence: High

Claim 4: Fee structure for excess waste.
Source: Email from Director, 2026-06-22
Confidence: Medium – details to be confirmed

=== UNCERTAINTY LOG ===
– Fee structure details: Source is an email, not an official policy.
– Implementation timeline: Official policy does not specify transitional provisions.

=== REVIEWER NOTES ===
– Verify Claim 4 against the official policy document.
– Confirm the effective date with the Environmental Department.

8. Traceability

Traceability is the foundation of accountability in AI-supported work. If you cannot trace a claim back to its source, you cannot verify it or defend it.

Why traceability matters

• Verification: You can check each claim against its source.
• Accountability: You can show where each piece of information came from.
• Error correction: You can identify where an error entered and correct it.
• Auditability: You can demonstrate that you followed a responsible process.
• Reusability: You can reuse traceable workflows with confidence.

How to preserve traceability

• Label every source with title, authority, date, and version.
• Link claims to sources in the output (e.g., using source IDs).
• Record stage outputs and include them in the next stage's context.
• Document prompts and versions for each stage.
• Record human decisions—what was reviewed, what was changed, and why.
• Use structured outputs that include source fields.

9. Workflow Assembly

A workflow is more than a prompt sequence. It includes the human decisions, verification points, and permission boundaries that govern the entire process.

Components of a complete workflow

1. Context preparation: Select sources, build a context package.
2. AI generation: Run stage-specific prompts.
3. Human review: Review outputs against criteria.
4. Verification: Check claims against sources.
5. Revision: Correct errors and improve outputs.
6. Approval: Obtain required approvals.
7. Permitted use: Use the output for its intended purpose.
8. Monitoring: Monitor for errors or drift.
9. Correction: Correct errors and document changes.

1. Context preparation (Human):
   – Select sources (official policy, supporting documents)
   – Build a context package with source hierarchy
   – Define constraints and requirements

2. Stage 1 – Source summarisation (AI):
   – Generate summaries of each source
   – Output: Source summaries with source IDs

3. Verification 1 (Human):
   – Check each summary against the source
   – Correct any errors

4. Stage 2 – Thematic synthesis (AI):
   – Synthesise summaries into themes
   – Output: Thematic synthesis with source references

5. Verification 2 (Human):
   – Check themes against summaries
   – Ensure all key points are covered

6. Stage 3 – Draft summary (AI):
   – Draft a 200-word summary for the public
   – Output: Draft summary

7. Verification 3 (Human):
   – Check the draft against the sources
   – Check readability and clarity

8. Revision (Human):
   – Revise the draft based on verification

9. Approval (Human):
   – Obtain approval from the communications director

10. Permitted use (Human):
    – Publish the final summary in the community newsletter

10. Verification and Review Gates

A gate is a point in the workflow where the process stops and a human must review, verify, or make a decision before proceeding.

Types of gates

Where to place gates

• After each stage that produces a significant output. Verify before moving to the next stage.
• Before any decision or action. Review before approving or publishing.
• When the task changes. Re-verify when scope, sources, or risks change.
• When errors are detected. Stop and correct before proceeding.

11. Permission Boundaries

Permission boundaries are essential for responsible AI use. They prevent outputs from being used inappropriately or without proper review.

Types of permissions

=== PERMISSION BOUNDARIES ===
– Draft summary: May be shared internally for review.
– Revised summary: May be shared with the communications team.
– Final summary: May be published in the community newsletter.
– The summary must not be used for legal advice.
– The summary must not be used for decision-making without
  additional review.

12. Provisional vs. Final

All AI-generated outputs should be treated as provisional until they have been reviewed and approved by a human.

Characteristics of provisional outputs

• Label clearly as "DRAFT" or "PROVISIONAL"
• Not used for decisions or actions
• Subject to verification and review
• May require revision
• Requires explicit approval to become final

Transitioning from provisional to final

1. Review: Human reviews the output against criteria.
2. Verify: Check claims against sources.
3. Revise: Correct errors and improve quality.
4. Approve: Obtain explicit approval from an authorised person.
5. Finalise: Label the output as "FINAL" and document the approval.

13. SCOPE: Produce, Review, and Permit Outputs

The SCOPE framework continues with the fourth element: P (Produce, Review, and Permit Outputs).

P: Produce, Review, and Permit Outputs

In this stage, you manage the production, review, and permission of outputs. You ensure that outputs are labelled as provisional, reviewed before use, and only permitted for appropriate purposes.

Guiding questions:

• What output is produced?
• How will it be reviewed?
• What may be used immediately?
• What requires approval?
• What must not be published, sent, modified, or acted upon automatically?

14. Process Walkthrough: From Structured Input to Reviewed Output

This walkthrough shows a complete process from designing a structured input to producing a reviewed and permitted output.

Step 1: Define the task and purpose

Task: Prepare a summary of a new provincial health guideline on nutrition labelling for a consumer advocacy group.

Step 2: Design the structured input

=== TASK INFORMATION ===
Purpose: Summarise the new nutrition labelling guideline for consumers.
Audience: Consumer advocacy group members (general public).
Format: 150-word summary with a claim inventory.

=== SOURCE RECORDS ===
Source 1 – Title: Provincial Nutrition Labelling Guideline 2026
Authority: Provincial Ministry of Health
Date: 2026-07-01
Content: [Extracted key points]

Source 2 – Title: Consumer Impact Assessment Report
Authority: Ministry of Health Research Division
Date: 2026-06-15
Content: [Extracted key points]

=== CONSTRAINTS ===
– Do not exceed 150 words.
– Use plain language (grade-8 reading level).
– Include a claim inventory with source references.

=== REQUIRED OUTPUT ===
A 150-word summary with a claim inventory linking each claim to its source.

Step 3: Run the AI-supported stage

Prompt: "Using the attached structured input, generate a 150-word summary of the nutrition labelling guideline for consumers. Include a claim inventory with source references. Output: A summary with a claim inventory."

Step 4: Produce a structured output

=== PROVISIONAL OUTPUT – DRAFT ===
Date: 2026-08-29
Status: DRAFT – NOT FOR DISTRIBUTION

=== SUMMARY ===
The new Provincial Nutrition Labelling Guideline requires food products
to display a simplified nutrition label on the front of packaging,
effective 2027-01-01. The label will highlight calories, sodium, sugar,
and saturated fat. Consumer research indicates that simplified labels
improve purchasing decisions.

=== CLAIM INVENTORY ===
Claim 1: New guideline requires front-of-package labels.
Source: Provincial Nutrition Labelling Guideline 2026, Section 2.1
Confidence: High

Claim 2: Labels will highlight calories, sodium, sugar, and saturated fat.
Source: Provincial Nutrition Labelling Guideline 2026, Section 3.2
Confidence: High

Claim 3: Effective date is 2027-01-01.
Source: Provincial Nutrition Labelling Guideline 2026, Section 9.1
Confidence: High

Claim 4: Simplified labels improve purchasing decisions.
Source: Consumer Impact Assessment Report 2026, page 4
Confidence: Medium – based on a single study

=== UNCERTAINTY LOG ===
– Claim 4 is based on a single study; may not be generalisable.
– The guideline does not specify enforcement mechanisms.

=== REVIEWER NOTES ===
– Verify Claim 4 against the original report.
– Consider adding a note about uncertainty in the summary.

Step 5: Review and verification

• Verification: Check each claim against the source. Claim 4 is verified against the report, but the uncertainty is noted.
• Review: Assess the summary for clarity, completeness, and accuracy. The summary is clear and accurate.
• Revision: Add a sentence about the uncertainty: "Early consumer research suggests simplified labels may improve purchasing decisions, though more research is needed."

Step 6: Permission and approval

• Approval: Obtain approval from the consumer advocacy group's director.
• Permissions: The final summary may be shared with members and published on the group's website.
• Prohibited uses: Not for legal advice or policy advocacy without additional review.

15. Worked Example: Complete Workflow for a Municipal Policy Summary

Context: A municipal policy analyst needs to prepare a summary of the city's Public Notification Policy for a new community engagement initiative.

Step 1: Structured input

=== TASK INFORMATION ===
Purpose: Summarise the Public Notification Policy for community
  engagement staff.
Audience: City staff designing engagement initiatives.
Format: 150-word summary with key provisions and a claim inventory.

=== SOURCE RECORDS ===
Source A – Title: City Council Bylaw 2024-12 – Public Notification
Authority: City Council
Date: 2024-03-15
Version: 2.0
Content: [Extracted key provisions]

Source B – Title: City Clerk's Guide to Public Notification – 2025
Authority: City Clerk's Office
Date: 2025-01-10
Content: [Extracted guidance]

=== CONSTRAINTS ===
– Do not exceed 150 words.
– Use plain language.
– Include source references.

=== REQUIRED OUTPUT ===
A 150-word summary with a claim inventory.

Step 2: AI-generated output (provisional)

=== PROVISIONAL OUTPUT – DRAFT ===
Date: 2026-08-29
Status: DRAFT – FOR INTERNAL REVIEW ONLY

=== SUMMARY ===
The Public Notification Policy (Bylaw 2024-12) requires the city to
provide public notice for major decisions, including planning changes
and policy amendments. Notices must be published at least 14 days
before the decision date, using at least two communication channels.
The City Clerk's Guide recommends using social media and the city's
website as primary channels.

=== CLAIM INVENTORY ===
Claim 1: The policy requires public notice for major decisions.
Source: Bylaw 2024-12, Section 2
Confidence: High

Claim 2: Notices must be published at least 14 days in advance.
Source: Bylaw 2024-12, Section 4.1
Confidence: High

Claim 3: Two communication channels are required.
Source: Bylaw 2024-12, Section 4.2
Confidence: High

Claim 4: Social media and the city's website are recommended.
Source: City Clerk's Guide 2025, page 3
Confidence: High

=== UNCERTAINTY LOG ===
– None identified.

=== REVIEWER NOTES ===
– Check that the 14-day requirement is correctly cited.
– Verify the recommendation on social media against the Guide.

Step 3: Review and verification

• Verification: All claims are correctly cited. The 14-day requirement is confirmed. The social media recommendation is confirmed in the Guide.
• Review: The summary is clear, accurate, and complete. No revisions are needed.

Step 4: Approval and permissions

• Approval: Approved by the City Clerk's Office.
• Permissions: May be shared internally with community engagement staff.
• Prohibited: Not for external publication or legal advice.

16. Guided Learning Activities

=== TASK INFORMATION ===
Purpose: Prepare a summary of the new cybersecurity policy for staff.
Audience: All staff (general audience, not technical).
Format: 150-word summary with bullet points.

=== SOURCE RECORDS ===
Source 1 – Title: Cybersecurity Policy 2026
Authority: IT Department
Date: 2026-08-15
Content: [Attached policy document]
Status: Authoritative

Source 2 – Title: Cybersecurity FAQ – Staff Edition
Authority: IT Department
Date: 2026-08-20
Content: [Attached FAQ page]
Status: Supporting

Source 3 – Title: Email from IT Director – Implementation Notes
Authority: IT Director
Date: 2026-08-22
Content: [Additional info from email]
Status: Supporting (unverified)

=== CONSTRAINTS ===
– Do not exceed 150 words.
– Use plain language (grade-8 reading level).
– Do not include technical jargon.

=== REQUIRED OUTPUT ===
A 150-word plain-text summary with bullet points on key provisions.

=== STRUCTURED OUTPUT ===
=== ANALYSIS ===
Option 1: [Name/description]
  Criteria scores:
  – Criterion 1: Score [1-5] with rationale
  – Criterion 2: Score [1-5] with rationale
  – Criterion 3: Score [1-5] with rationale
  Total: [Score]
  Strengths: [List]
  Weaknesses: [List]

Option 2: [Name/description]
  [Same structure]

Option 3: [Name/description]
  [Same structure]

=== RECOMMENDATION ===
Recommended option: [Option X]
Rationale: [2-3 sentences explaining why]

=== SUPPORTING EVIDENCE ===
Evidence 1: [Claim] – Source: [Reference]
Evidence 2: [Claim] – Source: [Reference]

=== UNCERTAINTY LOG ===
– [Uncertainty 1]
– [Uncertainty 2]

=== REVIEWER NOTES ===
– [Notes for reviewer]

1. Context preparation (Human):
   – Select sources
   – Build a context package
   – Define constraints and output format

2. Stage 1 – Source summarisation (AI):
   – Generate summaries of each source
   – Output: Source summaries with source IDs

3. Verification 1 (Human):
   – Check each summary against the source
   – Correct any errors
   – If errors are significant, revise and re-verify

4. Stage 2 – Thematic synthesis (AI):
   – Synthesise summaries into themes
   – Output: Thematic synthesis with source references

5. Verification 2 (Human):
   – Check themes against summaries
   – Ensure all key points are covered

6. Stage 3 – Draft summary (AI):
   – Draft a summary based on the synthesis
   – Output: Draft summary

7. Verification 3 (Human):
   – Check the draft against the sources
   – Check readability and clarity

8. Revision (Human):
   – Revise the draft based on verification

9. Approval (Human):
   – Obtain approval from the policy lead

10. Permitted use:
    – Draft: Internal review only
    – Final: Share with stakeholders

=== PERMISSION BOUNDARIES ===
Draft version (PROVISIONAL):
  – May be shared internally with the communications team for review.
  – May be revised based on feedback.
  – Must not be shared externally.
  – Must not be published.

Final version (APPROVED):
  – May be shared with the community newsletter editor.
  – May be published on the city's website.
  – May be used for public communication.

Prohibited uses (both draft and final):
  – Not for legal advice.
  – Not for policy decisions.
  – Not for internal decision-making without additional review.

17. Exercises

=== TASK INFORMATION ===
Purpose: Prepare a summary of the new employee benefits policy.
Audience: All staff (all-staff meeting).
Format: 100-word summary with key provisions.

=== SOURCE RECORDS ===
Source 1 – Title: Employee Benefits Policy 2026
Authority: HR Department
Date: 2026-07-01
Content: [Attached policy document]
Status: Authoritative

Source 2 – Title: Benefits FAQ – Employee Edition
Authority: HR Department
Date: 2026-07-10
Content: [Attached FAQ]
Status: Supporting

=== CONSTRAINTS ===
– Do not exceed 100 words.
– Use plain language.
– Cover key changes only.

=== REQUIRED OUTPUT ===
A 100-word plain-text summary with key provisions.

=== STRUCTURED OUTPUT ===
=== PROPOSAL SUMMARIES ===
Proposal 1: [Name] – [1-sentence summary]
Proposal 2: [Name] – [1-sentence summary]
Proposal 3: [Name] – [1-sentence summary]

=== EVALUATION CRITERIA ===
Criterion 1: [Description] – Weight: [%]
Criterion 2: [Description] – Weight: [%]
Criterion 3: [Description] – Weight: [%]

=== COMPARISON ===
| Criterion | Proposal 1 | Proposal 2 | Proposal 3 |
|-----------|------------|------------|------------|
| Criterion 1 | Score/Rationale | Score/Rationale | Score/Rationale |
| Criterion 2 | ... | ... | ... |
| Criterion 3 | ... | ... | ... |

=== RECOMMENDATION ===
Recommended proposal: [Name]
Rationale: [2-3 sentences]

=== SUPPORTING EVIDENCE ===
– [Claim] – Source: [Reference]
– [Claim] – Source: [Reference]

=== UNCERTAINTY LOG ===
– [Uncertainty 1]

1. Context preparation (Human)
2. Stage 1 – Source classification (AI)
3. Verification 1 (Human) – Check classifications
4. Stage 2 – Source summarisation (AI)
5. Verification 2 (Human) – Check summaries
6. Stage 3 – Thematic synthesis (AI)
7. Verification 3 (Human) – Check themes
8. Stage 4 – Draft briefing note (AI)
9. Verification 4 (Human) – Check draft
10. Revision (Human) – Revise based on verification
11. Approval (Human) – Obtain approval
12. Permitted use: Internal sharing only

=== PERMISSION BOUNDARIES ===
Draft version (PROVISIONAL):
  – May be shared internally with the policy team for review.
  – May be revised based on feedback.
  – Must not be shared externally.
  – Must not be published.

Final version (APPROVED):
  – May be shared with the city council.
  – May be published on the city's website.
  – May be used for public communication.

Prohibited uses (both draft and final):
  – Not for legal advice.
  – Not for policy decisions without additional review.
  – Not for external advocacy without council approval.

18. Homework Problems

Estimated time: 1.5–2 hours · Formative · Recommended for portfolio preparation.

=== TASK INFORMATION ===
Purpose: Prepare a summary of the new workplace safety policy.
Audience: All employees.
Format: 100-word summary with key provisions.

=== SOURCE RECORDS ===
Source 1 – Title: Workplace Safety Policy 2026
Authority: Health and Safety Committee
Date: 2026-08-01
Content: [Attached policy]
Status: Authoritative

=== CONSTRAINTS ===
– Do not exceed 100 words.
– Use plain language.
– Cover key changes only.

=== REQUIRED OUTPUT ===
A 100-word plain-text summary with key provisions.

=== TASK INFORMATION ===
Purpose: Synthesise findings on remote work and productivity.
Audience: Research team.
Format: Thematic synthesis with source references.

=== SOURCE RECORDS ===
[For each of 8 articles: Title, Author, Date, Journal, Key findings]

=== CONSTRAINTS ===
– Include only peer-reviewed articles.
– Identify 3-5 themes.
– Include source references.

=== REQUIRED OUTPUT ===
A thematic synthesis with source references for each theme.

=== STRUCTURED OUTPUT ===
=== SUMMARY ===
[100-word summary]

=== CLAIM INVENTORY ===
Claim 1: [Claim] – Source: [Reference] – Confidence: [High/Medium/Low]
Claim 2: [Claim] – Source: [Reference] – Confidence: [High/Medium/Low]
...

=== UNCERTAINTY LOG ===
– [Uncertainty 1]
– [Uncertainty 2]

=== REVIEWER NOTES ===
– [Notes for reviewer]

=== STRUCTURED OUTPUT ===
=== OPTIONS ===
Option 1: [Name] – [Description]
Option 2: [Name] – [Description]
Option 3: [Name] – [Description]

=== EVALUATION CRITERIA ===
Criterion 1: [Description]
Criterion 2: [Description]
Criterion 3: [Description]

=== COMPARISON MATRIX ===
| Criterion | Option 1 | Option 2 | Option 3 |
|-----------|----------|----------|----------|
| Criterion 1 | Score | Score | Score |
| Criterion 2 | Score | Score | Score |
| Criterion 3 | Score | Score | Score |

=== RECOMMENDATION ===
Recommended: [Option X]
Rationale: [2-3 sentences]

=== SUPPORTING EVIDENCE ===
– [Claim] – Source: [Reference]

Claim: The policy will reduce costs by 15%.
Source: Policy Impact Assessment 2026, page 3
Source wording: "...up to 15% under optimal conditions."
Confidence: Medium – conditional on optimal conditions
Reviewer note: Verify the conditions.

1. Context preparation (Human):
   – Select sources, build context package, define constraints

2. Stage 1 – Source classification (AI):
   – Output: Classified list
   – Verification: Human checks classifications

3. Stage 2 – Source summarisation (AI):
   – Output: Summaries with source IDs
   – Verification: Human checks summaries against sources

4. Stage 3 – Thematic synthesis (AI):
   – Output: Themes with source references
   – Verification: Human checks themes against summaries

5. Stage 4 – Draft briefing note (AI):
   – Output: Draft briefing note
   – Verification: Human checks draft against sources

6. Revision (Human):
   – Revise based on verification feedback

7. Approval (Human):
   – Obtain approval from the policy lead

8. Permitted use:
   – Draft: Internal review only
   – Final: Share with stakeholders

=== WORKFLOW: PUBLIC HEALTH RESEARCH SYNTHESIS ===

=== STRUCTURED INPUT ===
Task: Synthesise research on the effectiveness of nutrition labelling.
Sources: 12 peer-reviewed articles (attached)
Constraints: Include only articles from 2020-2026.
Required output: Thematic synthesis with source references.

=== STAGES ===
1. Source classification (AI) – Classify by date and relevance.
   Verification: Human checks classifications.

2. Source summarisation (AI) – Summarise each article.
   Verification: Human checks summaries.

3. Thematic synthesis (AI) – Identify 3-5 themes.
   Verification: Human checks themes.

4. Draft synthesis (AI) – Draft a 300-word synthesis.
   Verification: Human checks draft.

5. Revision (Human) – Revise based on verification.

6. Approval (Human) – Obtain approval from the lead researcher.

=== STRUCTURED OUTPUT ===
– Thematic synthesis with source references
– Claim inventory
– Uncertainty log

=== PERMISSION BOUNDARIES ===
– Draft: Internal review only
– Final: Share with research team

=== EXCEPTION HANDLING ===
– If a source is unavailable, document and exclude.
– If themes are unclear, escalate to the lead researcher.

19. Mini‑Project: Multi‑Stage Context and Output Workflow

=== WORKFLOW: PSYCHOLOGY STUDY GUIDE ===

=== STRUCTURED INPUT ===
Purpose: Create a study guide for the psychology exam.
Sources: Syllabus, lecture notes (12 files), textbook (selected chapters)
Constraints: 5-page maximum; include key concepts and definitions.

=== STAGES ===
1. Concept extraction (AI) – Extract key concepts from syllabus and notes.
2. Definition gathering (AI) – Gather definitions for each concept.
3. Synthesis (AI) – Synthesise concepts and definitions into a study guide.
4. Review (Human) – Review and revise the study guide.
5. Approval (Human) – Final approval.

=== STRUCTURED OUTPUT ===
– Concept list with definitions
– Source references for each concept
– Uncertainty log for unclear concepts

=== VERIFICATION POINTS ===
– After Stage 1: Verify concepts against syllabus.
– After Stage 2: Verify definitions against sources.
– After Stage 3: Verify the study guide against sources.

=== PERMISSION BOUNDARIES ===
– Draft: Personal study use only.
– Final: May be shared with study group.

20. Self‑Check Quiz

~20 minutes · Formative · Ungraded. Explanatory feedback is provided for every item.

21. Assignment Connections

22. Key Takeaways

23. Key Terms

Structured input

An organised, labelled set of information that reduces ambiguity.

Structured output

An organised, labelled set of information produced by an AI system.

Traceability

The ability to link outputs back to specific sources, stages, and decisions.

Workflow

A complete, interconnected sequence of stages, reviews, and decisions.

Verification gate

A point where the workflow stops for accuracy checking.

Review gate

A point where the workflow stops for quality assessment.

Approval gate

A point where formal permission is obtained.

Permission boundary

An explicit rule specifying what outputs may be used for.

Provisional output

A draft output that requires review before use.

Claim inventory

A list of claims with source references and confidence levels.

Uncertainty log

A record of uncertain or unverified claims.

Produce, Review, and Permit (SCOPE)

The element of SCOPE that focuses on managing outputs.

24. Looking Ahead: Lesson 5.4

In this lesson, you learned to design structured inputs and outputs, connect stages into complete workflows, and establish permission boundaries. In Lesson 5.4, you will learn to detect and address context drift, conversation drift, task drift, output drift, and workflow drift. You will also learn about prompt injection and how to apply the complete SCOPE framework—including the Evaluate, Escalate, and Evolve element—to maintain control over your workflows.

After Lesson 5.4, you will be ready to apply the complete SCOPE framework to your own workflows and to prepare for Unit 6: Human-AI Collaboration.

25. Sources & Further Reading

• Primary source: Chapter 5: Advanced Prompting and Context Engineering, COMP 3270 textbook.
• Supporting sources: Appendix B (Prompt Design Toolkit), Appendix E (AI Workflow Design Toolkit), and Appendix D (Responsible-Use Toolkit) from the COMP 3270 course materials.
• Institutional guidance: [INSERT VERIFIED REFERENCE TO INSTITUTIONAL AI POLICY]
• Further reading: [INSERT VERIFIED REFERENCE TO RELEVANT PEER-REVIEWED LITERATURE]

Note: All page numbers, specific definitions, and direct quotations from the textbook are to be confirmed from the supplied course materials. [PAGE NUMBER TO BE CONFIRMED]