COMP 3270 / Module 2 / Unit 6 / Lesson 4

Lesson 6.4 · Week 4 · Reading: Chapter 6

PARTNER Integration, Escalation, and Stop Conditions

Integrate PCAFE-V, SCOPE, and the complete PARTNER framework; design exception handling, escalation routes, and stop conditions; document AI contributions and human decisions; and evaluate whether a collaboration should continue, be redesigned, or be retired.

Estimated study time: 180–220 min Prerequisites: Lessons 6.1–6.3; PCAFE-V, SCOPE, PARTNER Module 2 · Week 4

Table of Contents

1. Lesson Overview

This is the culminating lesson of Unit 6. You have learned to classify AI roles, allocate tasks, retain human authority, and design meaningful oversight. Now we bring everything together through the full PARTNER framework—with a special focus on the final three elements: N (Note Contributions and Decisions), E (Escalate Exceptions), and R (Review and Refine). You will also learn how to integrate PCAFE-V (prompt design) and SCOPE (context and workflow control) with PARTNER to create a complete collaboration plan.

This lesson covers designing testing and verification points, documenting AI contributions and human decisions, defining exceptions and escalation pathways, establishing stop conditions, and evaluating whether the collaboration should continue, be redesigned, or retired. By the end, you will have a comprehensive plan that prepares you for Assignment 2 and future professional collaborations.

Why this matters: A collaboration plan that is not documented, tested, and reviewed is a plan waiting to fail. This lesson teaches you to close the loop—capturing what happened, escalating what cannot be handled, and refining what can be improved. That closure is what turns a promising pilot into a responsible practice.

2. Learning Outcomes

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

  1. Integrate PCAFE-V, SCOPE, and PARTNER in a bounded workflow.
  2. Apply all seven PARTNER elements: Purpose, Allocate Tasks, Retain Human Authority, Test and Verify, Note Contributions and Decisions, Escalate Exceptions, Review and Refine.
  3. Design testing and verification points for prompts, outputs, and the review process.
  4. Document AI contributions, human revisions, decisions, approvals, and exceptions.
  5. Define exception triggers, escalation routes, and stop conditions.
  6. Evaluate whether the collaboration should continue, be redesigned, or be retired.
  7. Complete the required preparation for Assignment 2: Prompt Engineering Portfolio.

3. Before You Begin

4. Retrieval Practice: Concepts from Lessons 6.1–6.3

Recall concepts from previous lessons to answer these questions.

Question 1: What is the difference between review authority and decision authority?

Show answer

Review authority is the right to examine and critique AI outputs; decision authority is the right to choose among alternatives. A reviewer may flag issues but not make the final choice.

Question 2: In PARTNER, what is the purpose of Test and Verify?

Show answer

To ensure that prompts, outputs, sources, calculations, and the oversight process are reliable and that the workflow behaves as expected, especially in edge cases.

Question 3: How does PCAFE-V contribute to prompt design?

Show answer

PCAFE-V guides the construction of effective prompts by focusing on Purpose, Context, Audience, Format, Evaluation, and Version control—ensuring clarity and traceability.

5. Opening Scenario: "The Expanding Assistant"

📌 Scenario — The Expanding Assistant

A legal team uses an AI drafting assistant to generate initial contract clauses. Originally, the AI was a simple Assistant—generating alternatives for lawyers to review. Over time, the team added an external tool that checks clauses against a regulatory database, and the AI now suggests which clauses are "low risk" and "high risk." The review workload has increased, and a junior lawyer now approves most low-risk clauses without accessing the original negotiation notes. An unusual case arises involving a novel regulatory interpretation; the AI's recommendation conflicts with the junior lawyer's intuition, but the lawyer approves it anyway due to time pressure. Accountability is unclear—was it the AI, the tool, or the lawyer?

Questions:

  1. What role expansion and workflow drift have occurred?
  2. What oversight failure is evident?
  3. What escalation trigger should have been activated?
  4. What stop condition might apply?
  5. Where is the accountability gap?
Show guided analysis

Role expansion: The AI moved from Assistant to Recommender (risk labelling) and may be approaching Agent if it integrates with the tool automatically.

Workflow drift: The junior lawyer is approving without original evidence, relying on the AI's risk label.

Oversight failure: The lawyer lacks adequate time and evidence; oversight is nominal.

Escalation trigger: Novel regulatory interpretation—should escalate to a specialist (senior lawyer or external counsel).

Stop condition: If the AI's recommendation conflicts with the lawyer's professional judgment, the workflow should pause and escalate.

Accountability gap: No one is clearly responsible for the final clause; the process diffuses accountability among the AI, tool, and junior lawyer.

6. Integrating PCAFE-V, SCOPE, and PARTNER

These three frameworks are complementary and should be used together:

In practice, you start with Purpose (PARTNER P), then use PCAFE-V to design the prompt and SCOPE to define the workflow boundaries. Then you Allocate Tasks and Retain Human Authority (A & R). Next, Test and Verify (T) the prompts and workflow, Note contributions and decisions (N), Escalate exceptions (E), and Review and Refine (R) the entire process.

Figure 1 (placeholder): [INSERT INTEGRATION DIAGRAM: PCAFE-V + SCOPE + PARTNER as layers]
Section summary
  • PCAFE-V designs prompts; SCOPE controls context and workflow; PARTNER governs the whole collaboration.
  • The frameworks are used together, not in isolation.
  • Integration happens through deliberate mapping of each element.

7. PARTNER: Note Contributions and Decisions, Escalate Exceptions, Review and Refine

N: Note Contributions and Decisions

Documentation is essential for transparency, accountability, and reconstructing the process. You must record:

Guiding questions: What did the AI contribute? What did the human change? What was rejected? Who approved the result? Could the process be reconstructed?

E: Escalate Exceptions

Exceptions are situations that fall outside the normal workflow and require special handling. Common exceptions include:

An escalation process must specify: trigger, receiving role, information required, status while unresolved, timeline, interim restrictions, final authority, and documentation.

R: Review and Refine

The final element is ongoing evaluation. Ask:

Refinement may lead to changes in prompts, task allocation, authority, oversight, or even discontinuation of AI use.

Section summary
  • N documents contributions, decisions, and versions for accountability.
  • E defines how exceptions are identified and escalated.
  • R evaluates the collaboration and adjusts it over time.

8. Exception Handling and Escalation

Design a clear process for each exception type. For example:

Table 1. Exception escalation example
Exception Trigger Receiving role Information package Interim action
Novel legal interpretation AI cannot find a relevant precedent Senior legal counsel AI output, original sources, AI's reasoning Stop workflow; do not finalise
Conflicting sources AI identifies contradictory evidence Subject-matter specialist All conflicting sources, AI analysis Flag for review; proceed only after resolution
Suspected prompt injection AI produces unusual or harmful output Security/IT Full prompt and output history Immediately stop and quarantine
Section summary
  • Escalation requires a specific trigger, a receiving role, an information package, and interim action.
  • Escalation is not a failure—it is a designed control.
  • Exception handling must be tested and refined.

9. Stop Conditions

A stop condition is a predefined situation that triggers an immediate halt to the workflow. Stop conditions are distinct from escalations—they are for situations where the risk is too high to continue without fundamental review. Examples:

Stop authority must be assigned to a specific human role, and the stop process must be tested.
Section summary
  • Stop conditions are safety triggers that halt the workflow immediately.
  • They must be predefined, assigned to a person, and tested.
  • Stop conditions differ from escalations in their urgency and consequence.

10. Testing and Verification

In addition to testing prompts and outputs (which you learned in SCOPE), you must test the workflow itself, including:

Testing should be documented, and the results should inform refinement.

Section summary
  • Testing includes prompts, outputs, permissions, stop behaviour, and the review process.
  • Test results must be documented and used for refinement.
  • Untested controls cannot be relied upon.

11. Documentation and Recordkeeping

A robust documentation system supports accountability and continuous improvement. Key records include:

This documentation is essential for audits, appeals, and future refinement.

Section summary
  • Documentation must capture both AI contributions and human decisions.
  • It supports audits, appeals, and refinement.
  • A record that cannot be reconstructed cannot be defended.

12. Review and Refinement Process

Schedule regular reviews of the collaboration. Use a structured approach:

  1. Collect data on performance, errors, exceptions, and user feedback.
  2. Assess whether the purpose is still valid and being achieved.
  3. Evaluate task allocation—are tasks appropriately assigned?
  4. Review oversight quality—is it still meaningful?
  5. Check for role drift, scope creep, or permission expansion.
  6. Identify new risks or hidden work.
  7. Make decisions: continue, continue with safeguards, limit, redesign, pause, use non-AI alternative, require specialist review, or retire.
  8. Document changes and communicate to stakeholders.

The refinement cycle is ongoing; collaboration designs are never finished.

Section summary
  • Review is periodic and structured.
  • It assesses purpose, allocation, oversight, and drift.
  • Decisions range from continuing to retiring the workflow.

13. PARTNER Walkthrough — Note, Escalate, Review and Refine

We now apply the remaining PARTNER elements to the academic example (generating discussion questions) that we have been building.

N: Note Contributions and Decisions

Documentation plan:

  • Record the prompt version (using PCAFE-V) and the context package (readings, criteria).
  • Save the AI's draft questions and the instructor's revised versions.
  • Note which questions were rejected and why.
  • Document the verification step (pilot feedback).
  • Record the final approval (instructor name, date).
  • Include an AI-use disclosure (e.g., "AI was used to generate draft questions; final selection and revision by instructor.").

E: Escalate Exceptions

Escalation triggers and routes:

  • Trigger: AI generates questions that are off-topic or factually incorrect despite context. Route: Instructor revises prompt; if persists, escalate to AI support or department head.
  • Trigger: A student challenges a question's validity. Route: Instructor reviews; if unresolved, escalate to curriculum committee.
  • Trigger: The AI attempts to generate sensitive or inappropriate content. Route: Stop immediately; escalate to IT and academic integrity office.

R: Review and Refine

Review schedule: After each use, the instructor briefly reflects. At the end of the semester, a formal review:

  • Did the AI save time? Was quality maintained?
  • Were there any issues with question quality or student engagement?
  • Should the AI role change (e.g., add a critic function)?
  • Is the oversight still meaningful? (review workload, time)
  • Decision: Continue as is, adjust prompts, or stop using AI.

14. Process Walkthrough: Complete Collaboration Cycle

We walk through a complete cycle for the academic use case, integrating all frameworks.

1. Purpose (PARTNER P): Instructor needs draft discussion questions.

2. Prompt Design (PCAFE-V): Instructor crafts a prompt with purpose, context (readings), audience (300-level students), format (list of questions), evaluation criteria (critical thinking, alignment), and versioning.

3. Context Control (SCOPE): Instructor defines source hierarchy (readings first, then general knowledge if needed), context package, output format, permission (only generate drafts), escalation (if off-topic).

4. Task Allocation (PARTNER A): AI generates; instructor reviews and revises; final selection human-only.

5. Retain Authority (PARTNER R): Instructor has decision, action, stop authority.

6. Test and Verify (PARTNER T): Test the prompt with a sample; verify output against criteria; test the stop button.

7. Note (PARTNER N): Document prompt, context, outputs, revisions, approvals.

8. Escalate (PARTNER E): If AI produces problematic questions, instructor escalates to department or revises prompt.

9. Review and Refine (PARTNER R): After each use, reflect; after semester, formal review; decide to continue, adjust, or stop.

This cycle ensures that the collaboration is purposeful, controlled, and continuously improved.

15. Worked Example: "Travel Itinerary Planner"

A travel agent uses an AI to generate personalised travel itineraries for clients. The client provides preferences, budget, and dates. The AI generates a draft itinerary with flights, hotels, and activities. The agent reviews, adjusts, and presents it to the client.

PARTNER application:

  • P (Purpose): Generate a draft itinerary that meets client preferences, within budget, and compliant with travel regulations.
  • A (Allocate): AI generates draft; agent reviews and edits; client final approval.
  • R (Retain): Agent has decision authority (which itinerary to present), action authority (booking), stop authority (if AI suggests unsafe or illegal options).
  • T (Test): Test the AI on previous client data; test edge cases (e.g., last-minute changes, cancellations).
  • N (Note): Document prompt, client preferences, AI draft, agent edits, booking confirmations.
  • E (Escalate): If AI suggests a destination with travel advisories, escalate to agent for manual verification; if booking fails, escalate to technical support.
  • R (Review): After each trip, review client satisfaction and AI performance; refine prompts and context based on feedback.

Stop conditions: If AI repeatedly suggests unaffordable options, stop and redesign the prompt. If AI attempts to book without agent approval, stop immediately.

Integration: PCAFE-V used for prompt design (purpose, audience, format); SCOPE defines source boundaries (e.g., only use approved travel databases).

Show analysis of accountability

The agent is accountable for the final itinerary. The AI is a tool; the agent must verify all bookings and ensure compliance. Responsibility is not laundered because the agent makes the final decision and can be held accountable.

16. Self-Check Quiz

Eight questions · Formative · Ungraded. Expand each for feedback.

Question 1: What does the N in PARTNER stand for?

Select one:
Show answer and explanation

Answer: a) Note Contributions and Decisions.

Question 2: Which of the following is a stop condition?

Select one:
Show answer and explanation

Answer: b) Reviewer workload exceeds capacity.

Stop conditions are serious triggers that halt the workflow; off-topic outputs might be escalated, but capacity overload is a stop condition.

Question 3: What should be included in documentation (N) for an AI-assisted task?

Select one:
Show answer and explanation

Answer: b) Prompt versions, AI output, human revisions, and approvals.

Question 4: When an exception occurs, what is the first step in escalation?

Select one:
Show answer and explanation

Answer: b) Identify the trigger and assemble the information package.

Escalation requires a clear trigger and relevant information to hand off to the receiving role.

Question 5: In the review and refine (R) step, what should you evaluate?

Select one:
Show answer and explanation

Answer: b) Purpose, task allocation, authority, oversight, and overall suitability.

Review and refine is comprehensive, looking at all aspects of the collaboration.

Question 6: How does PCAFE-V relate to PARTNER?

Select one:
Show answer and explanation

Answer: b) PCAFE-V is used for prompt design, while PARTNER is for overall collaboration design.

They are complementary; both are used together.

Question 7: Which of these is a suitable exception trigger?

Select one:
Show answer and explanation

Answer: b) AI identifies a novel case not covered by the source hierarchy.

Novel cases are exceptions that require human judgment.

Question 8: What is the purpose of testing the stop mechanism (T)?

Select one:
Show answer and explanation

Answer: b) To ensure the stop authority can effectively halt the workflow.

Stop authority must be operational and tested to be meaningful.

17. Exercises

Complete these six exercises to apply integration, escalation, and stop conditions.

📝 Exercise 6.4.1: Apply PARTNER to a Bounded Scenario
  • Learning outcome: Apply all PARTNER elements
  • Estimated time: 20 min

Scenario: A small business owner uses an AI to generate social-media post ideas and draft captions. The owner reviews, edits, and posts.

Instructions: Briefly apply all seven PARTNER elements to this scenario.

Show sample solution design
  • P: Generate engaging posts to increase brand awareness.
  • A: AI generates drafts; owner reviews and edits.
  • R: Owner has decision and action authority; stop authority.
  • T: Test AI on past successful posts; verify tone and accuracy.
  • N: Record prompts, drafts, edits, and post dates.
  • E: If AI suggests inappropriate content, owner escalates to themselves (revise) or stop.
  • R: Review engagement metrics monthly; refine prompts based on what works.
📝 Exercise 6.4.2: Integrate PCAFE-V, SCOPE, and PARTNER
  • Learning outcome: Integrate frameworks
  • Estimated time: 20 min

Scenario: A researcher uses an AI to summarise academic articles. The researcher defines the research question, selects articles, and uses a prompt to generate summaries. The researcher reviews and incorporates summaries into a literature review.

Instructions: For each framework (PCAFE-V, SCOPE, PARTNER), identify one key element and explain how it applies.

Show sample answer
  • PCAFE-V: Purpose – summarise key findings; Context – specific articles; Audience – academic; Format – bullet points; Evaluation – accuracy; Version – track prompt changes.
  • SCOPE: Source – only selected articles; Context – research question; Output – concise summaries; Permission – generate only; Escalation – if summary is inconsistent, re-prompt.
  • PARTNER: Purpose – integrate findings; Allocate – AI summarises, researcher reviews; Retain – researcher decides which summaries to use; Note – record summaries and revisions; Escalate – if AI misinterprets, consult original; Review – assess if summaries improve literature review.
📝 Exercise 6.4.3: Create a Testing and Verification Plan
  • Learning outcome: Design testing
  • Estimated time: 15 min

Scenario: A teacher uses an AI to generate quiz questions. The AI produces questions based on learning objectives.

Instructions: List five tests that should be performed before using the AI in a live class.

Show sample plan
  1. Test the prompt with a sample of learning objectives to ensure questions are relevant.
  2. Check for factual accuracy by comparing questions to known sources.
  3. Test edge cases: ambiguous objectives, complex topics.
  4. Check for bias or inappropriate content.
  5. Test the review process: have a colleague review a set of questions and check if they catch any issues.
📝 Exercise 6.4.4: Create a Contribution and Decision Record
  • Learning outcome: Document contributions
  • Estimated time: 15 min

Scenario: A project manager uses an AI to generate weekly status reports. The AI pulls data from project tools and drafts the report. The manager reviews, edits, and sends it.

Instructions: Draft a sample documentation record for one week, including at least the prompt used, AI output, human edits, and approval.

Show sample record

Date: 2026-03-15

Prompt v2.1: "Generate a status report for project X using data from the attached CSV. Highlight risks and milestones."

AI Output: [Draft report text]

Human Revisions: Added context on client feedback; corrected a milestone date; reworded risk section.

Approved by: PM (Jane Doe) at 10:30 AM

Notes: Verified data against source; all correct.

📝 Exercise 6.4.5: Design Exception and Escalation Procedures
  • Learning outcome: Design escalation
  • Estimated time: 20 min

Scenario: A financial analyst uses an AI to generate investment recommendations. The AI analyses market data and suggests a portfolio. The analyst reviews and presents to clients.

Instructions: Define three exception triggers and the corresponding escalation route for each.

Show sample escalation design
  • Exception 1: AI recommends a stock with negative earnings surprise. Escalation: Analyst manually investigates; if uncertain, escalate to senior analyst.
  • Exception 2: AI suggests an investment that violates client's ethical restrictions. Escalation: Analyst stops and consults client or compliance officer.
  • Exception 3: AI generates a recommendation that is significantly different from market consensus. Escalation: Analyst verifies with external data; if conflict persists, escalate to investment committee.
📝 Exercise 6.4.6: Define Stop Conditions and a Review Decision
  • Learning outcome: Define stop conditions and review decisions
  • Estimated time: 15 min

Scenario: A news editor uses an AI to draft article summaries. The AI uses a large language model and sometimes makes factual errors. The editor reviews every summary.

Instructions: List at least three stop conditions for this workflow. Then, after a review, decide whether to continue, redesign, or stop using the AI, and justify.

Show sample stop conditions and decision

Stop conditions:

  • AI makes factual errors in more than 5% of summaries.
  • Editor spends more time correcting than if they wrote from scratch.
  • AI produces content that is offensive or inaccurate.

Review decision: After 1 month, error rate is 3%, time saving is 20%. Proceed with safeguards (improve prompts and add a fact-checking step).

18. Homework Problems

Formative · Ungraded · Recommended for portfolio preparation and Assignment 1 evidence.

🧠 Problem 6.4.1: PARTNER Element Matching

Match each PARTNER element to its description:

  1. Purpose
  2. Allocate Tasks
  3. Retain Human Authority
  4. Test and Verify
  5. Note Contributions and Decisions
  6. Escalate Exceptions
  7. Review and Refine

Descriptions: (a) Document the process, (b) Define what the AI and humans do, (c) Ensure humans can stop and decide, (d) Define the legitimate goal, (e) Handle unusual situations, (f) Check that everything works, (g) Evaluate and improve over time.

Show solution
  • 1-d, 2-b, 3-c, 4-f, 5-a, 6-e, 7-g
🧠 Problem 6.4.2: Complete PARTNER Analysis

For the travel itinerary planner worked example, write a brief PARTNER analysis for each element.

Show solution design

Refer to the worked example; elaborate on each element with specific actions and roles.

🧠 Problem 6.4.3: Distinguish PCAFE-V, SCOPE, and PARTNER

Explain the primary focus of each framework and provide an example of what each would contribute to a project proposal review workflow.

Show solution design
  • PCAFE-V: Crafting the prompt to evaluate proposals (e.g., purpose: summarise and evaluate; context: proposal documents; audience: review committee).
  • SCOPE: Defining source boundaries (only the proposal and evaluation criteria), output format (score and comments), and escalation (if proposal is outside scope).
  • PARTNER: Allocating roles (AI generates draft evaluation, reviewer finalises), retaining decision authority (committee decides), testing (pilot with past proposals), documenting, escalating, and refining.
🧠 Problem 6.4.4: Testing Plan

Design a testing plan for an AI system that generates press releases from a company's internal data. Include at least five tests.

Show solution design
  • Test accuracy of key facts against source data.
  • Test tone and appropriateness for public release.
  • Test edge cases (e.g., negative news, complex data).
  • Test consistency across multiple runs.
  • Test the review process: can a human catch errors in a simulated audit?
  • Test the stop mechanism.
🧠 Problem 6.4.5: Documentation Record

Create a sample documentation record for a single use of an AI that drafts performance reviews for employees. Include prompt, AI output, human edits, and approval.

Show solution design

Provide a realistic example with a specific employee, performance data, prompt, AI draft, and manager's changes.

🧠 Problem 6.4.6: Exception and Escalation

A hospital uses an AI to recommend discharge summaries. Define three exception triggers and describe the escalation route for each.

Show solution design
  • Trigger: AI recommends a medication that conflicts with patient allergies. Route: Stop, notify attending physician immediately.
  • Trigger: AI summary omits a critical diagnosis. Route: Escalate to the department head for review.
  • Trigger: AI generates a summary with inconsistent dates. Route: Flag for nurse review and correction.
🧠 Problem 6.4.7: Stop Conditions

A customer service chatbot is used to handle returns. What stop conditions should be in place?

Show solution design
  • Chatbot cannot understand the customer's request after three attempts.
  • Customer requests to speak to a human.
  • Chatbot suggests an action that violates policy.
  • Chatbot provides incorrect product information.
  • System detects a possible fraud attempt.
🧠 Problem 6.4.8: Review and Refine Decision

After six months of using an AI to generate legal contract summaries, the review team finds that error rates are low, but lawyers spend 30% more time reviewing than before because they must verify the AI's summaries thoroughly. What decision should the team make? Justify.

Show solution design

Possible decisions: continue with improvements (e.g., better prompts, training), redesign (e.g., use AI only for first draft), or stop if no efficiency gain. Justify based on cost-benefit and quality.

🧠 Problem 6.4.9: Transfer Problem — Apply to a New Context

You are a human resources manager implementing an AI to screen resumes and shortlist candidates. Develop a full collaboration plan using PARTNER, incorporating PCAFE-V and SCOPE.

Show solution design

Detailed plan with all seven PARTNER elements, prompt design (PCAFE-V), source and context control (SCOPE), and integration.

🧠 Problem 6.4.10: PARTNER Element-Matching

Match the following activities to the PARTNER element they support:

  • Conducting a pilot test
  • Recording the AI's draft and the final version
  • Defining who can stop the workflow
  • Identifying a novel case that requires supervisor input
  • Surveying users about the AI's usefulness
  • Defining the problem the AI should address
Show solution
  • Pilot test → T (Test and Verify)
  • Recording → N (Note)
  • Stop authority → R (Retain Human Authority)
  • Novel case → E (Escalate)
  • Survey → R (Review and Refine)
  • Defining problem → P (Purpose)
🧠 Problem 6.4.11: Documentation of Exception

Write a sample exception log entry for a situation where an AI-generated report contained a factual error that was caught by the reviewer.

Show solution design

Date: 2026-03-20

Exception: AI report cited incorrect sales figures.

Trigger: Reviewer noticed discrepancy with source data.

Action: Corrected figures; escalated to AI support to investigate the prompt.

Resolution: Prompt was revised to include explicit instruction to use only supplied data.

🧠 Problem 6.4.12: Stop Condition Design

A professor uses an AI to grade multiple-choice exams. The AI has been reliable, but the professor wants to have stop conditions. List three stop conditions and explain why each is necessary.

Show solution design
  • If the AI scores a student with a perfect score when the professor knows they are weak—this could indicate a bug.
  • If the AI marks correct answers as wrong for more than 5% of questions—systematic error.
  • If a student appeals and the professor finds a pattern of misgrading—stop and manually regrade all.

19. Mini-Project: Complete PARTNER Collaboration Plan

📁 Project: Complete PARTNER Collaboration Plan
  • Purpose: This final mini-project integrates all your work from Unit 6. You will produce a comprehensive collaboration plan for your chosen use case, incorporating all PARTNER elements and integrating PCAFE-V and SCOPE. This is the capstone preparation for Assignment 2.
  • Estimated effort: 120–150 minutes.
  • Deliverable: A structured document (approx. 1200–1800 words) addressing 43 items.

Instructions

  1. Use the same bounded AI-supported activity you have been developing (or a new one if necessary).
  2. Do not use sensitive or institutional information.
  3. Cover all 43 sections; you may combine related items.
  4. Complete the self-review checklist.

Required Sections

  1. Use-case title
  2. Human problem
  3. Legitimate purpose
  4. Intended outcome
  5. AI role
  6. Role-classification rationale
  7. AI capabilities
  8. AI limitations
  9. Task inventory
  10. Task allocation
  11. Human-only activities
  12. Prohibited AI activities
  13. Process owner
  14. Information owner
  15. Human reviewer
  16. Specialist reviewer
  17. Reviewer qualifications
  18. Evidence available to reviewer
  19. Review timing
  20. Decision authority
  21. Action authority
  22. Appeal authority
  23. Correction authority
  24. Stop authority
  25. Oversight arrangement
  26. Automation-bias risks
  27. Responsibility-laundering risks
  28. Hidden human work
  29. Testing and verification plan
  30. Contribution and decision record (template)
  31. Exception triggers
  32. Escalation route
  33. Stop conditions
  34. Capability-preservation analysis
  35. Relationship analysis
  36. Review and refinement schedule
  37. Complete PARTNER assessment
  38. Integration with PCAFE-V
  39. Integration with SCOPE
  40. Final recommendation (Proceed, Proceed with safeguards, Pilot/limit, Redesign, Obtain specialist review, Use non-AI alternative, Pause, Do not proceed, Retire)
  41. Assignment 2 portfolio connection
  42. Assignment 1 reflection connection
  43. Preparation for Module 3

Self-Review Checklist

  • Have I covered all 43 sections (or combined logically)?
  • Is the purpose clearly defined and legitimate?
  • Are tasks allocated with clear human authority retained?
  • Have I identified testing and verification points?
  • Is there a documentation plan for contributions and decisions?
  • Are exception triggers and escalation routes defined?
  • Are stop conditions specified?
  • Have I integrated PCAFE-V and SCOPE?
  • Is the final recommendation justified?
  • Does the plan connect to Assignment 2 and Module 3?
Show Hint 1: Purpose and problem

Be specific: what is the human need that AI helps address? Avoid vague statements.

Show Hint 2: Allocation

Use the seven allocation categories; justify each one based on consequence, verifiability, and required expertise.

Show Hint 3: Stop conditions

Think of conditions that would make you immediately halt the workflow—not just minor issues.

Show Hint 4: Integration

Explain how PCAFE-V shaped your prompt and how SCOPE defined context and permissions.

Show illustrative partial example (different use case)

Use case: Using AI to generate project status reports.

Purpose: To save time on routine reporting while maintaining accuracy.

AI role: Assistant – generates draft from project data.

Task allocation: Data gathering – AI-supported; draft generation – AI-generated + human-reviewed; final approval – human-only.

Stop conditions: If AI misinterprets data; if human reviewer overburdened; if project changes scope significantly.

Integration: PCAFE-V for prompt (context: project data, audience: team); SCOPE for source hierarchy (only approved data).

20. Key Takeaways

Remember these points
  • PARTNER is a comprehensive framework for human-AI collaboration: Purpose, Allocate, Retain, Test, Note, Escalate, Review.
  • Integrate PCAFE-V and SCOPE to ensure prompts are well-designed and workflows are controlled.
  • Documentation (N) is essential for transparency, accountability, and auditability.
  • Exception handling and escalation (E) must be planned for common and novel situations.
  • Stop conditions provide a safety net; test them.
  • Review and refine (R) is ongoing; collaboration designs must adapt.
  • This lesson completes your preparation for Assignment 2 by giving you a structured way to present your portfolio contexts.

21. Key Terms

PARTNER
A framework for human-AI responsibility and collaboration: Purpose, Allocate Tasks, Retain Human Authority, Test and Verify, Note Contributions and Decisions, Escalate Exceptions, Review and Refine.
PCAFE-V
A prompt-design framework: Purpose, Context, Audience, Format, Evaluation, Version.
SCOPE
A context and workflow control framework: Source, Context, Output, Permission, Escalation.
Note Contributions and Decisions (N)
The PARTNER element concerned with documenting AI outputs, human revisions, approvals, and exceptions.
Escalate Exceptions (E)
The PARTNER element concerned with defining triggers, routes, and information packages for exceptional situations.
Review and Refine (R)
The PARTNER element concerned with periodic evaluation and improvement of the collaboration.
Exception
A situation that falls outside the normal workflow and requires special handling.
Escalation route
The defined path by which an exception is passed to a receiving role, along with required information.
Stop condition
A predefined situation that triggers an immediate halt to the workflow.
Stop authority
The human authority to halt the workflow or AI system.
Documentation
Prompt versions, AI outputs, human revisions, approvals, exception logs, and final sign-offs.
Refinement cycle
The ongoing process of evaluating and improving a collaboration design.

22. Looking Ahead: Module 3

Congratulations—you have completed Unit 6! You now have a robust toolkit for designing responsible human-AI collaboration. Your next tasks are:

Remember, responsible AI use is a continuous process of learning, testing, and refining. The frameworks you've learned are not one-time checklists but living guides for thoughtful engagement with AI systems.

Continue to Module 3 Return to Unit 6 Home

23. Sources & Further Reading

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]

⚠️ Note

Some textbook content, definitions, examples, and figures are marked with placeholders. These will be supplied by the course coordinator or subject-matter expert.

👤