How the portfolio is evaluated
The portfolio is graded on the quality of the design process and evidence, not on whether an AI system produces an impressive answer. High achievement requires purposeful prompt design, fair comparison, information minimization, staged workflow design, meaningful human responsibility, verification, and clear documentation.
Criteria and weighting
| Criterion | Points | What is assessed |
|---|---|---|
| Use-case definition and suitability | 10 | Purpose, audience, boundaries, information classification, criteria, feasibility, and low-risk suitability |
| Weak-prompt diagnosis | 15 | Identification of missing requirements, assumptions, ambiguity, source needs, verification, and information risks |
| PCAFE-V redesign and rationale | 20 | Completeness, precision, proportional context, evaluability, privacy, verification, and justification of design choices |
| Output comparison and evidence | 15 | Pre-defined criteria, fair comparison, specific evidence, recognition of trade-offs, and separation of fluency from reliability |
| Context minimization and SCOPE design | 15 | Data minimization, source boundaries, task decomposition, intermediate outputs, evaluation gates, revisions, and stop conditions |
| PARTNER responsibility and human oversight | 10 | Appropriate AI and human roles, qualified review, decision authority, automation-bias controls, escalation, and accountability |
| Verification, correction, and residual risk | 10 | Claim checking, source quality, error correction, uncertainty, limitations, fallback, and stopping |
| Communication, documentation, accessibility, and disclosure | 5 | Organization, traceable prompt records, citation integrity, accessible presentation, privacy, and accurate AI-use disclosure |
| Total | 100 |
Performance levels
| Level | Descriptor |
|---|---|
| Excellent, 85–100% | Purposeful, precise, evidence-rich, and fully traceable. Designs and compares prompts rigorously, minimizes information, creates an effective staged workflow, assigns meaningful human authority, verifies consequential content, and explains limitations and residual risk with excellent clarity. |
| Proficient, 70–84% | Complete and well reasoned. Prompt design, comparison, context, workflow, responsibility, and verification are effective, with minor gaps that do not undermine the portfolio. |
| Developing, 55–69% | Meets many core requirements but analysis or evidence is uneven. Some design choices are under-explained, comparison criteria are limited, or workflow and responsibility controls need strengthening. |
| Limited, 40–54% | Partly complete and primarily descriptive. Prompt changes may be cosmetic, criteria unclear, context excessive, oversight nominal, or verification and documentation weak. |
| Insufficient, below 40% | Does not demonstrate a responsible prompt-engineering process. Major components or evidence are missing, the case is unsuitable, outputs are accepted without analysis, or serious integrity and information-handling concerns prevent assessment. |
Criterion-specific guidance
1. Use-case definition and suitability, 10 points
Full achievement: The case has a legitimate low-risk purpose, clear audience and output, bounded scope, pre-defined success criteria, appropriate information classification, and feasible alternatives.
Common deductions: Broad topic, unclear audience, high-consequence use, criteria created after viewing outputs, or protected information.
2. Weak-prompt diagnosis, 15 points
Full achievement: Diagnoses purpose, context, audience, format, criteria, sources, verification, assumptions, ambiguity, and information risks with specific evidence.
Common deductions: Vague comments such as “needs more detail,” redesign without diagnosis, or failure to identify hidden assumptions.
3. PCAFE-V redesign and rationale, 20 points
Full achievement: The redesigned prompt is complete, bounded, proportional, privacy-aware, source-conscious, evaluable, and supported by a persuasive rationale.
Common deductions: Merely longer prompt, unnecessary persona, excessive context, conflicting instructions, or no verification requirements.
4. Output comparison and evidence, 15 points
Full achievement: Applies the same pre-defined criteria to both outputs, cites specific evidence, identifies regressions and trade-offs, and distinguishes usefulness from reliability.
Common deductions: Preference without criteria, cherry-picking, comparing different tasks, or equating fluency with correctness.
5. Context minimization and SCOPE design, 15 points
Full achievement: Removes unnecessary information, defines source boundaries, sequences stages and outputs logically, and includes meaningful evaluation and stop gates.
Common deductions: One oversized prompt, unclear sources, no intermediate review, unnecessary personal data, or workflow diagram without text alternative.
6. PARTNER responsibility and oversight, 10 points
Full achievement: Assigns suitable roles to the AI system, user, qualified reviewer, and final decision-maker with real authority, escalation, and automation-bias controls.
Common deductions: “Human in the loop” without duties or authority, AI making decisions, or no reviewer for specialized claims.
7. Verification, correction, and residual risk, 10 points
Full achievement: Verifies important claims with suitable sources, documents corrections, acknowledges uncertainty, and provides fallback and stopping.
Common deductions: Unverified citations, superficial fact-checking, hidden corrections, or claims that the revised prompt eliminated risk.
8. Communication, documentation, accessibility, and disclosure, 5 points
Full achievement: The portfolio is coherent, professional, traceable, accessible, privacy-preserving, correctly cited, and transparently disclosed.
Common deductions: Missing prompts, inaccessible diagram, weak labels, incomplete disclosure, citation errors, or confidential information.
Grade calculation
The rubric score is calculated out of 100 and converted to Assignment 2's 15% course weight.
Late penalties, extensions, reassessment, and academic-integrity procedures follow the official course and institutional requirements.