D3 · Assessment Redesign Sheet — Interactive Assessment Session 5 · Workshop Part 2 · Online

D3 · Assessment Redesign Sheet

Audit an existing assessment, assign every component a lane (🔴🟡🟢), and swap-test the result with a partner.

Honesty guard

"Red is not failure — it is protection. If every component is 🟢 the assessment has no core; if every component is 🔴 the pedagogical load is unworkable. The constraint is structural: ≥1 🔴 and total student workload flat or lower. Name the exploit your partner found — a real patch means you actually stress-tested the design."

Entries are temporary until saved.

Assessment & Learning Outcome

Redesign — Components

Assign every assessment component a lane. 🔴 Restricted — no AI (protects the core outcome). 🟡 Permitted with disclosure — AI allowed, process evidence required. 🟢 Required — AI use is part of the task.

# Component Lane Weight Process evidence
Design constraints ≥1 🔴 component protecting the core outcome · 🟡 components name their process evidence · total student workload flat or lower than the pre-redesign version

Swap-test results

Have a partner try to break your redesigned assessment. What exploit did they find? What patch did you apply?

How the Assessment Redesign Sheet works

The D3 Redesign Sheet is the core of Session 5 — it takes you through the audit → redesign → swap-test cycle. The output goes into your portfolio and becomes evidence for Guideline component 3 (the permission architecture) in S7.

  1. Name the assessment — one real assessment you currently run. Be specific: course name, word count, submission format.
  2. State the protected learning outcome — the one skill this assessment exists to verify. If AI can produce a passing answer without that skill being practised, the assessment is vulnerable.
  3. Audit it — run P09 (vulnerability audit). Paste the AI you used, the prompt, and the estimated grade the AI could achieve. This is your pre-redesign baseline.
  4. Redesign every component — split the assessment into components (e.g. "in-class viva", "annotated source table", "final analysis"). Assign each a lane and a weight. The whole must add to 100%.
  5. Meet the constraints — at least one 🔴 component protecting the core outcome. Every 🟡 component must name its process evidence. Total workload flat or lower.
  6. Swap-test — give the redesign to a partner. They try to break it. Record the best exploit they found and the patch you applied.
  7. Save your snapshot — name it "S5 Redesign Sheet". Export as JSON for backup or submission.

Lane definitions

  • 🔴 Restricted — No AI use permitted. Protects the core outcome directly. Process evidence is the work itself (in-class, handwritten, annotated printout).
  • 🟡 Permitted with disclosure — AI use allowed but must be disclosed and documented. Process evidence shows the student's interaction with AI (prompts, iterations, verification).
  • 🟢 Required — AI use is part of the task. The student's ability to delegate, discern, and diligence is what is being assessed. Evidence includes the final output and the verification log.

Quality bar

  • Protected learning outcome is specific and observable — not "critical thinking" but "construct a sociological argument using Marx and Weber's theories of inequality."
  • Components are non-overlapping — a colleague could assign a grade from the description alone.
  • Weights sum to ~100% — no mystery weighting.
  • Swap-test exploit is concrete — "the student could ask AI to write the annotated table" not "students might cheat."
  • Patch is specific — "require inline page-number citations" not "monitor more closely."
Assessment

Assessment name: Theory Application Essay (1,500 words) — Introductory Sociology

Protected learning outcome: Analyse a social phenomenon (economic inequality in Malaysia) using Marx, Weber, and Durkheim — citing specific concepts from each theorist.

AI-audit grade before redesign (P09): B+ — GPT-4 produced a competent essay on Malaysian inequality citing all three theorists with plausible sources. Only in-class vs AI review revealed over-reliance on Weber's rationalisation concept and absence of Durkheim's anomie. Estimated grade without detection safeguards: B+/A-.

Component table
ComponentLaneWeightProcess evidence
In-class viva (10 min, post-submission)🔴25%Verbal defence — tutor probes which theorist's concept the student chose and why; handwritten notes taken during viva
Annotated source table with process draft🟡30%PDF of AI chat log showing prompts + iterations + inline citations to specific page numbers of assigned readings
Final analysis (submitted essay)🟢45%AI-assisted draft with disclosure statement and verification log — graded on argument quality and source use, not drafting
Swap-test

Best exploit found by partner: "A student could ask the AI to draft the annotated source table and fabricate page-number citations to the readings, then claim the AI log is just for the final analysis. The in-class viva would catch some of this, but if the student prepares AI-generated talking points, they might slip through a shallow viva."

Patch applied: "Viva questions now include one from the tutor's own reading — 'on page 47 of the Marx chapter, what does he say about commodity fetishism?' — requiring genuine reading, not summary. Annotated source table must reference at least two passages from the readings the student actually quotes, with page numbers verified against a shared reading list."