Case C — new rubric
5 criteria × 4 levels, 40 points (mapped to the 30% course weight in the gradebook table). Three criteria are process criteria — the capabilities the stress test showed the original rubric couldn’t see.
The rubric
| Criterion | 10 — Accomplished | 7.5 — Proficient | 5 — Developing | 2.5 — Beginning |
|---|---|---|---|---|
| 1. Research design & ownership (10) | Twist is justified against the shared protocol; design choices follow from the research question; every named member can defend one design decision | Twist justified; design mostly coherent; choices defensible | Twist present but weakly justified; design choices unexplained | No meaningful design engagement; template decisions |
| 2. Data practice & integrity (10) | Data de-identified and well documented; collection problems surfaced in the notebook and handled; decisions traceable | Data documented; problems noted; minor traceability gaps | Data thin or undocumented; problems glossed over | Data provenance unclear; notebook post-hoc or absent |
| 3. Analytic reasoning (10) | Test choice justified with assumptions checked; “what would change my decision” answered concretely; interpretation tied to this data | Correct test with partial justification; interpretation specific | Correct mechanics, thin justification; generic interpretation | Test mismatched; interpretation absent or wrong |
| 4. Communication (10) | Report is accurate, APA-consistent, and reads as a professional record of this study; AI use disclosed and appropriate | Accurate and structured; minor APA slips; disclosure complete | Readable but uneven; disclosure vague | Disorganized; undisclosed AI shaping analysis or results |
| 5. Reflective practice (10) | Notebook shows genuine problem-solving and change of course where needed; Q&A answers show learning from the process | Notebook consistent; Q&A mostly confident | Notebook present but shallow; Q&A hesitant | Notebook thin; Q&A evasive or unprepared |
Alignment to the course outcome
| Outcome element (“conduct a small study responsibly”) | Rubric criterion | Evidence source |
|---|---|---|
| Design a measure / adapt a protocol | 1 | Twist registration + Q&A (method question) |
| Collect and manage data responsibly | 2 | Dataset + collection notes + notebook |
| Choose and justify the analysis | 3 | Analysis-choices memo + Q&A |
| Communicate to professional standards | 4 | Report (+ AI-use statement) |
| Reflect and adjust | 5 | Notebook + Q&A (limitation/next-step question) |
Gradebook mapping
| Component | Weight | Criteria |
|---|---|---|
| Team registration + twist (week 8) | 2% | 1 (checklist) |
| Data collection + dataset (weeks 8–9) | 8% | 2 (checklist + spot-check) |
| Lab notebook (weeks 8–11) | 8% | 2, 5 |
| Analysis-choices memo (week 11) | 4% | 3 |
| Report (week 12) | 10% | 3, 4 |
| Structured Q&A (weeks 12–13) | 6% | 1, 3, 5 (per member) |
| Total | 38% → scaled to 30% |
(The raw total is 38 points; the course handbook caps the task at 30% of the final grade, so component weights are scaled by 30/38 — shown here to keep the arithmetic honest.)
Why this rubric replaced the original
- The original’s Method (5) + Results (7) = 12 of 30 points rewarded the prose of statistics — exactly what models produce. Criterion 3 now rewards the decision and its defence (memo + Q&A), not the description.
- The introduction/literature criterion (6 marks) folds into Communication (criterion 4) — grounding matters, but it is no longer a third of the assurance story.
- Criteria 1, 2 and 5 are new process criteria: they grade what AI cannot supply, and they make the notebook and Q&A load-bearing rather than decorative.
- Five criteria still mark fast: the memo is 500 words, the notebook is checklist-spot-checked, and the Q&A is scored per member on three anchors.