Reasoning Rubric (v2)¶
Audience: Faculty (primary), students (for transparency)
Use this rubric to evaluate engineering reasoning, not answer production.
Scoring scale per criterion: 0-4.
How to interpret phase weights¶
Each phase uses its own complete weighting model. Do not add weights across phases. Select one mode that matches your course stage.
Scoring modes by week¶
Week 1 pilot (core only)¶
Score only criteria 1-3.
- Problem framing: 35%
- Assumption quality: 35%
- Quantitative discipline: 30%
Week 2 pilot (add decision quality)¶
Score criteria 1-5.
- Problem framing: 20%
- Assumption quality: 20%
- Quantitative discipline: 25%
- Tradeoff clarity: 15%
- Decision defense: 20%
Full implementation (with AI accountability)¶
Score criteria 1-6.
- Problem framing: 18%
- Assumption quality: 18%
- Quantitative discipline: 22%
- Tradeoff clarity: 16%
- Decision defense: 16%
- Verification trace: 10%
Criteria and performance descriptors¶
1) Problem framing¶
- 0: No clear decision question.
- 1: Topic stated, decision unclear.
- 2: Decision question present but boundary is vague.
- 3: Decision and boundary are explicit.
- 4: Decision, boundary, and success metric are explicit and testable.
2) Assumption quality¶
- 0: Assumptions missing.
- 1: Assumptions listed but untestable.
- 2: Assumptions listed with partial rationale.
- 3: Assumptions justified and linked to model behavior.
- 4: Assumptions include failure conditions and triggers.
3) Quantitative discipline¶
- 0: No calculations/checks.
- 1: Calculations present but inconsistent units or basis.
- 2: Basic calculation is correct.
- 3: Includes reasonableness or bounds check.
- 4: Includes sensitivity or uncertainty check tied to decision.
4) Tradeoff clarity¶
- 0: Single-metric argument only.
- 1: Mentions tradeoff without quantification.
- 2: Identifies at least two competing objectives.
- 3: Shows which objective dominates and why.
- 4: Explicitly compares alternatives and states switching condition.
5) Decision defense¶
- 0: No recommendation.
- 1: Recommendation without evidence.
- 2: Recommendation with limited evidence.
- 3: Recommendation with evidence and caveats.
- 4: Recommendation includes evidence, uncertainty, and what would change the conclusion.
6) Verification trace (AI accountability)¶
- 0: No disclosure of AI-assisted content.
- 1: Mentions AI use but no validation method.
- 2: Lists AI-assisted steps and one basic check.
- 3: Clearly separates AI-assisted vs student-generated reasoning and documents checks.
- 4: Provides a reproducible verification trace with checks tied to the final decision.
Verification trace examples¶
Score 2 example: "AI suggested an alternative and I checked one equation manually."
Score 3 example: "AI drafted alternatives; I separated accepted/rejected items and validated accepted claims with units and bounds checks."
Score 4 example: "AI generated options; I documented each accepted claim, verification method, and how verification changed the final recommendation."
Policy reference: AI Use Policy
ABET-style mapping (adapt to your program)¶
| Outcome focus | Rubric criteria | Evidence artifact in notebook |
|---|---|---|
| SO1: Solve complex engineering problems | 1, 2, 3 | Decision question, assumptions, quantitative checks |
| SO2: Apply engineering design under constraints | 1, 4, 5 | Alternatives compared with switching conditions |
| SO3: Communicate effectively | 5 | Concise recommendation with evidence and caveats |
| SO4: Professional responsibilities and judgment | 2, 5, 6 | Assumption limits, uncertainty, verification trace |
| SO6: Experimentation and data interpretation | 3 | Bounds, sensitivity, and reasonableness checks |
| SO7: Acquire and apply new knowledge | 2, 6 | Explicit belief update and validated external support |
Anchor exemplars (what 2 vs 3 vs 4 looks like)¶
Problem framing anchors¶
- Score 2: "Should we improve conversion or utility first?" (decision present, boundary unclear)
- Score 3: "For the current feed and purity target, should sprint effort prioritize utility intensity over yield?" (decision + boundary explicit)
- Score 4: "For this module and cost horizon, should we prioritize utility intensity over yield to keep total cost below the weekly target?" (decision + boundary + testable success metric)
Assumption quality anchors¶
- Score 2: "Assume tray efficiency is 70%."
- Score 3: "Assume tray efficiency is 70% based on prior run behavior; this drives separation load."
- Score 4: "Assume tray efficiency is 70%; if fouling increases pressure drop beyond design, this assumption fails and recommendation must be revised."
Decision defense anchors¶
- Score 2: "Recommend utility reduction because results look better."
- Score 3: "Recommend utility reduction based on lower modeled duty with stated caveats."
- Score 4: "Recommend utility reduction now; reverse if selectivity drops below threshold or utility price assumptions change materially."
Two-week pilot success criteria¶
At the end of week 2, continue adoption if most of the following are met:
- At least 80% of submissions include an explicit decision question.
- Median score on criteria 1-3 is at least 2.5.
- At least 70% of submissions include an explicit failure condition for one assumption.
- At least 70% of submissions include one reasonableness or bounds check.
- Instructor grading time is stable (target: 4-6 minutes per notebook with rotating criteria).
Calibration process for faculty¶
- Blind-score 3 sample notebooks independently.
- Compare criterion-level differences.
- Align on one anchor example per score level (2, 3, 4).
- Re-score a new sample after calibration.