ReviewBench
A prototype for comparing rubric decisions and reviewer feedback across programming assignments.
ReviewBench is private because it is tied to course workflow assumptions, but it represents the kind of teaching software that belongs in the project portfolio.
Problem
In large programming courses, multiple reviewers can apply the same rubric differently without noticing the drift early.
Implementation
ReviewBench models rubric items, reviewer decisions, and feedback notes so disagreements become easier to inspect.
What it does
- Surfaces rubric disagreement patterns before they become grading inconsistencies.
- Keeps reviewer calibration tied to concrete submissions and rubric items.