Adaptable assessment and feedback
These activities complement a master course and also work independently. They do not replace, reproduce, or set grades for Canvas assignments. Public feedback is intentionally available to self-study learners; it is unsuitable as a secret answer key.
Choose the evidence
For each selected module, name the repository outcome, an observable artifact, and the support allowed. Require a purposeful change plus an explanation when implementation is the goal. A copied report or generated explanation alone does not show the learner can reason about the workflow.
| Criterion | Developing | Meets the chosen outcome | Extends the outcome, optional |
|---|---|---|---|
| Framing and data | Task or units unclear | States task, observation, target, source, and relevant limits | Compares defensible alternative contracts |
| Implementation | Cannot reproduce the central step | Runs and modifies shared code; saves command/configuration and evidence | Adds a justified bounded extension |
| Evaluation | Reports a number without denominator/baseline | Uses the appropriate split/baseline and explains errors or missingness | Examines robustness without tuning final-test decisions |
| Communication | Claims exceed evidence | Connects named results to a decision and limitation | Adapts the explanation for another stakeholder |
| Reproducibility and access | Evidence depends on private state or screenshots | Provides readable code, text/table results, provenance, and an effective access route | Documents a reusable adaptation and its checks |
Select applicable criteria; early modules do not require a trained model or train/test split. Publish local weights, submission formats, resubmission rules, and any deadlines before grading. Optional depth must not be necessary to meet the core standard.
Discussions and independent study
Ask learners to make an initial decision, cite a specific observed result, and respond to an alternative interpretation. The solo route compares two defensible positions and explains the learner's choice. Grade reasoning against published criteria rather than message count or use of an AI assistant.
GenAI and authorship
Use the AI collaboration guide. Allow an equivalent no-assistant route. Where local rules permit assistance, request a short disclosure, what was verified, and one decision the learner can explain independently. Do not ask learners to publish private prompts, personal data, or entire conversation logs.
Accessibility
Accept code plus readable text/CSV or an equivalent effective format. Do not require color identification, screenshots, mouse actions, or a recorded face. An alternative still needs to demonstrate the same selected outcome. Consult the learner privately for interface barriers through the institution's established support route.
Instructor check
Try the task from a clean setup, confirm the public feedback fits the chosen data/version, check the intended outcome against the authoritative syllabus, and pilot the workload. Keep restricted assessments and student submissions outside the open repository.
Module mastery and independent transfer
Each module's guided page now identifies required changed code, checked evidence, a diagnosed failure and an explanation. Use its specific question and worked feedback to diagnose misconceptions. Public solutions support independent learning; use a changed task or explanation to establish authorship rather than hiding feedback. A diagnostic routes preparation and does not lower the shared outcome standard. Reading-only evidence supports interpretation; implementation requires execution and an explainable modification.