Learner pilot, workload and support record
This instrument supports independent review and course adoption. No learner-study results are recorded in this edition. Recruit consenting learners with no programming, programming without ML and prior ML experience. Use institutional processes; keep identities and submissions private.
For each route, observe acquisition, first output, one guided construction, a broken example and an independent transfer. Record preparation time, reading time, setup/download time, execution time, practice time, help requests, stopping point, misconception and successful independent explanation. Distinguish elapsed downloads from active study. Ask what instructions were missing and which terms needed explanation.
| Decision | Evidence to collect | Revision trigger |
|---|---|---|
| Preparation route | Changed diagnostic example solved unaided | Repeated basic syntax/path barriers |
| Pacing | Total time by starting experience, plus current assignments | Workload exceeds the instructor's course budget |
| Learning | Required construction and fresh transfer artifacts | Expected output achieved without an explainable method |
| Advanced depth | Declared experiment and defended tradeoff | Additional work is repetitive rather than deeper |
| Access | Complete keyboard/AT/zoom tasks in real interface | Any blocked required outcome |
Publish class support contact, response expectations, escalation and recovery arrangements in Canvas before assigning work. Do not copy an unconfirmed contact or invent a service promise in the OER. During the first offering, review anonymized patterns weekly and adapt core/optional scope. Keep outcome standards stable while increasing assistance and planned preparation time.
Use the accessibility task log for browser/OS/assistive-technology versions, task, observed barrier, severity, repair and retest. Human screen-reader and student testing remain required. Automated checks establish a narrower scope.