Case C — faculty notes

Cost, scaling and equity notes for the redesign. Setting is the fictional PSY 250 at Lakeside State University — a fictional US public university preparing for ADA Title II digital accessibility compliance — ~120 students in tutorials of 20, one coordinator and three TAs.

Grading time

Component	Original	Redesign	Notes
Per-report marking	~45 min × 120 (≈ 90 h/term)	~13.5–14.5 min per student total (≈ 27–29 h)	Rows below sum to the total
Notebook	—	~3 min (checklist + 1 spot-check of two entries)	The spot-check looks for dated, problem-centred entries
Memo (500 words)	—	~4 min	Criterion 3; the Q&A covers it again per student
Report	(the 45 min above)	~4–5 min	Marked on structure/accuracy/interpretation; analysis correctness is re-checked only where the memo and Q&A disagree with it
Q&A	—	~2 min per student (10-minute team session in tutorial)	Three anchors per member: method / result-or-limitation / next step
Twist registration	—	~0.5 min per student (the coordinator reviews all 24 team registrations in one ≈ 1 h batch)	Amortised: ≈ 2.5 min per team

Where the grade weight comes from: nowhere new — the redesigned components’ raw weights sum to 38 points, and the course handbook caps the task at 30% of the final grade, so every component scales by 30/38 (shown in the rubric’s gradebook mapping). The original task also carried 30%.

Net: roughly 30% of the original marking load (≈ 27–29 h vs ≈ 90 h), with better assurance — because the Q&A replaces the impossible task of verifying every student’s statistics by re-reading prose.

Scaling to 200+ students

1. Team sessions, not individual orals. One 10-minute team Q&A inside each tutorial covers 5 students ≈ 2 min/student. Four tutorials per TA ≈ two Q&A sessions per TA per week for two weeks.
2. Notebook spot-checks, not audits. Check every notebook against the checklist (~3 min), deep-read two entries from every fifth team. Q&A performance drives who gets looked at more closely.
3. One shared protocol, 24 team twists. The coordinator reviews twist registrations in one batch against three approved-variation rules (same measures; no new populations; no sensitive topics). Anything outside → team uses the standard variant (still fully gradeable).
4. Report marking stays conventional — TAs calibrate on three reports per section before marking week.
5. If TAs are cut: fold the Q&A to 6 minutes per team (one question per member), and grade the memo by checklist. Never cut the notebook spot-check — it’s what makes the Q&A honest.

Accessibility and accommodations

• The Q&A has three equal-weight modes (live / recorded / written, ~250 words per question), chosen once at term start. Students with communication disabilities, anxiety or speaking accommodations are never required to speak — no diagnosis disclosure needed.
• The data-collection step has a provided-dataset fallback for anyone who cannot collect (remote students, care duties, disability that makes approaching strangers costly, visa restrictions on recruitment). The fallback dataset is written to have the same shape of messiness (missing responses, one skewed variable) so the notebook criterion stays meaningful.
• Notebook format is flexible — text, bullets, audio-transcribed, screen-reader-friendly templates provided. Grading criteria never depend on format.
• Team roles are chosen, not assigned, so a student’s accommodation (e.g., not handling recruitment) can be a legitimate role; tagged notebook entries make each member’s contribution explicit.
• All briefs, the question bank and templates ship in accessible formats (same baseline as this site: README accessibility section).

Multilingual students

• The Q&A’s written mode and the notebook’s format freedom remove the oral-English penalty.
• AI language help in the report is explicitly allowed with disclosure (AIAS level 3) — multilingual students get to write in their strongest language, get editing support openly, and are assessed on analysis and defence, not idiom.
• Memo and notebook are graded on meaning: a plain, precise entry beats fluent vagueness.
• No detectors anywhere in the marking chain (Liang et al. 2023; Weber-Wulff et al. 2023) — multilingual students are the population detectors harm most, and this redesign gives them no reason to be near one.

Common pushback, with answers

• “Letting undergraduates collect data is a governance nightmare.” The shared topic runs on the department’s standing class-study approval; team twists are constrained to pre-approved variation (same measures, no new populations). One coordinator reviews 24 registrations in an hour.
• “The report is now AI-tolerant — that lowers standards.” The report stops being the assurance mechanism (the memo, notebook and Q&A are). Grading the prose while knowing AI can write it was the false standard; this one is honest about what each part certifies.
• “Teams mean free-riding.” Tagged notebook entries + per-member Q&A + role choice make contribution visible. Not airtight — but the original’s individual report had its own unenforced fiction.

What we would measure after the first run

• Grading minutes per component vs. the estimates above.
• Fallback-dataset uptake (if >15%, collection design needs revisiting).
• Q&A mode distribution and score parity across modes (watching for a written-mode gap — the equity canary).