HUM-01 · Authorship after autocomplete

Design a study of voice change with AI assist; measure something falsifiable about style or revision.

Student’s own drafts or a teacher-approved public text set. AI-assist condition and unaided condition. One falsifiable measure declared before running (e.g. revision-type counts, type-token features, or a blinded peer “voice” ranking with a pre-registered rubric). Method appendix template. Explicit non-claim list about mind, meaning, and authorship required.

Measuring “sounds better.” Inventing a stylometry paper. Claiming the measure proves who the real author is. Hiding the AI condition.

Research question; pre-registered measure; results table; method appendix; essay section that interprets without mind-talk; non-claim list (what this does not prove about authorship or understanding).

CT: falsifiable style vs vibe. CR: pre-registration. CO: peer re-applies the measure to two samples. CM: method appendix. CZ: readers who would mis-credit the prose.

English/humanities teacher reviews the measure before data collection. Standing instruction: reject invented citations and reject “proves authorship.” School-supervised.

R1: one question, one measure. R2: appendix reproducible. R3: unaided baseline. R4: limits stated. R5: essay scorable. R6: non-claim list present.

(a) “The AI ruined my voice” with no measure. (b) A footnote to a paper that does not exist.

• Week 1–2: Question and measure approved; corpus/draft set locked; non-claim list drafted early.
• Week 4–5: Conditions run; table complete; peer re-measure.
• Week 7–8: Essay + appendix; non-claim list final; mentor review.

Lane A immediately (AP Research, IB EE, humanities senior paper). Natural precursor to HUM-04. No verified credit claim.