Research plan · Not yet conducted
Research plan: reviewable AI edits
A proposed study of understanding, accepting, and recovering from an AI edit.
The question
Can someone understand an AI-proposed edit, decide whether to accept it, and recover when the underlying file has changed?
Current evidence
The existing app exposes proposed changes as cards with Apply and Dismiss controls. Published engineering work documents preserving untouched values and validating before saving. UI-test captures use fixture snippets and simulated assistant responses. These establish interface behavior, not participant findings.
Prototype scope
Compare inline changes and proposal cards, then prepare an interactive flow covering working, proposed, rejected, applied, interrupted, and conflict states. The comparison and prototype are planned. Use synthetic snippet files throughout.
Visual review
Review hierarchy, typography, spacing, alignment, action prominence, contrast, keyboard focus, and narrow-window behavior. Retain the before-and-after screens and record the reason for each significant adjustment. External design critique is planned.
Participants and sessions
The proposed first round involves five consenting people who regularly use snippets or similar editing tools. Observe realistic tasks without teaching the interface. Ask participants to explain their thinking. Recruitment and sessions have not started. Recordings require consent and remain private.
Proposed tasks
Change a greeting while preserving its placeholder.
Handle a suggestion that changes the wrong snippet.
Resolve an edit after the file changes elsewhere.
What to record
Record completion without help, misunderstandings, accidental actions, and recovery. Keep observation separate from interpretation. A small qualitative study identifies issues to investigate; it does not establish population-wide success rates.
Revision and retest
Connect each significant revision to an observation, then retest the affected tasks. Publish anonymised findings and remaining limitations only after the sessions and revision exist.
Status · 7 October 2026
Planned. No participant findings, revised prototype, or retest results are claimed.
Method reference: NN/g: Usability Testing 101.
