Reviewer Walkthrough

1. Start at the fictional Cedar workspace. Follow the routing guide and context card through delivery, owner, adoption, and hypothetical benefit. Inspect Context card and check its shared lookup plus re-entry atoms.
2. Read compatible baseline B1: 200 minutes/week for Cedar’s 20 handoffs, Sept 7–11. Observation O1: 140 minutes/week for the same exposure, Sept 21–25. The descriptive change is 60; read the causal caveat.
3. Select Missing baseline. The raw 140 remains; no recognized change is available. Select Incompatible baseline and explain why Elm’s 30 handoffs cannot supply Cedar’s starting point. Return to Compatible baseline.
4. Inspect the overlap section: routing guide 40 + context card 30 − shared lookup 10 = 60 hypothetical minutes/week. Read each full-use assumption. Actual adoption is lower. The forecast and observation are separate; their equal values are coincidental.
5. Choose Investigate, owner, rationale, and open questions. Preview decision. Check versions and references, then Cancel or Escape. No review is recorded and focus returns to the preview control.
6. Preview again and Confirm review. Inspect its preserved evidence in Decision record. Change the current scenario to Missing baseline; the original review still contains B1 and the compatible descriptive change.
7. Export report. The JSON contains current evidence, complete historical review evidence, owners, questions, references, versions, assumptions, and causal limits. It needs no live application to interpret those records.
8. Preview withdrawal, enter a reason, and cancel once. Then confirm a withdrawal. The original review remains, with the appended withdrawal reason.
9. Preview reset. Check the counts and Export before reset if needed. Cancel once; history remains. Confirm reset to recover a new compatible ledger. Twelve reviews disables further reviews with an export-and-reset recovery route; it never evicts history silently.

Meaningful recovery exercises

With an invalid saved record, the page announces preservation and rejects attempts to save a review over it. Export current work, preview reset, and confirm only when the raw record remains unchanged. A competing tab that changes saved bytes after preview causes confirmation to reject; Load saved state and re-review. If storage cannot be read, reviews remain memory-only and reset is unavailable; refresh or close may lose unsaved work. Restore access and load saved state before proceeding.

These are reproducible software scenarios, not evidence of a completed human usability study.