# Informed UX candidate retest

This is a short, informed retest of the revised game at `http://terminal.local:4173/`, separate from the blind R6 report. The original blind report is unchanged. I used only `cua_repl` for game interaction and inspected rendered UI attributes, not application source. No other reviews were read.

## Scope and infrastructure

The normal browser-entry call succeeded. No recovery loop or alternative browser technology was needed. The canonical game contained root's existing QA workspace. I dismissed the start-new confirmation rather than erase it. A brief attempt to use a reviewer query still showed the same save; I returned to the canonical URL. Root subsequently confirmed that the canonical save was disposable QA material but should be preserved and authorized using the empty Malawi case.

I resumed the existing QA workspace and inspected its Malaysia comparison and Russia completion views without editing authored content. In Malawi I opened MW3 and its source details, selected MW3 as Supporting evidence 1, prepared the two questions, and selected narrowing mode without applying a decision. No prose fields were changed. This was not a fresh full playthrough, screen-reader audit, mobile test, export test, or validation of a newly completed record.

## Verified revised behaviors

- **Decision state is now exposed.** In Malawi question 1, the rendered “Make a step more precise” element was a `BUTTON` with `aria-pressed="false"`. After selection it had `aria-pressed="true"`; the accessibility snapshot reported its selected state and changed “Keep it, with a reason” from selected to unselected. This resolves the specific missing-state observation from the blind review for the tested interaction. I did not audit every decision or keyboard-only operation.
- **Primary action names identify the question and action.** Examples observed were “Consider reading for question 1: Official figures have a history and a limit,” “Record retention reason for question 1,” and, after selecting narrowing, “Apply revised step for question 1.” Decision-choice text itself remains repeated across questions, but the submitted actions are substantially easier to distinguish.
- **Card and source controls identify their targets.** Observed accessible names included “Read the full card MW3,” “Source details for MW3,” and original-source links naming MW-B, MW-C, and MW-D plus their titles. Serbia's comparison cards similarly exposed RS card IDs, and the original RS-A source link identified that record. This directly improves navigation through repeated disclosure controls.
- **An empty explanation does not receive a misleading revision warning.** Starting with Malawi's empty paragraph, I selected MW3 for the board, then visited Connect and Hearing. Both displayed the empty “Your current explanation” field and zero-word count without the old “Your board changed. Review your explanation…” banner. This covers an initial card selection before drafting, not every possible empty-draft change sequence.
- **The paragraph-location promise is corrected.** The editor now says, “Edits carry through wherever this paragraph appears,” rather than claiming the paragraph appears in every stage. Comparison offers “Review my current main explanation.” I observed the control but did not exercise its complete return path in this retest.
- **Comparison prompts require a defensible comparison.** The “Test a similar advantage across cases” route asks the learner to name each result and period before comparing another actor's option. “Compare routes to a result” asks which specific result or step is comparable and states that a failed analogy can limit an explanation. The comparison writing prompt also asks for the comparable event/period, action, affected option, and a limiting or testing difference. These are improvements in task clarity without supplying a preferred historical explanation.
- **Incomplete-case guidance identifies a missing item.** Empty Malawi's Finish view and other stages said “Next required item: First explanation kept,” consistent with the visible inclusion check. The button name repeats that destination instead of the generic “Go there.” I did not test every missing-item destination.
- **Saving guidance has a working focus destination.** On Russia's existing record, “Go to saving and hand-in guidance” focused the “Keep or hand in your work” container. The explanatory copy clearly distinguishes local completion, readable HTML, editable JSON, and teacher submission. I did not download or submit anything.

## Observed completion contradiction in the existing save

Before any authored fields were edited, Russia's existing saved record displayed “Record complete for human review. Save the readable report…” as its next useful move. Finish also said “Record complete for human discussion · reasoning not assessed” and “Your local record is complete.” On that same screen, current Inclusion checks displayed **To review** for the competing account and for the paragraph/board review confirmation. The competing-account field was empty and that confirmation was unchecked.

This is an observed saved-state compatibility issue. I have not established its underlying cause or shown that a freshly completed candidate record behaves the same way. Nevertheless, the simultaneous messages contradict each other and can send a returning learner toward hand-in while current requirements remain missing.

Minimal fix: determine current completion from both the stored completion event and current inclusion checks. Retain the earlier completion timestamp as history, but label the record as needing review when current required items are missing. Do not infer meaning from prose or automatically supply a competing explanation. Root acknowledged this observation and requested a pause while implementing that fix.

## Post-fix recheck — verified after UPDATED

After root's UPDATED notification, I reloaded once, resumed the preserved workspace, switched to Russia, and opened Finish. No authored fields were edited. The same two current inclusion checks still required review, so this recheck tested the original contradictory saved state without filling its gaps.

The contradiction is resolved in the tested state:

- The next useful move now says “Next required item: A competing account of the same observation written (for human review).” It no longer offers a complete-record saving action.
- The Finish status now says “Earlier completion retained; current record needs review.” It retains the earlier timestamp, **Oct 30, 2026, 2:31 PM**, and explicitly asks the learner to review missing current items before finishing again.
- “Preview current record” now labels the output “Draft record · current inclusion checks not yet finished.” Its next line retains the same earlier completion timestamp and says current items still need review. The preview's self-check and inclusion record consistently show the missing confirmation and competing account.

This passes the requested saved-record compatibility recheck. It does not establish freshly completed-record behavior or all possible completion transitions. I hid the preview, captured the corrected status alongside the missing checks, and closed QA tab 74 successfully. The browser portion is complete.

## Screenshot evidence

I read the documented `cloud-shared-files` workflow, captured the revised entry once, visually inspected those exact bytes, and saved the same bytes as a JPEG. The synchronized main-container file was confirmed non-empty at 134,338 bytes. The image contains only the game's QA entry and saved-workspace summary, with no personal data.

- Browser path: `/home/oai/share/informed-ux-entry-1793403077226.jpg`
- Main path: `/workspace/scratch/informed-ux-entry-1793403077226.jpg`

A second viewport capture documents the post-fix status, with both missing checks and the retained timestamp visible. It was captured once, visually inspected, and saved using the same documented workflow:

- Browser path: `/home/oai/share/informed-ux-completion-recheck-1793403277399.jpg`
- Main path: `/workspace/scratch/informed-ux-completion-recheck-1793403277399.jpg`

## Remaining limits

The revised labeling, pressed state, empty-draft guidance, comparison copy, and saved-completion correction were verified as described. The one reload also retained Malawi's selected card and narrowing mode, but it was not a broad persistence test. Full narrowing/replacement application, stale-summary behavior after successive changes, new-record completion, all focus-return routes, narrow layouts, and exports remain outside this short retest. The report does not turn these unavailable checks into failure claims. The original blind R6 report remains unchanged.
