# R6 — Independent first-use UX and accessibility review

Review target: the frozen game at `http://terminal.local:4173/?reviewer=R6`, reviewed only through its current browser UI. Main case: Russia. Second case: Malawi. No source code, prior reviews, or other reviewers' findings were consulted.

## First understanding, recorded at entry

I understood this as a historical explanation exercise, not a simulation in which I change an election outcome. I would choose a main case, investigate how an advantage changed another actor's options, select two supporting observations and one complication, take questions, and test the explanation in another country. The outcomes were explicitly known. The three-space board limited what I used, not what I could read. Class mode and Solo were already selected; Russia was the default main case. Progress would save on this device. The roughly 55-minute estimate and absence of a timer made the task feel manageable.

I did not initially know the exact final writing lengths. The Opening stage subsequently made them concrete: a 60–85-word explanation, a 30–45-word comparison, and short phrases for another explanation and needed evidence. This was a useful early statement of the deliverable. The introductory language also correctly discouraged ranking different countries on a common vote or seat score.

## Scope and confidence

I verified entry and stages 1–5. I recorded the “Restricted challenge” opening, opened full cards and source details for RU1, RU2, and RU6, and built that board. I authored a three-step connection and a 69-word paragraph. I considered both practice readings, applied one narrowing decision, replaced RU2 with RU5, reopened RU5 and its source details, and revised the shared paragraph to 72 words. I filled the alternative-explanation and further-evidence fields and selected Malawi as the second case.

The last confirmed comparison screen still had an empty comparison. A batch intended to open MW3, enter a comparison, and visit stage 6 encountered a browser `DOMSnapshot.captureSnapshot` protocol timeout. The single reload/Resume attempt timed out too. I closed the QA tab and stopped. Thus I cannot verify stage 6, a completed-state screen, exports/imports, or persistence across reload. The timeout is a browser-tool limitation, not an observed game defect. An initial browser-creation protocol error had recovered on one retry.

This was a desktop browser review, including screenshots around 1348 × 926, accessibility snapshots, and limited read-only inspection of rendered control attributes. It was not a full keyboard-only completion or a screen-reader audit. The advertised browser API offered no viewport-resize control, so I did not test a narrow viewport or physical phone.

## What worked in live use

The first Tab exposed “Skip to the workspace” with a strong visible focus outline. Enter activated it, and the next Tab reached “How are you playing?” rather than making me traverse the header. Entering the game and changing stages consistently moved focus to the stage heading. This gave a dependable orientation point.

“How to play” and “Glossary” opened with focus on their explicitly named close controls. Escape closed both and restored focus to the launch button. In the glossary, Tab reached “Find a term”; typing “complication” reduced the list to one definition and updated the result count. A draft survived a help visit and return. These are valuable foundations to preserve.

The current paragraph remained the same between Connect and Hearing. Narrowing changed the selected connection step while leaving the paragraph intact, and the UI explicitly said it had not rewritten it. “Your board changed. Review your explanation below.” focused the paragraph field directly, with a clear outline and the change summary nearby. Reopening RU5 focused the card heading; “Return to Take questions” returned to the paragraph. These interactions support revision without doing the interpretation for the learner.

The replacement confirmation named both card IDs and the affected slot. After acceptance, RU5 occupied Supporting evidence 2, its reason appeared on the board, and “Earlier board choices” increased to 1. I did not expand that archive, so its contents are unverified. Trying to apply a decision before considering a source reading produced an explanatory message and kept the entered replacement selections and reason.

I also checked the rendered Russia table after the accessibility snapshot represented its headers as cells: it actually uses column `TH` elements with `scope="col"` and a row header with `scope="row"`. I would not report missing table headers. Source limitations, denominator distinctions, and the separation between practice feedback and human judgment were consistently explicit.

## Findings and minimal fixes

### 1. Decision selection is visual but not exposed as a selected state — Medium, observed

In Hearing, click “Make a step more precise.” Its fields appear and its rendered button receives the `chosen` class. Read-only inspection found no `aria-pressed` attribute or alternative selection role; the accessibility snapshot continued to announce an ordinary button. The same set also contains “Keep it, with a reason” and “Change supporting evidence.”

A learner navigating controls with assistive technology can discover the resulting fields, but cannot directly ask which of the three mutually exclusive decisions is selected. This is especially costly after leaving and returning to a question.

Use a labeled native radio group for each question, or expose true/false pressed states on all three buttons. Keep focus stable when the choice changes. Verify by keyboarding through all decisions and checking that exactly one selected state is announced for the correct question. This changes access to the choice, not its intellectual difficulty.

### 2. Repeated control names lose their context in control lists — Medium, observed; assistive impact inferred

Evidence cards repeat “Read the full card,” “Source details,” and “Open original source.” Hearing repeats “Consider this reading,” “Apply to my board,” and the decision controls for two questions. The board also repeats “What does it show?” and “Why use this evidence? optional phrase.” The surrounding headings provide context in sequential reading, but the controls' accessible names do not carry the card or question identity.

This makes a button list or form-control jump ambiguous and increases the chance of changing the wrong question or slot. The card-use buttons already demonstrate a better pattern: “Use RU1 as Supporting evidence 1” is unambiguous.

Retain concise visible text but add the card ID, source ID, question number/title, or board slot to each accessible name. Verify the resulting names in an accessibility control list. Do not add hints about which evidence or reading to choose.

### 3. The paragraph handoff shows an older change after a later replacement — Medium, observed

With RU1/RU2/RU6 on the board, I first narrowed “Whose options changed, and how?” and applied it. I then used the second hearing question to replace RU2 with RU5. The local applied message correctly said “RU2 removed; RU5 put on the board.” However, the shared banner immediately above the current paragraph still presented the earlier narrowing's “Before” and “Now” wording. It did not show the replacement as that banner's change. The paragraph still cited RU2 until I revised it myself.

Leaving the paragraph unchanged is appropriate. The problem is the handoff's incomplete orientation: its generic “Your board changed” heading can make the old before/after pair look like the relevant latest change. A learner may review the wrong dependency and miss a stale citation.

Show either the latest change or a short list of all changes awaiting paragraph review, including “Supporting evidence 2: RU2 → RU5.” Link each change to the paragraph and relevant step. Keep human authorship and source-limit requirements intact. Verify the sequence narrow → replace → return to paragraph, and the reverse order.

### 4. “Next useful move” does not track the visible state closely enough — Medium, observed copy; destination behavior unverified

Before preparing questions, Hearing said “Use the practice feedback…” even though there was no feedback and the main control was “Prepare my two questions.” Later, reopening RU5 returned to Evidence with a full three-card board, a completed connection, and an existing paragraph, yet the banner still said “Put two supporting cards and one complication on your board, then connect an action to a changed option.”

The instruction is valid in general but inaccurate as a personal next action. This can suggest work is missing or unsaved. I did not test “Go there” in these states and cannot claim it loops or misroutes.

Derive the banner from the first actually incomplete action: prepare questions, consider a reading, review a changed citation, or proceed to comparison. Give “Go there” an accessible name that identifies its destination and focus that destination. Verify the banner after each relevant state transition and again when returning to a completed earlier stage.

### 5. The shared-paragraph promise overstates where it is available — Low, observed

The editor says “This same paragraph appears in every stage.” It was available in Connect and Hearing, but I did not find an editor in Opening, Evidence, or the confirmed Malawi comparison view. Comparison instead displayed the action/changed-option chain, which still contained my earlier court-imprisonment wording after the support-card replacement.

That mismatch raises a navigation question precisely when the learner is testing and revising a claim. Change the promise to “Edits here update your current explanation across the workspace,” and provide a clear “Review current explanation” return near the comparison writing area. Preserve the distinction between connection steps and the paragraph; do not silently synchronize their content. Verify that opening the editor preserves comparison work and returns to the same place.

## Reading burden and completion follow-up

The desktop layout is readable, with a useful board alongside the main task and progressive disclosure for full cards and source records. Hearing nevertheless becomes a long reading surface: two complete practice questions, decision forms, applied summaries, short responses, a change banner, the paragraph, and decision history. The identical labels and state-inaccurate next-action text compound that burden. On a narrow layout, distance between these regions could be greater, but that remains an inference.

Keep all six cards, original-source access, alternative explanations, the second-case test, revision, and human review. Reduce navigation effort through named sections, accurate next actions, and compact summaries of already considered questions that remain reopenable. A follow-up should verify narrow-screen order, focus visibility when regions expand, reload/Resume preservation, stage-6 links to missing work, and whether a structurally completed record stops offering already completed steps. I make no claim that those untested behaviors currently fail.
