# S2 informed retest — revised candidate

This is an **informed retest**, not a fresh blind student session. It does not establish human timing or learning. The initial S2 report is frozen and unchanged. Scope: rendered UI only; Malawi, For class, Solo; repeat MW3/MW5/MW6 and a comparison. No source-code inspection.

## Environment and result

The supplied directory URL `http://terminal.local:4173/v2/?reviewer=S2R` displayed `Cannot GET /v2/`, including after one reload. Root supplied the corrected exact URL `http://terminal.local:4173/v2/index.html?reviewer=S2R`, which loaded the revised landing page. Treat the first route as QA routing, not game behavior.

After selecting Malawi and Enter, two accessibility captures and a screenshot hit browser protocol timeouts. A single reload recovered the tab and showed a saved Malawi hearing. **Resume saved hearing** returned to the opening without loss. Testing paused at root's request, then continued as the only active browser reviewer. Subsequent browser calls succeeded. These protocol errors are not evidence of a game defect.

**Completed:** courts opening recorded; MW3 Supporting evidence 1, MW5 Supporting evidence 2, MW6 Complication; all three full cards and Source details read; one narrowed consequence applied; unchanged prose inspected and manually revised; two source-practice readings and board decisions recorded; required rival and evidence request completed; Russia comparison written using RU1; both self-checks confirmed. The final state displayed **“Hearing finished · explanation ready for discussion.”** No product blocker prevented completion.

## Reading, language, and orientation

- The landing page uses **main case**, **evidence board**, **supporting cards**, and **short explanation**. These are clearer than anchor/exhibit/account for a skimming learner.
- The opening displays the final deliverable early: one 60–85-word explanation using both supporting IDs and a source limit, one 30–45-word comparison, and short phrases naming another explanation and needed evidence.
- **Next useful move** initially says to choose the first explanation, then changes to **“Open the full text and Source details of the evidence you use.”** Its specific instruction is a substantial improvement.
- The evidence screen visibly says **“Before you explain: open the full text and Source details of the cards you use.”** It also explains that a complication can identify another explanation, a source limit, or a missing step and need not contradict everything.
- The table warning now says **“Check who or what each percentage is out of, and when it was recorded.”** This conveys the denominator instruction in ordinary language.
- The expandable **People, groups, and terms in Malawi** defines MEC and the relevant actors. **“Majority in this case”** says **“More than half of valid presidential votes. Having the largest share under 50% is a different rule.”** This directly resolves my original majority confusion.
- The same panel explains that the two official 2020 publications are the commission's later report and polling-station results; their exact totals differ slightly while one-decimal shares match. This is available before the hearing asks about it.
- Live reading counts now work in the observed flow. Opening MW3 full text and details changed the display immediately from 0/6 and 0/5 to **1/6 full cards and 2/5 source records**. After opening MW5 and MW6 text/details, it showed **3/6 and 3/5**. The source count is unique records, not cards, which is consistent with MW3 opening MW-B and MW-C and MW6 adding MW-E.
- The board adds **“Shared source: MW-C”** and warns that different steps are not automatically independent checks of a claim. Useful precision, especially because my repeated choice uses two cards from the same administrator report.

## Narrowing-to-paragraph handoff: passed

I deliberately reused the original rough 66-word paragraph, including **“This helped Chakwera win instead.”** My connection's consequence began **“A different winner in June 2020.”** In the first practice question, I chose the reading that separates a legal opening, implementation, and winner. I selected **Make a step more precise**, selected **What might that help explain?**, and applied:

> A new vote became possible; explaining its winner needs other evidence

The interface immediately displayed **Before** and **Now**, alongside my reason naming MW3. It then offered **“Your board changed. Review your explanation below.”** That button moved directly to and focused the paragraph editor. A screenshot confirmed a pink handoff panel directly above the editor, saying **“Your board changed. Review your explanation below; we have not rewritten it.”** My original paragraph was visibly unchanged.

This is a substantial fix to my original confusion. The screen makes clear what changed, what did not change, and what I should do next. The current paragraph is large and usable. It is the same paragraph in connection, hearing, and finishing, rather than a separate working-view field that feels like another deliverable. I revised it myself and the count updated to 67 words. The handoff remained visible until I completed final review, then disappeared. That was coherent rather than an apparent failure to notice my edit.

The dropdown now includes each step's actual current wording. This is much clearer than choosing an abstract **Changed option** or **Possible consequence** without seeing the sentence being edited.

## Hearing and finishing: passed with remaining density

The revised distractors are materially better. They now ask whether the ruling records show changed preferences, reasons for institutional compliance, or the opening of a contest. The answer is less discoverable merely by rejecting an absurd claim about judges appointing a president. The figures question identifies the two publications in ordinary language rather than starting with unexplained “two series.”

Both question cards are still displayed together. Each still requires an answer, feedback, a board response, and a reason. Applying a precise step adds a dropdown, current wording, new wording, before/after feedback, and another review action. Collapsing extra reflection and source details helps, and calling the multiple choice **practice** is honest, but this remains the longest and busiest stage. A future one-question-at-a-time option could help without removing the evidence or reasoning.

The new required rival field gave me a concrete next task: **“The joint Chakwera–Chilima ticket may explain the winner better than the court decision.”** The evidence request was **“Voter evidence separating the joint ticket from turnout changes, rather than just comparing the final percentages.”** It was eventually present in Hearing, Compare, and Finish & save. The final page correctly carried the rival, request, main paragraph, comparison, and applied decisions.

I deliberately tried **Finish this hearing record** before checking the two review boxes. It did not finish. It reported **“2 items to review”**, named the source-limit confirmation, and focused its **Go to this item** control. That link focused the actual checkbox. After I checked both source-limit and board-paragraph review confirmations, every inclusion item became Included and finishing succeeded. This is a clearer recovery path than the original bare count of missing requirements. The interface correctly explains these are student confirmations, not software judgments about meaning.

## Concrete remaining problems and limits

1. **Medium: hearing density remains.** The prose is clearer, but two full tasks, their decision controls, the new short responses, handoff, and paragraph still form a long page. This retest shows completion is possible; it does not show an uninformed skimmer will patiently use every part.
2. **Low: selected-card reading guidance and the final gate differ.** The evidence screen says to open full text and Source details for cards used. The inclusion check requires only **at least one main-case source record opened**. I read all three selected cards in this pass, so I did not test bypassing that advice again. This is a design tradeoff, not a discovered blocker. A per-selected-card reading reminder would make the intended next action more concrete without equating opening with reading.
3. **Low: narrowing reports clearing an attachment even when none existed.** I left all optional source attachments empty. Applying the narrowed consequence still said **“The old source attachment for this step was cleared.”** Use conditional wording, or “This step now has no source attachment.” No work was lost in this case.
4. **Low: the top next-action copy is sometimes generic.** After finishing successfully, it still said **“Review the missing items below, check your explanation, then save your record.”** There were no missing items. An all-complete/finished state would make this panel match the result.
5. **Unconfirmed input-loss observation:** in one batched browser interaction I entered a rival, attempted to fill the evidence request, and immediately navigated to comparison. The rival persisted but the request appeared blank in Compare and on returning to Hearing. I had not captured the request's value before that first navigation, so this is not proof that the app lost valid input. I reentered it, visibly verified its value, then navigated; it carried correctly to Compare and Finish. Reproduce fast entry followed by navigation before classifying this as an application bug. Recovery was straightforward.

## Actual finished work

**Main paragraph — interface count 67 words:**

> The courts stopped the first declared result from being final by ordering another election (MW3). Parliament and the election commission then made a new vote possible (MW5), giving voters another choice. These records help explain the new contest, but not why Chakwera won it. The joint ticket and changed turnout could also matter. The official 2019 figures were set aside, so they cannot reconstruct a fair count.

**Comparison — interface count 39 words:**

> In Malawi, courts let challengers get another election. In Russia, the commission refused to register Nadezhdin (RU1), so voters could not choose him. This suggests institutions can open or close a route, but does not prove he would win.

**Verdict:** the original board-versus-paragraph problem, missing Malawi majority explanation, stale reading counters, and vague missing-item recovery are addressed in this observed route. The next action and revision handoff are now clear. The main remaining usability concern is hearing density; there is also minor state-dependent copy to polish. This informed pass supports functional improvement, not a claim about first-use learning or a 55-minute human lesson.
