# S4 independent first-use pressure test

**Persona and scope.** I simulated an average 18-year-old with confident general political opinions, weak research habits, and a tendency to collect confirming facts. This is a role simulation, not a human usability sample or a validated timing study. I used only the visible game at `http://terminal.local:4173/?reviewer=S4`, including its help, glossary, source records, and final preview. I did not inspect source code, hidden answers, design history, or other reviews. I played Serbia, Just for fun, solo, and chose Russia for comparison. I reached **“Hearing filed · interpretation awaits discussion.”** The advertised approximately 55 minutes remains a design target; this run cannot establish human completion time.

## First understanding and actions

My simulated first understanding was: “Pick a government, collect three facts, explain how it stayed in power, and defend that explanation.” The landing page's “choices change the analysis, not history” made the historical premise clear. I initially treated “two anchors and one complication” as two strong supporting facts plus an awkward fact to explain away. The word “anchor” also names the country, which briefly blurred two different jobs.

I selected **Unequal access and pressure**, recorded **Fairly confident**, and went straight to evidence without opening help. Serbia's introduction usefully specified a parliamentary election and explicitly said Vucic was not a parliamentary candidate. I understood the known result before reading cards. I skimmed the table, read RS3, RS4, and RS5 in full, and selected RS3/RS4 as anchors and RS5 as the complication. I left RS1, RS2, and RS6 unopened. That was plausible confirming behavior: television advantage, paid work linked to voting, and unsuccessful challenges all seemed to tell my preferred story.

My exhibit notes actually included **“Government control of television gives them propaganda power,” “People were paid to vote for the government,”** and, under complication, **“Opposition protested but the unfair system stopped them changing the result.”** Those are student overstatements or interpretations, not established game facts. The UI accepted them without semantic feedback, consistent with its stated limits.

## Prioritized findings

### 1. High: the complication can remain another confirming fact

**Observed:** My RS5 explanation supported my original account instead of unsettling it. It survived through filing, with **“Included A complication reason recorded; anchor notes are optional.”** The selected questions concerned national-scale effects of RS3/RS4 and the distinction between a parliamentary contest and presidential reelection. Neither asked me what RS5 made less certain. The “Why these questions?” explanation said the second prompt was selected because my exhibit included RS5, but the proposed claim was about Vucic and the government's rationale. As a first-time student, that connection was difficult to see.

**Simulated response:** I could feel I had met the “complication” task by naming protests, while retaining the same totalizing explanation. This is a scaffold gap, not evidence that the advertised structural checker malfunctioned.

**Fix:** Beside the complication field, ask “Which sentence or step becomes less certain because this is true?” Give a short neutral example. During the hearing, explicitly revisit the selected complication and offer a human self-check: “Have you used this only as another reason your opening is right?” Preserve the freedom to defend the original account; require the student to identify the tension first.

### 2. High: useful narrowing can leave contradictory current claims behind

**Observed:** The first question genuinely challenged my leap from monitored coverage and one operation to national voter effects. I selected the supported reading, chose **Narrow a step**, and changed the option to: “Opposition candidates could organize, but had less favorable ordinary television news coverage; we do not know how many voters changed their minds.” The UI clearly reported **“Board step changed.”** I then revised the main paragraph myself.

The final preview nevertheless retained my current action **“Controlled television and paid people to vote for them [RS3]”** and consequence **“Why the government won and the opposition could not change anything [RS5].”** The revised changed-option step still carried my earlier RS4 attachment, even though the new wording primarily concerned television. The exhibit notes also retained the earlier overstatements. These were presented as current causal connection and notes, alongside the narrower final paragraph.

**Fix:** After applying a revision, present a small reconciliation task: review the adjacent step, its evidence attachment, the exhibit note, and the final paragraph. Do not rewrite student prose automatically or pretend to grade its meaning. Clearly distinguish retained historical versions from claims still being filed as current. This would add intellectual pressure rather than remove rigor.

### 3. Medium: source-reading credit depends on the route used

**Observed:** I opened **“Reopen the source records behind this question”** and read the displayed RS-A and RS-C records and their limitations. At filing, **“At least one anchor-file card source record opened”** still said **To review**. My first filing attempt produced **“1 inclusion items still need review. Your draft can be exported now.”** I resolved it by using **Reopen RS3 & sources**, then returning to filing. The check became Included and filing succeeded.

**Simulated response:** “I already read those sources; what counts as opening one?” The check is accurately narrower than “read a source,” but that distinction is not educationally obvious.

**Fix:** Count viewing a source record wherever it appears, or explain the required route and provide a direct review link within the unfinished check. Keep the distinction between opening a record and reading an original source explicit.

### 4. Medium: correct options are easy to recognize without doing the hard work

**Observed:** The first question contrasted “Every voter...” and “court conviction...total number...nationwide” with the carefully qualified source-scope answer. The second contrasted obvious office/source errors with a fully qualified answer. I answered both correctly on the first attempt, as well as the optional Vale transfer question.

**Simulated response:** I could pick the least absolute answer quickly. The real learning came from writing the narrower step, not from selecting the radio button. **Try the strongest rival account** opened a blank optional phrase; I supplied “Maybe people really supported the government, and the opposition was not convincing enough.” It did not make me test that against the unused cards. My vague **“More unbiased polls and media reports”** also satisfied the next-evidence inclusion check.

**Fix:** Use more plausible near-miss readings, then request the missing observation in plain language: whose exposure, which behavior, over what period? Offer a contrast template for next evidence: “To distinguish X from Y, I would need to observe Z.” Keep human judgment explicit.

## Comprehension, appeal, and effort

The cream paper, dark green navigation, restrained red accents, serif headings, and persistent exhibit made this feel credible and attractive. The result table was readable. The visible final record looked like something worth keeping. I enjoyed choosing evidence and seeing a narrower explanation emerge; there was less game-like payoff during repeated form filling. Long pages, repeated source-scope reminders, and maintaining several overlapping versions were the tiring parts.

Exact phrases that demanded extra effort were **“recorded backing,” “effective opportunities,” “bounded account,” “mechanism,”** and **“What does it unsettle?”** The glossary clearly explained observation/inference, anchor/complication, source limitation, and turnover, but did not define bounded or mechanism. Help's invented library-hours example was excellent; I only found it after the hearing. A compact version near the first causal chain would have helped earlier.

What software checks was unusually clear. The hearing said **“No software has read your prose for meaning.”** Filing explained that inclusion checks count items, lengths, and IDs, and cannot judge truth or logic. I did not interpret my supported source readings as a browser grade for the thesis.

## Sample completion and verdict

The filed account was 78 words, beginning **“Serbia's governing list had an unfair advantage through propaganda and pressure”** and ending with uncertainty about persuasion, the single investigated operation, and continued protests. My 38-word Russia comparison used RU2 prison sentences against Serbian campaigning/protest options. I chose Russia partly because it confirmed my prior political intuition; the capsule's reported support and no-pressure survey figures provided useful counterpressure, but I did not investigate those cards.

**Verdict:** Finishable without designer help, visually strong, and capable of improving a confident student's source claims. No unresolved completion blocker occurred. The most important improvement is to make the selected complication and all current causal claims answerable to the revision, so a student cannot complete the careful paragraph while leaving the larger confirming story untouched elsewhere in the record.
