# T1 — low-preparation student: partial first-use browser review plus cold read

This is an AI learner simulation, not a human trial. The role is a willing first-year community-college student who skims and has little country or methods vocabulary. No code, designer guide, previous Hearing feedback, other reviewers, hidden answers, or outside research was used. The earlier wrong-game attempt is excluded; its separate blocker report is unchanged.

## Scope and initial understanding

The repaired URL `http://terminal.local:4173/?reviewer=T1` showed **The Hearing · Who Holds the Advantage**, **Who holds the advantage?**, and Russia/Malaysia/Malawi/Serbia before any game action. Russia was the selected main case. My own-words first reading was: “This is for figuring out why one side has an advantage in an election. I think I have to choose a country, pick two cards that help my explanation and one that causes a problem, then answer questions about it. I might learn how to back up an explanation instead of just saying an election was unfair.”

Actual interaction reached Stage 2 only. After a browser failure, the parent authorized a text-only review of the remaining Russia/Malawi screens in **Player_Screens.txt** and **How_to_Play.txt**. Everything below labels that boundary. There is no completed hearing record, interactive comparison, verified revision, timing result, learning-effect finding, or mastery claim.

## Actual first-use actions and difficulties

I kept the For class/Solo defaults and left the optional name blank. In Stage 1 I opened **People, groups, and terms in Russia**, chose **Restricted challenge**, and recorded it with **A first idea · I expect to test it**. In Stage 2 I read the six summaries, opened RU1 and RU2's full text and source details, and confirmed RU1 as Supporting evidence 1. My intended board was RU1/RU2 supporting restricted challenge and RU4 complicating it with reported fairness and lack of personal pressure. RU2 selection was not confirmed; RU4 was not opened or selected.

The introduction's known outcomes and fictional research role helped: I understood I was changing an explanation, not election history. The definition of **incumbent** and the country-specific people panel supplied necessary background. I did not use the general glossary or How to play during the browser portion; nearby country help felt more useful.

I skimmed the future writing specification: first idea, current paragraph, 60–85-word final explanation, 30–45-word comparison, competing account and needed evidence were a lot to retain before handling a card. The six stage buttons and **NEXT USEFUL MOVE** oriented me better than the suggested pace. I initially treated “complication” as evidence against everything I thought; Stage 2's “a source's limit, or a missing step” corrected that.

The RU0 **Who answered** column helped distinguish survey replies from an official election result. My first instinct on seeing 80%, 94%, and 78% was to treat them as general election facts; the concrete labels corrected it. **Denominator** was less familiar. I did not open the table's source limitations, so full table comprehension was not demonstrated.

“RU1–RU6 = cards · RU0 = table · RU-A = original source” needed a reread. Two numbered cards sharing one source is easy to mistake for independent confirmation. RU1's **Survey** label initially seemed mismatched to a registration refusal; its full text clarified that the report also described limited recognition and 3% hypothetical support. “Does not adjudicate the registration grounds” was harder than the simpler idea that this source cannot show an unrestricted campaign's result. RU2's “before/after loss of national electoral capacity” similarly needed translation into what organizing became possible and what vote effects remained unknown.

The attempted RU2 support click returned **Input.dispatchMouseEvent timed out**. Fresh accessibility state showed no change. The one authorized reload and one fresh capture both returned **Accessibility capture changed before it was committed**. I stopped and closed QA tab 79 (`t1verified`). Those are browser-operation limits, not demonstrated product defects. No ordinary learner action was abandoned because of a game rule; later tasks were interrupted by this blocker.

## Text-only comprehension of the remaining stages

The library example in Stage 3 clarified the central skill: name who acted, what changed, and whose option opened or closed; an attendance list cannot prove why every visit happened. The optional inference step made sense. I could still overreach from “a candidate could not enter” to “that caused Putin's victory” without the repeated source-limit prompts.

Stage 4's two supplied claims made important distinctions understandable: most respondents reporting no personal pressure does not describe opposition organizers' opportunities; a low January measure does not establish the ceiling of a campaign that never happened. I could choose the cautious readings from the text. That is recognition, not proof I could construct an equally careful argument. **Keep it, with a reason** usefully permits justified persistence. “One competing account, one evidence request” reduces duplicate writing, although “different expected patterns under the accounts” is still the hardest instruction for this role.

Malawi gave useful context for unfamiliar names, running mates, valid votes, majority, and the two separate questions: why another contest occurred, and why it had a different winner. MW2/MW3/MW5 distinguish petition, court order, appeal and implementation. MW4/MW6 prevent simply adding former candidates' vote shares or treating the new rule as the sole cause of victory. The precise 2020 source discrepancies and upload-date qualifications are honest, but the repeated full source records are a substantial reading burden beside this basic sequence. Keep that evidence available; layer its presentation.

Stage 6 clearly says inclusion checks cannot judge logic and finishing does not submit to a teacher. That protects against mistaking completion for mastery. I cannot assess actual navigation, validation, feedback, saving, or whether my rough answers would complete anything. The packet's illustrative board is not my board or a recommended answer.

## Prioritized improvements

1. Bring the neutral actor → action → changed-option example into the opening. Keep the substantive causal task and inferred-link requirement.
2. Lead each source limit with ordinary language; retain full qualifications below. Explain **adjudicate**, **capacity**, and **denominator** in context.
3. Keep card/source/table identifiers visible in a compact legend, including that shared documents are not independent confirmation.
4. Label the immediate writing task prominently; make the complete final requirements a nearby checklist. Preserve both supporting IDs, complication, claim-specific limit, rival, revision, and comparison.
5. Add a neutral worked example of a weak evidence request becoming two distinguishable expectations. My “a fair poll” request below shows this remaining gap without needing a historical answer supplied.
6. Give Malawi's court order → appeal → implementation sequence a compact timeline and keep the winner question separate. Retain alliances, turnout, rule changes and all source caveats.
7. Consolidate repeated source metadata through expandable shared records, with each card's specific limitation still visible. Retain the unresolved numerical discrepancy and inaccessible-original-judgment disclosure.
8. Make open counters secondary to understanding prompts. Preserve the explicit human-review boundary and record both justified keeping and actual revision; neither opening nor finishing should imply mastery.

## Responses: actual versus text-only

**Actually recorded:** supplied opening, “Exclusion and pressure on opposition organizations may have narrowed the routes through which opponents could build a challenge.” Confidence: **A first idea · I expect to test it**. RU1: Supporting evidence 1; default role **An action**; optional reason blank. No original prose was entered. No Malawi selection, hearing answer, final explanation, or finished record exists.

**Cold-read rehearsal only, never entered or validated:**

- Connection: “Election commission / refused registration / Nadezhdin could not be on the ballot / made a challenge harder.”
- Rival: “Maybe most people wanted Putin anyway, so a challenger would struggle.”
- Needed evidence: “A fair poll showing whether people wanted another candidate.” This is still vague: I have not specified who, when, or different expected patterns.
- Comparison: “In Russia a challenger could not get on the ballot (RU1). In Malawi the courts gave opponents another election in 2020 (MW3). The comparison suggests the courts can change opportunities, although it does not prove why the opposition won.”

I can now restate a research distinction: a source recording restrictions does not by itself measure how many votes changed, and reopening a contest does not explain its winner. This is a prompted text-only restatement after partial interaction, not evidence of retained learning or independent transfer.
