# F2 — informed expert verification of revised questions

## Scope and verdict

This is a separate expert content review, not a second blind user test. I reviewed the twelve active `pressureTests` whose IDs end in `-v2` in `revised_case_enrichment.json`, using the original `case_content.json` for the supplied historical evidence. I did not evaluate the `legacyPressureTests`, reopen or amend the frozen blind review, edit the question files, or test this revision in the browser. This is internal consistency and pedagogical verification against the supplied evidence, not a new audit of the linked primary documents or a test with human learners.

All twelve intended keys match a defensible reading of the supplied evidence. No active keyed answer reverses the historical outcome, confuses the office being elected, or asserts a measured causal effect that the file cannot establish. However, **RU-P1-v2 needs a precision edit before claiming that its alternatives are unambiguously discriminating**. Several other distractors remain invented-event or chronology errors instead of plausible nearby inferences. I recommend the concrete replacements below before sign-off.

The key distribution is balanced: four A, four B, and four C. That is useful, but it is not evidence of substantive validity. The reasons for the keys follow.

## Item-by-item verification

| Active item | Key | Judgment against supplied evidence |
|---|---|---|
| RU-P1-v2 | A | RU0/RU4 assign 94% to all respondents; RU2 concerns organizers' opportunities. A is warranted. B's wording needs correction because “most” of a subgroup is not an explicit misassignment of the 94% denominator. C clearly changes the unit measured. |
| RU-P2-v2 | B | RU1 documents both a limited January hypothetical preference and subsequent exclusion. Their separate effects are not isolated. Feedback is accurate; A invents a completed campaign and C's “later result” is insufficiently specific. Replace both. |
| RU-P3-v2 | C | RU3 concerns named international observers, while RU5 documents attempted protest or monitoring. Neither missing observation nor detention measures fabricated ballots. The key and feedback are accurate; the domestic/international distinction is a reasonable near miss. |
| MY-P1-v2 | B | MY1 describes district contests: a national popular-vote majority did not produce enough district victories for parliamentary control. C mistakes a vote/seat difference for a count of irregular ballots. A's delayed national allocation is an invented process; replace it. |
| MY-P2-v2 | A | MY0, MY3 and MY5 establish an earlier alliance, changed leadership/membership, PAS outside PH, and PH's 113 seats excluding Warisan's eight. B and C are source-checkable misconceptions. Their feedback accurately corrects them. |
| MY-P3-v2 | C | MY4 records a 2016 legal action and allegations, without measuring voters' responses. A adds a nonexistent electoral measurement. B unnecessarily imports the post-election reform episode; its chronology is wrong, but it is a weak near miss. Replace B and align `whyItMatters` with MY4/MY-D. |
| MW-P1-v2 | A | MW0 gives declared, subsequently set-aside 2019 figures and rounded reported 2020 figures. Neither corrected voter preferences nor an isolated ticket effect follows. Feedback accurately retains the unresolved publication discrepancy. B should express a mistaken inference rather than invent a verified replacement count. |
| MW-P2-v2 | B | MW4 records a changed choice on the ballot; MW6 names simultaneous changes. A confuses group shares with individual movements; C confuses shares with voter counts. Both are good near misses. Add a caution that local shifts provide clues rather than directly identifying individual transfer. |
| MW-P3-v2 | C | MW3/MW5 support legal opening and implementation; they do not measure preference change or establish administrators' reasons. A and B are plausible causal overextensions. Limit B's language to the supplied implementation account rather than implying that an entire report has been exhaustively checked. |
| RS-P1-v2 | C | RS3 monitors broadcast content; RS4 records a particular recruitment/payment operation. A substitutes typical exposure for content; B substitutes national prevalence for one investigation. These are clear and plausible scope errors. Feedback accurately locates the missing links. |
| RS-P2-v2 | A | RS0 uses all ballots cast; RS6 concerns seat allocation. Neither isolates voters' motives. C's valid-vote substitution and B's pressure-free interpretation are precise near misses. The key and feedback are sound. |
| RS-P3-v2 | B | RS1 supplies the government's public rationale; RS2 establishes campaigning without parliamentary candidacy. C confuses a government's claim with independent verification. A's invented presidential election should become a mistaken inference about the documented parliamentary campaign. |

## Corrections sent to root

### Required for fair discrimination: RU-P1-v2

Current B says, “Most self-reported Putin voters reported no personal pressure.” This does not actually say that the **94%** uses that subgroup as its denominator. A majority of the subgroup could be true even though the published 94% concerns everyone; no cross-tab is supplied. Feedback should reject an explicit denominator error, not imply that an unreported subgroup claim is necessarily false.

Replace B with:

> The 94% is the share of self-reported Putin voters reporting no personal pressure; opposition organizers' opportunities remain a separate question.

The current denominator feedback then resolves the error directly. For simpler wording, replace A's awkward “Most of all respondents” with:

> Of all respondents, 94% reported no personal pressure; opposition organizers' opportunities remain a separate question.

### Strongly recommended: replace invented episodes with inference errors

**RU-P2-v2 A:**

> The January 3% gives a reliable ceiling on the support Nadezhdin could have won if allowed onto the ballot.

Feedback:

> A pre-election hypothetical choice is not a ceiling on support after a campaign. The file does not observe a campaign with Nadezhdin on the ballot.

**RU-P2-v2 C:**

> The January 3% measures how much support the later exclusion removed from effective competition.

Feedback:

> The share selecting him before exclusion does not estimate the effect of exclusion. The file does not observe what support or organization would have developed with ballot access.

**MY-P1-v2 A:**

> A national popular-vote majority means the opposition must also have led in most districts, so the seat count must be wrong.

Feedback:

> National totals do not show how votes were distributed across districts. A coalition can receive more votes overall while winning fewer districts; the gap alone does not establish miscounting.

**MY-P3-v2 B:**

> Its official legal status lets us treat the alleged misappropriation as a proved fact when explaining the campaign.

Feedback:

> A filed complaint records allegations and legal action. The announcement alone is not a judgment establishing those allegations.

Revise `whyItMatters` to keep it on the triggered source:

> A pre-election legal action can form part of campaign context. The announcement itself does not prove the allegations or measure voters' responses.

**MW-P1-v2 B:**

> The court's annulment lets us use the displayed 2019 shares as corrected estimates of voters' true choices.

The current feedback remains appropriate: the figures are declared figures, not a new verified count.

**RS-P3-v2 A:**

> Parliamentary seats were at stake; the president's campaigning role therefore shows that he was standing for one of those seats.

Feedback:

> Campaigning for a list is different from being a candidate on it. RS2 states that Vucic campaigned without being a parliamentary candidate.

This preserves a recognizable actor/office confusion without introducing an entirely different election. In keyed B, “public rationale” is more precise than “public purpose,” since the source records what the government said rather than verifying its motives.

## Feedback, vocabulary, and load

The revised alternatives place correct details in some wrong answers, making source distinctions more consequential. Particularly strong examples are RS-P2's denominator substitution, RS-P1's content/exposure distinction, and MW-P2's share/count distinction. These require more than choosing the longest cautious statement. Feedback generally identifies the precise erroneous step rather than merely labeling an answer wrong.

The repeated “Discuss or note” demand is compatible with oral use and short revision notes. It should not imply a second mandatory written response when the actual task is to revise the same account. Several items still contain an objection, rationale, demand, question, three alternatives, feedback, and a board decision. That remains a substantial reading sequence; content inspection alone cannot establish improved classroom completion time.

Vocabulary is clearer where it names an actual group, action or record. Some expert language remains: “units of observation,” “independently varied,” “counterfactual,” “aggregate,” and “institutional compliance.” These can be replaced in student-facing feedback with “different people or activities,” “changed separately,” “what would have happened,” “group totals,” and “institutions following the ruling.” The methodological distinction should remain intact.

For MW-P2, add: “Local totals can supply clues but still do not track individuals; linked voter evidence is stronger for testing transfer.” For MW-P3 B, use “the supplied implementation account already explains administrators' reasons” so the item assesses the available record, not assumptions about unseen pages of a whole report.

**Sign-off:** The set is substantively coherent and the answer keys are defensible. Make the RU-P1 precision edit and the grounded replacements above, then confirm the final stored wording and key mapping. This review does not establish actual novice difficulty, timing, transfer, or learning gains.
