# F5 — informed second-pass verification of the revised MC bank

This is an informed content review after the frozen blind F5 walkthrough, not another blind review. I checked only `pressureTests` with IDs ending `-v2` in `revised_case_enrichment.json`, against the four cases, tables and cards in `/workspace/sites/who-holds-advantage-game/content/case_content.json`. I did not edit either input or the blind review. I did not retest rendering, selection logic or runtime scoring in this pass. The reviewed enrichment file had SHA-256 `da578a16c8c7faf78bfd6abd1f32e9577531a9996b8dc4deca4f81875f409189`.

**Verdict:** approve the bank as source-reasoning practice after one necessary wording fix to RU-P1-v2. The keyed choices are defensible against the supplied evidence, and the distractors are substantially closer than in the blind version. A few items still disclose their answers through a uniquely cautious option or through the stem itself. The proposed authored rival explanation and specific evidence request are appropriate complements, but neither MC success nor filled authored fields should become evidence of mastery without human evaluation.

## Correctness and discrimination by item

| Active item | Key | Content judgment and targeted concern |
|---|---:|---|
| RU-P1-v2 | A | Correct denominator/unit distinction: RU0/RU4 use all respondents for 94%; RU2 concerns organizers. **Fix B:** “Most self-reported Putin voters reported no personal pressure” is not necessarily false and does not actually assign the 94% to that subgroup. A high overall proportion may coexist with, or under additional subgroup information imply, a majority within that group. The intended error must explicitly concern the percentage’s denominator. A and B otherwise provide a useful matched pair. |
| RU-P2-v2 | B | Correctly separates a hypothetical January preference measure from a later entry barrier and avoids estimating a missing campaign. A’s “full campaign” is contradicted by RU1; C mistakes a hypothetical share for an exclusion effect. Both are longer, more involved errors than simple absolutes, but B remains the only answer saying the contribution is not isolated. This item can still be answered by the generic rule “choose uncertainty.” |
| RU-P3-v2 | C | Correctly distinguishes the missing named international assessment from separate domestic attempts to protest or monitor in RU5. A’s international/domestic conflation is a worthwhile close distractor. B’s conversion of missing observation into a manipulation estimate is clearly unsupported. “The missing mission prevented this assessment” is slightly awkward; name the OSCE assessment and the missing invitation directly. |
| MY-P1-v2 | B | Correctly applies MY1’s district contests rather than treating national votes as a proportional seat entitlement or a count of fabrication. A’s “local totals delayed that allocation” is less plausible than the rest of the new bank: it invents a delayed national allocation. It could instead misidentify the electoral system while preserving a plausible relationship between votes and seats. |
| MY-P2-v2 | A | Strong item. It requires the earlier PR alliance, different PH membership/leadership, PAS outside PH, and PH’s 113 excluding Warisan’s eight. B and C are concrete, plausible errors rather than merely reckless causal claims. The key does not win by sounding more cautious. |
| MY-P3-v2 | C | Correct about MY4’s pre-election civil forfeiture action/allegations and the unmeasured voter response. A tests scope well, but B’s post-election timing is already refuted by the stem’s explicit “2016, before the May 2018 election.” Replace B with a plausible mistake about allegation versus adjudicated finding, retaining a cautious voter-effect clause. |
| MW-P1-v2 | A | Correctly distinguishes historical reported figures from a corrected count and rejects matching rules/participation. However, the stem tells the learner both key caveats before asking them to recognize them. Reduce that disclosure if actual table/source consultation is the aim. The supplied original case JSON does not itself set out the exact two-series 2020 discrepancy; see the source limitation below. |
| MW-P2-v2 | B | Correct: MW4 records a new ballot/ticket; aggregate shares cannot identify individual transfers, and a percentage increase does not establish a count increase. A and C test real interpretation errors. “Tracking voters or local shifts” should not imply that ecological shifts identify individual transfers. Local evidence can help assess alternatives without proving voter-level movements. |
| MW-P3-v2 | C | Correctly distinguishes reopening/implementation from changing preferences or explaining institutional motives. B is a meaningful improvement over “no other institution mattered”: it tests description versus explanation of compliance. C alone explicitly reserves a further evidentiary need, so caution remains a partial cue. |
| RS-P1-v2 | C | Correctly distinguishes monitored content, one firsthand operation, population prevalence and national vote effects. A’s exposure/content confusion and B’s operation/prevalence confusion are close and specific. C is still the only option including an explicit limit; add a sensible limitation to a distractor while keeping its first measurement claim wrong. |
| RS-P2-v2 | A | Strong item. C retains a sensible motive limitation while changing the 46.75% denominator from all ballots to valid votes. This directly counters the caution shortcut and requires checking RS0. B confuses recorded ballots with support purged of pressure. A’s conversion language is acceptable; it does not say invalid ballots themselves receive seats. |
| RS-P3-v2 | B | Correctly identifies the parliamentary office, Vucic’s campaigning without candidacy, and a government statement as evidence of stated purpose. A and C each preserve one correct distinction while making another error. This requires both office identification and source-role judgment. The key is two words longer, an immaterial difference on its own. |

## Required fix and suggested wording

For RU-P1-v2, replace the vague “Most…” denominator pair with explicit statements about the number being interpreted:

- **A:** “The 94% describes all respondents’ reported personal experience; it does not measure the opportunities available to opposition organizers.”
- **B:** “The 94% describes self-reported Putin voters’ personal experience; it does not measure the opportunities available to opposition organizers.”
- **C:** “The 94% describes all respondents’ personal experience and also measures the opportunities available to opposition organizers during campaigning.”

This makes B specifically wrong about the stated denominator rather than potentially true about an unreported subgroup. Adjust its feedback to say that the source supplies no 94% estimate for Putin voters. Also remove the ungainly phrase “Most of all respondents.”

For MY-P3-v2, a better B would be: “It records a pre-election judgment establishing misappropriation; connecting that judgment to voting requires evidence about voters’ responses.” The key remains the complaint/allegations option. The cautious second clause then cannot rescue a wrong account of the legal source’s status. The file explicitly says civil forfeiture complaints; do not inadvertently imply that a press release itself adjudicates the allegations.

For MW-P2-v2, consider “The joint ticket changed the available choice; following the same voters and comparing local turnout would help test rival explanations.” Feedback should still distinguish help in testing from identification of every individual transfer. For RU-P3-v2, “Without an invitation, OSCE bodies could not assess the election; separate reports document attempted protest and monitoring despite detention risk” makes the actor/action/option more direct.

## Cues, source burden and limits of approval

The 12 keys are evenly distributed: four A, four B, four C. Option lengths differ by no more than two whitespace-delimited words within any item. A key is uniquely longest in only three items, tied for longest in seven, and shorter in two. That is a substantial improvement; I do not regard answer length as a material remaining problem. These counts describe the text, not observed learner behavior.

The more significant remaining cue is epistemic style. RU-P2, MY-P3, MW-P1/P2/P3 and RS-P1 let a learner favor the one answer that explicitly leaves an effect uncertain. Some of this is inherent in teaching evidentiary limits, but not all limits must appear only in keyed options. RS-P2 C demonstrates the better pattern: a cautious clause accompanies a wrong denominator. Add similarly bounded distractors to a few items rather than forcing equal wording everywhere.

Some source distinctions are now substantive and case-specific: denominator versus subgroup, international versus domestic monitoring, old versus new alliance membership, seat system versus vote totals, complaint versus finding, and content versus exposure. The bank still practices recognition rather than requiring an independently generated explanation. MW-P1 and MY-P3 are especially answerable from the stem plus general caution. Shortening the stem’s supplied caveats and asking the authored follow-up to name an exact observation would increase the evidentiary burden without adding much writing.

All four countries’ evidence-card arrays in the enrichment file matched the original case JSON exactly. That original file does not contain the exact conflicting 2020 MEC totals mentioned in MW-P1’s stem; it only cites the MEC series and gives rounded values. I previously saw the discrepancy disclosed in the visible source records during blind play, so this is not a newly discovered contradiction. This pass nevertheless cannot independently verify those exact totals from the two specified JSON inputs. Keep their source-record explanation and locators available wherever MW-P1 is shown.

I have not reviewed the new authored rival/evidence-request implementation, tested student performance, calibrated item difficulty or audited primary PDFs in this pass. Subject to the RU-P1 fix and those limits, the bank is suitable for a classroom pilot as practice with reviewable attempt history, not as a validated assessment instrument.
