Source: qa/wave2-cases-QA2-crosscheck.md
# QA2 cross-check — wave-2 case sample Reviewed QA1's exact 12 case files and six additional files, one from each of the six statutes in QA1's assignment. The added cases were Wrolstad, Leal, Bozzuto, Malouf, Lexmark, and CBOCS West. Selection favored issue-specific or procedural cases recorded in the directories' research notes. I opened the relevant majority/opinion passages in the pinned IR text; quote offsets were checked first across the corpus. `python3 neov2/tools/casedsl.py check neov2/cases` reports 165 files, 2,182 quotes, zero errors, and one existing warning: Moncrieffe's small-marijuana exception is declared as an exception in LAW. The review below concerns whether the case labels reflect each opinion, not whether a quote matches its source. ## Per-file verdicts | Case file | QA1 | QA2 | Reason | |---|:---:|:---:|---| | `29usc623/f3d-900-923-obrien-v-caterpillar.case` | PASS | PASS | The opinion finds the particular practice and significant disparity prima facie, then separately sustains RFOA. Opposite polarity is preserved. | | `29usc623/f3d-914-480-kleber-v-carefusion.case` | PASS | PASS | En banc majority excludes outside applicants from §623(a)(2) impact. The file scopes this to the impact claim and does not treat it as a §623(a)(1) treatment bar. | | `21usc846/f2d-525-881-oviedo.case` | PASS | PASS | Subjective belief is accepted, while objective acts are held insufficient under the Fifth Circuit's attempt test. The file avoids turning the sham-drug facts into a universal impossibility rule. | | `21usc846/f3d-934-1251-hawkins.case` | PASS | PASS | Sufficiency is assessed using all evidence admitted at trial, including the agent testimony later held inadmissible; the procedural scope and retrial result prevent it reading as a final guilt finding. | | `29usc158/f3d-894-0707-innout.case` | PASS | PASS | The blanket pin rule is presumptively unlawful; the Fifth Circuit upholds the Board's rejection of both public-image and food-safety defenses. The result follows the charging party despite the employer's petition. | | `29usc158/f-supp-792-1459-watson.case` | PASS | PASS | §10(j) interim bargaining is kept as background/procedural, with no final ULP merits instance inferred. | | `15usc78j/us-564-135-janus-v-first-derivative.case` | PASS | PASS | Rule 10b-5(b) maker attribution is limited to ultimate authority; the file does not infer immunity from scheme liability under (a)/(c). | | `15usc78j/f3d-935-721-sec-v-hui-feng.case` | PASS | PASS | SEC summary-judgment findings on materiality, scienter, and distinct schemes are separated; private reliance and loss elements are not imported. | | `15usc1125/us-529-205-wal-mart-v-samara-brothers.case` | PASS | PASS | The Court announces that product design cannot be inherently distinctive and remands without deciding secondary meaning. `not-reached` is used for the ultimate protectability application. | | `15usc1125/f3d-932-589-sportfuel-v-pepsico.case` | PASS | PASS | Fair use is the summary-judgment ground; likely confusion is expressly not decided. The case does not turn fair use into an infringement element. | | `42usc1981/f3d-927-81-hu-v-city-of-new-york.case` | PASS | PASS | The Second Circuit vacates dismissal on comparator pleading; own-contract, §1983-remedy, and post-*Comcast* but-for questions remain not reached. The IR-cap limitation is disclosed. | | `42usc1981/us-546-470-dominos-pizza-v-mcdonald.case` | PASS | PASS | The Court holds McDonald lacks rights under his corporation's contracts; the file reserves third-party-beneficiary status and does not bootstrap a claim from personal injury. | | `29usc623/f3d-911-450-wrolstad-v-cuna.case` | — | PASS | The actual age-treatment claim fails at summary judgment; retaliation is separately untimely under the applicable charge deadline, not abandoned. | | `21usc846/f3d-921-951-leal.case` | — | PASS with caveat | The court decides whether two charged conspiracies are the same for double-jeopardy purposes, not whether either lacks an agreement. `scope procedural` and the gloss disclose this; the element-shaped block is still a limitation of v1. | | `29usc158/f3d-927-0672-bozzuto.case` | — | PASS | The selected interrogation is held noncoercive in its full context. `result mixed` and the IR-cap note keep other, incompletely preserved Board findings from being imputed to this issue. | | `15usc78j/f3d-933-1248-malouf-v-sec.case` | — | PASS | The SEC's scheme and scienter findings are affirmed on substantial evidence. The file does not confuse this with private reliance or §78ff criminal willfulness. | | `15usc1125/us-572-118-lexmark-v-static-control.case` | — | PASS | Zone-of-interests and two proximate-cause routes are treated as pleading holdings; commercial promotion is expressly assumed, not found. | | `42usc1981/us-553-442-cbocs-west-v-humphries.case` | — | FIX | The separate `remedies : not-reached` block conflicts with its rule/gloss: the Court responds to the Title VII circumvention argument by reaffirming that Title VII and §1981 remedies are separate and independent, and says it has no reason to reach a different conclusion. Specific damages/fees are not decided, but the remedial-independence rule is. Split those propositions or mark the rule announced/decided. | ## Disagreement table vs QA1 | Comparison | QA1 | QA2 | Outcome | |---|---:|---:|---| | QA1's exact 12-file sample | 12 PASS | 12 PASS | No per-file verdict disagreement. | | Six added files | Not sampled | 5 PASS, 1 FIX | CBOCS West adds a status/holding mismatch that QA1's sample did not test. | ## DSL-v2 report cross-check The “do not migrate wholesale” conclusion remains sound: v2 is not in the production loader, still cannot aggregate linked rulings into a claim result, and its own examples document remaining semantic gaps. But the report's size table does not reproduce against the current case twins. `python3 dsl2.py stats cases` gives 855 v1 lines, 1,439 v2 lines, 84 v1 blocks, 181 v1 quotes, and 46 v1 misfits; the report gives 636, 1,439, 83, 190, and 50 respectively. The v2 totals (133 rulings, 328 quotes, 38 instances plus 7 findings-below, 78 uncertainty fields, 45 misfits) reproduce. V2 is therefore 1.68× the current v1 line count, not 2.3×; its greater structure and cost remain, but that comparison needs correcting. I checked three re-expressed cases against their actual opinions: - **Boyle — PASS with caveat.** V2 separates the jury's enterprise finding (reviewed under an approved instruction), the majority's rule/relationship holdings, and Stevens's dissent. This accurately avoids claiming the Supreme Court reviewed evidentiary sufficiency. The `requirement-rejected` ruling on the exact wording of a jury instruction is still attached to an element target; the report itself marks this as a remaining type mismatch. - **American Manufacturers — FIX to the case model, accurately disclosed by the report.** The opinion treats three state-action tests as alternative routes. V2 writes each as a separate ruling but labels their relationship `ground cumulative`, which is not the relation the court applied. Since `ground` is ignored by the classifier, the party-asymmetric Part II/Part III alternatives are stored but not consumed; the report correctly says so. - **Moncrieffe — PASS with caveat.** The majority holds the state conviction may correspond to either §841(b)(1)(D)'s felony tier or §841(b)(4)'s misdemeanor tier, with neither the default; ambiguity means the record does not necessarily establish the felony. V2 captures that relation and keeps the federal-sentencing burden discussion separate from the categorical immigration inquiry. The compared Georgia law and actual INA disposition remain outside the modeled §841 section, as the report acknowledges. ## Judgment on QA1 reliability QA1's PASS verdicts on the exact 12 selected files hold up under this independent spot-check. That is meaningful evidence that its case-level calls are careful on this purposive sample, but it does not establish a 12/12 quality rate for the full wave: one added CBOCS case exposes a concrete `not-reached`/decided-rule mismatch. The broader review should therefore trust QA1's specific PASSes provisionally, while continuing independent checks of scope and holding type; the earlier Fable review's 4/8 FIX results concern a different author/sample and are not a direct counterexample to these 12 judgments. Verification: casedsl 165 files / 2,182 quotes / 0 errors (one unrelated warning); DSL-v2 checker 19 files / 367 quotes / 0 errors; DSL-v2 `stats` rerun recorded above. No case, law, or source-text files were edited.