Source: cases/15usc78j/RESEARCH.md
# Wave 2 securities case research
Ten opinions, 2004–2019: four Supreme Court opinions and six circuit opinions (First, Second, Fourth, Ninth, Tenth, Eleventh). Selection deliberately combines private dismissal losses, an SEC restored jury verdict, SEC administrative affirmance, SEC summary judgment, criminal instruction review, and class certification. The latter two are not merits instances. Canonical IR was read in contextual windows before quotations were selected; isolated keyword matches were used only to locate the later Amgen discussion. No corpus text was edited.
## Reading record and pins
All character ranges below refer to canonical `text`, with paragraph-context expansion from `casedsl.py text`. The files themselves carry source SHA1, exact length, URL, date, and quote offsets.
| Opinion | Read windows | Pin / encoded issues |
|---|---|---|
| Dura, 544 U.S. 336 (2005) | 0–18487 (complete) | Parts I–II; spray-device loss and loss causation, distinguished from earlier earnings disclosure |
| Janus, 564 U.S. 135 (2011) | 0–17700 | Majority II.A–B; ultimate authority, independent fund/adviser, no attribution; dissent not encoded |
| Stoneridge, 552 U.S. 148 (2008) | 0–18500, targeted reread 10300–12200 | Parts I–III.B; undisclosed supplier transactions and private reliance, conduct can be deceptive; no transaction-connection ruling inferred |
| Amgen, 568 U.S. 455 (2013) | 0–20500; contextual paragraphs 39630–43287 | Majority I–II.A and III; certification/materiality timing and truth-on-market rebuttal, not merits materiality or reliance |
| SEC v. Ginsburg, 362 F.3d 1292 (11th Cir. 2004) | 0–18000; 18400–28000 | I–III.A.2.a; family calls/trades, disputed innocent explanations, EZ and Katz materiality; excludes Rule 14e-3 substantial-steps branch |
| Xia Bi, 927 F.3d 177 (4th Cir. 2019) | 0–26800 | I–II.C; ascertainable factual statements, absent specific exposure, alternative unread-documents ground |
| Metzler, 928 F.3d 151 (1st Cir. 2019) | 0–33000 | I–III.A–B; separate safety/usage scienter allegations; corporate-scienter analysis read for context but not encoded |
| Afriyie, 929 F.3d 63 (2d Cir. 2019) | 0–11400 | Background and I; duty-instruction harmlessness and criminal intent/government burden, not sufficiency affirmance |
| Malouf, 933 F.3d 1248 (10th Cir. 2019) | 0–26500 | Background and II.A–B; scheme and scienter administrative findings; distinguishes Investment Advisers Act analysis |
| Hui Feng, 935 F.3d 721 (9th Cir. 2019) | 0–12500; 28400–40500 | Background and III.B.1–2 (fraud discussion); materiality, intent, separate nominee/rebate schemes; Howey and broker registration not converted into invented IDs |
Amgen's IR has 59,999 characters, with the majority judgment present but later separate-opinion material capped. Other selected records do not hit the cap. Reading windows are not claims of full-opinion review outside the listed ranges.
## Annotation decisions and remaining limits
- **Final summary judgment for the SEC:** Hui Feng lacks an exact basis token. `law` is used for the appellate legal conclusions on the established undisputed record, including Feng's admissions; `proved` would imply trial factfinding and `triable` would imply merely surviving summary judgment. Each file/posture and misfit states this limitation. Reviewer should examine whether a dedicated basis is needed for future data.
- **Issue-specific holdings:** Janus has one substantive block because the only resolved element is maker attribution. Dura has two loss-related blocks. Ginsburg has three, separating the two corporate-information episodes. No blanket declaration of all fraud elements follows from a win.
- **Roles and scope:** Amgen is rule-announced/procedural throughout. Afriyie is rule-announced, with procedural scope for duty-instruction prejudice and a merits-scope not-reached willfulness rule. Stoneridge's scheme rule is rule-announced. Xia Bi's second reliance rationale is alternative. None of those blocks is counted as a merits instance.
- **Basis and burden:** Malouf's agency findings affirmed for substantial evidence and Ginsburg's restored jury verdict use proved. Private complaint failures use pleaded. No civil recklessness standard is imported into criminal willfulness, and private reliance/loss elements are not imposed on the SEC.
- **Ginsburg scope:** scienter captures the specific tipping/knowledge-use sufficiency dispute. The statute catalogue does not encode a complete tipper/tippee duty or personal-benefit test. No general modern insider-trading checklist is inferred from this historical decision.
- **Temporal limits:** Hui Feng's 2019 duty/omission discussion predates the Macquarie boundary documented in the law catalogue; the file does not announce unrestricted pure-omission liability. Janus is limited to (b), and Malouf supplies a distinct (a)/(c) application. Amgen's 2013 materiality rule is not a claim that every kind of present-day price-impact rebuttal is barred at certification. No independent citator/current-good-law review was undertaken for this historical annotation batch.
- **Mixed sentences/polarity:** factual allegations of purchase inflation support the claimant's proposed loss inference even though Dura rejects the inference. Neutral contextual investment facts in Xia Bi receive supports, but do not themselves prove reliance. A lack of pleaded exposure/knowledge has absent polarity. The binary tag remains coarse for mixed factual sentences.
- **Court metadata:** Hui Feng's truncated IR court caption was completed to United States Court of Appeals for the Ninth Circuit. Supreme Court files explicitly identify the deciding Court.
- **Frozen excerpt limitation:** the short Stoneridge reliance rule slice ends midword ("act" from "action"). This was detected after fill; hard quote-freeze instructions prohibit repair. Exactness is unaffected, and complete holding quotes capture the reliance ruling. Treat as an excerpt-boundary quality issue for review, not source corruption.
## Verification and quote freeze
Original corpus tools: `python3 neov2/tools/casedsl.py fill neov2/cases/15usc78j/*.case`; `check` reports **10 files, 98 quotes, zero errors**. `variants` regenerated the catalogue: **23 blocks, 17 merits instances**, six excluded blocks (including the alternative reliance ground). No quotes changed after fill. A local frozen tuple snapshot (file, kind, start, end, text) is `/tmp/78j-frozen-quotes.json`, SHA256 `c859e48891d1c0fd5505c46db6070be2d9e39ea381fe648f51255b9ecf72a9db`; it permits comparison before integration. Independent author-separated QA remains outstanding.