US Pre-Market Brief | 2026-09-08 (ET) Tuesday

• Coverage window: 2026-09-04 (Friday) 16:00 ET regular-session close → 2026-09-08 about 08:10 ET
• Note: the previous trading day was 9/4 (Friday). 9/7 (Monday) was Labor Day, and the NYSE, Nasdaq and the bond market were closed all day. Every US single stock read on this machine shows last trading day = 2026-09-04, which is machine evidence that "Monday did not trade."
• This piece therefore covers a three-day news gap: Friday after hours → the weekend → Monday (Europe and Asia traded as usual, US equities did not) → this morning's pre-market. Everything that happened on Monday is completely unpriced in US equities — that is the most important structural fact today.
• Price basis: 9/4 closing prices and this morning's pre-market prices for single stocks are both taken from the stockanalysis quote API, and every one of them has been verified by back-computing "pre-market price − change = prior close"; ⚠️ but pre-market quotes are cumulative, and the read times across sections of this piece are not exactly the same (most are 08:00–08:10 ET; ORCL in §5.10 and the Dow constituents in §0.2 are 08:18–08:23 ET), so a few names show small differences between sections (for example AMGN is −6.24%/$409.95 in §5.2 but −6.00%/−$26.23 in the Dow attribution in §0.2 — that is a difference in read time, not a conflict of basis); indices/futures/commodities are taken from the CNBC quote API (with trade timestamps); the closing benchmark for Treasuries is taken from the official US Treasury yield-curve CSV (9/4 is the last day with data; the absence of a 9/7 data row is official evidence that the bond market was closed), while this morning's intraday readings come from the CNBC quote API — these are two different sources, do not mix them.
• Pre-market snapshot time: 08:00–08:10 ET. Pre-market quotes are cumulative, not final; they will still move before the open. Read every pre-market percentage move in this piece as of that time.
• Pre-market liquidity is measured by "notional amount," never by share count (1 million shares of a $10 stock ≠ 1 million shares of a $1,000 stock).

---

0. One-Sentence Summary of the Day

1. The biggest price event today is not macro and not AI — it is a pharma accident, and on "the ones listed in the US" it is brand new. Novartis suffered three failures in one week: ① the lipid-lowering drug pelacarsen in Phase 3 (Lp(a)HORIZON, 8,323 patients) successfully lowered Lp(a) but did not lower cardiovascular event risk; ② del-desiran in Phase 3 (HARBOR, 159 patients) for myotonic dystrophy (DM1) missed its primary endpoint; ③ the cell therapy rap-cel was halted after 3 patient deaths. But NVS itself and the US stocks being extrapolated from it must be looked at separately — this is the most important disaggregation in this piece:

   • The Swiss exchange (SIX) was open as usual on US Labor Day. Novartis's Swiss line NOVN: 9/4 close 129.58 CHF (−1.33%) → 9/7 close 125.46 (−3.18%, with only the pelacarsen news that day) → 9/8 open 113.40 (−9.61%, mainly del-desiran).
   • The two-day cumulative move 129.58→113.40 = −12.49%, while the NVS ADR is −12.51% pre-market — a difference of only 0.02pp. ⚠️ This piece does not treat that as an independent test — an "implied-FX residual of 2bp" is algebraically identical to "a 0.024pp difference between the two percentage moves," the same thing said another way, introducing no external information; and the ADR pre-market quote was quite possibly posted mechanically by market makers off the Swiss line, in which case the "synchronization" is structural rather than discovered. The two read times are also not aligned (the Swiss line uses the 9/8 open = 03:00 ET, the ADR uses 08:00–08:10 ET). ⇒ What can be said is: NVS's decline matches the Swiss line in direction and magnitude, and is more likely an ADR catch-up than new US pricing; but this is not exclusionary proof that it "contains no incremental US information."
   • And this disaggregation overturns the center of the narrative: the market is pricing a loss from del-desiran (−9.6%) roughly 3 times the size of pelacarsen (−3.2%). Every headline is about Lp(a), but the genuinely more expensive one is DM1.
   • What is truly unpriced is the handful of US-listed names: Ionis's 8-K was only accepted at 9/4 17:00 ET (one hour after the close), which is why IONS fell just 0.07% on 9/4; and US equities were shut on Monday. Today is the first chance IONS / AMGN have had to trade this news.

2. The Dow is far weaker than the Nasdaq, and the largest single piece of that is one stock — but not all of it. Dow futures −0.78% vs Nasdaq 100 futures −0.03%. I divided each of the 30 Dow constituents' pre-market dollar change by the Dow divisor (≈0.168; this piece did not obtain a primary source for that value, flagged as pending verification) to build a bottom-up attribution: about −375 points in total. Of that, AMGN alone contributes about −156 to −162 points (depending on read time, see below), roughly 40%, which is 3.4 times the second-largest (GS at about −45 points). ⚠️ Three things must be made explicit, or this attribution will be overused:

   • The 40% share does not depend on the divisor value (numerator and denominator are divided by the same number), so that part is robust; but the absolute "point" figures do depend on the divisor — read them as estimates.
   • Do not reconcile it precisely against "Dow futures −389 points": the constituent sum runs over the "9/4 cash close → this morning's pre-market" window, while futures moves are referenced to their own prior settlement — different rulers, plus basis. This piece does not assert precise claims like "96% agreement."
   • The sum includes quite a few constituents with tiny pre-market notional (TRV and SHW have no quote at all this morning), so those individual readings are themselves noise; the conclusion is robust only at the level of "AMGN is the overwhelming single contributor." This piece found no company-specific news for the other two pharma names, JNJ (about −31 points) and MRK (about −8 points), while XLV is −1.10%, underperforming S&P futures by 0.8pp — which says there is a layer of sector-level pharma selling that cannot all be charged to AMGN. So the correct statement is: about 40% of the Dow's weakness comes from this one stock, and the rest is the pharma sector falling as a whole; both are true at once; reading the whole thing as "a macro rotation out of value/defensives" is still an over-read, but you also cannot say "it is just one stock."

3. Friday's big semiconductor green candle has not gotten a second candle of confirmation this morning — but it has not been negated either, and that distinction matters. MU and SNDK are the two largest pre-market notional turnovers in the entire market this morning (US$1.174 billion and US$1.150 billion), yet they are up only +0.48% and +0.11%; on Friday they were +6.10% and +11.90% respectively. WDC +0.65% and STX +0.88% have likewise stalled. US$2.3 billion of pre-market chips changed hands for almost no price change. ⚠️ This piece's wording here must be restrained: "every trade has an equal buyer and seller" is a tautology, and distinguishing "absorption/distribution" from "accumulation at highs" requires trade-direction data, which this piece does not have; moreover this pre-market window spans a weekend plus Labor Day, so the accumulation period is far longer than a normal overnight, which means "largest in the market" itself contains a calendar factor. The only thing that can be said safely is: as of 08:10 ET, Friday's big green candle has not yet gotten a second candle of confirmation. This morning's SOXX +1.28% is carried mainly by three stocks that each have their own news (INTC +4.94%, ASML +3.33%, TSM +1.54%); strip them out and the storage chain is flat.

4. The line that is actually rising is "AI infrastructure / neocloud," and it is diverging from mega-cap tech. ORCL is +4.99% pre-market, CRWV +2.31%, NBIS +2.13%, VRT +1.24%, GEV +0.51%; while at the same moment Microsoft −0.67%, Amazon −1.05%, Google −0.70%, the software ETF IGV −0.55% and the data-center REITs (DLR −0.73%, EQIX −0.35%) are all down. This same group rose together on Friday as well (CRWV +5.68%, NBIS +7.48%, VRT +4.35%, ORCL +3.08%). So ORCL is not a lone oddity — it is the biggest riser within this group. The most popular explanation for it finds no support in this piece (though it is not genuinely falsified either): ORCL has recently been seen as trading like "a highly leveraged credit instrument sensitive to long-end rates" (its single-day −5.4% on 9/1 was attributed to the 30-year yield). Long-end yields are slightly higher this morning (10Y and 30Y each about +2bp) and TLT is −0.06%, the opposite direction from "a rally driven by falling rates." ⚠️ But this test must be acknowledged as weak: ① both ±2bp and −0.06% are within intraday noise; ② the variable that actually corresponds to "a credit instrument" is credit spreads (HY/IG, CDX, or ORCL's own CDS), and this piece did not obtain that data — and "Treasury yields up + credit spreads tighter" is a classic risk-on combination, fully compatible with this group's strength. So the correct statement is "the rates explanation is unsupported," not "it has been falsified." This piece could not find a common catalyst for the group, and honestly labels it attribution-unknown; but that "this is a sector line rather than an isolated single stock" is clear.

5. Oil is the only macro-level new variable today, but the equity market clearly does not believe it will last. Houthi forces struck Saudi energy facilities with ballistic missiles and drones on 9/8, injuring more than 70 people and setting multiple fires; Brent is +2.51% to $98.70 and WTI +2.62% to $93.88. But what was hit were the Jizan refinery and southern facilities such as Abha and Najran, not the eastern export hubs of Abqaiq / Ras Tanura — crude export capacity is most likely unaffected, so this looks more like a risk premium than an actual supply cut. Corroboration: XLE is only +1.20% and oil services OIH is 0.00%, far behind oil's +2.5%.

6. Driver ranking: clinical data (S) > geopolitics/energy (A+) > index rebalancing (A+) > AI-infrastructure sector rotation (A) > rating changes (A) > industry intent (A) > macro (B, an empty window today).

7. The policy baseline is still "hike," not "cut": CME FedWatch shows the probability of a 25bp hike at the 9/16 FOMC at about 60% (⚠️ this piece retrieved two different readings, 58.7% and 66%, with the cited source and the number not matching each other, so only the "about 60%" order of magnitude is kept and the precise value is flagged as pending verification; also note: 9/7 was Labor Day and rate futures did not trade normally, so any probability labeled a "9/7 reading" is essentially a stale Friday number). The source of it is August nonfarm payrolls at +162,000 (versus expectations of about +53,000) and Fed Chair Warsh's hawkish turn at Jackson Hole. August CPI this Friday, 9/11 at 08:30 ET, is the last piece of the hiking-path puzzle; today through Thursday is a macro vacuum, and risk appetite will lean toward waiting.

8. Pre-market state today (08:00–08:10 ET): S&P futures 7,699.25 (−0.29%), Nasdaq 100 futures 29,556.25 (−0.03%), Dow futures 53,025 (−0.78%); VIX 15.66 (⚠️ the quote API shows +2.35%, but its "prior close" is the phantom 15.30 print from Labor Day with no trading; against the real 9/4 close of 14.53, VIX is actually +7.8% — the rise in volatility is much larger than the headline number); the 10-year Treasury at 4.798%, the 30-year at 5.258%, the 2-year at 4.383% (intraday readings from the CNBC quote API; versus the Treasury CSV's official 9/4 closes (4.78 / 5.24 / 4.37) that is about +2bp / +2bp / +1bp respectively, a mild bear steepening); the dollar index at 98.967 (−0.21%); gold at $4,446.60 (−0.67%); copper +2.25%. Asia overnight: Nikkei 65,269.33 (−1.70%), KOSPI 6,954.52 (−0.58%), Hang Seng 25,317.18 (−0.38%), Shanghai Composite 3,940.55 (+0.20%), Taiwan Weighted 47,105.78 (−0.47%); Europe is roughly flat at midday (FTSE +0.18%, DAX −0.12%, CAC +0.06%, Euro Stoxx 50 0.00%).

9. An easily overlooked Asian clue: Korea held onto its gains, Japan did not. Two-day cumulative moves against the 9/4 US close: Korea +4.00%, Japan only +0.38% (on Monday Korea +4.61% and Japan +2.12%; today Korea gave back only 0.58% while Japan gave back 1.70%). Korea is precisely the market with the highest storage weight — the storage bid stayed in place in Korea. But Japanese semiconductor equipment fell hard today (Disco −5.70%, Tokyo Electron −3.30%, Advantest −1.97%), the opposite direction from the strength in European and US equipment names this morning, and this piece found no clean explanation for that divergence and honestly labels it attribution-unknown.

In one sentence: today is a bottom-up day, not a top-down day — the biggest alpha sits in the mispriced selling and mispriced buying around a pharma accident, and the biggest trap sits in the storage chain that "rose Friday but is flat this morning on huge volume" and in the index-inclusion names that "you only think to buy after a 14% two-day run"; the macro window is empty until Friday's CPI, so do not read a Dow decline caused by one stock as a change in the wind.

---

1. News Overview

1.1 This Piece's Self-Check on Three Key News Items

① Has pelacarsen actually been priced in US equities? — confirmed twice over, using primary timestamps and the Swiss line.

• Release times (SEC acceptance times, primary source): the Ionis 8-K at 09-04 21:00 UTC = Friday 17:00 ET (one hour after the close); the Novartis pelacarsen 6-K at 09-04 21:22 UTC; the Novartis del-desiran 6-K at 09-08 06:05 UTC = 02:05 ET this morning.
• The price evidence agrees: IONS fell only 0.07% on 9/4 — as co-developer it could not possibly have been motionless if the news had landed during the session.
• ⇒ For the US-listed names (IONS / AMGN): genuinely unpriced, and today is the first tradable session.
• ⚠️ But this does not hold for NVS itself: the SIX Swiss exchange was open as usual on Monday and has already priced pelacarsen (−3.18%). Calling NVS's −12.51% an "unpriced opportunity" is wrong; it looks more like an ADR catch-up (matching the Swiss line's two-day −12.49% in direction and magnitude; but as noted in §0.1, that "match" is not an independent test). On that basis this piece handles NVS separately from IONS/AMGN.
• ⚠️ This conclusion does not depend on the FX check above: it stands on three mutually independent pieces of hard evidence — the 8-K acceptance time of 9/4 17:00 ET (after the close), IONS down only −0.07% that day, and the US cash market being closed on 9/7.

② The claim that "ASML won a $400M order" is not accepted by this piece. Some aggregator sites (parameter.io, coincentral and others) wrote it up as a "$400M machine order," but the original CNBC and Techzine reports describe only "plans/commitments to adopt" and "a four-party agreement," with no dollar amount, no tool count and no binding purchase agreement. This is intent, not an order, and the volume-production windows are 2028 (Samsung) / 2030 (TSMC) — the impact on near-term revenue is close to zero. ASML's +3.33% this morning (+7.64% over two days) is pricing a narrative, not revenue.

③ Intel's price increase is a rumor, not company disclosure. This machine checked SEC EDGAR: Intel's most recent 8-K is dated 8/12, with no disclosure since; the sole source of the price-increase story is DIGITIMES citing the supply chain. The same DIGITIMES piece also writes that Qualcomm and MediaTek are eroding its IPC/IoT share — "defending gross margin through price increases while losing share" is a double-edged story, and you cannot read only the first half of the sentence. This piece gives INTC only 6/15 points for source authority.

④ ORCL is +4.99% this morning (notional US$243 million): no company-level driver can be found, but it is not an isolated case; and the most popular "rate-driven" explanation is unsupported. This machine checked SEC EDGAR: Oracle's most recent 8-K remains 2026-06-10, with no company disclosure since; searches also turned up no rating or order news dated 9/8. But "no company news found" does not equal "this stock is rising in isolation for no reason" — this piece ran two control tests:

• Peer-group test (holds): the AI-infrastructure/neocloud group rose together — CRWV +2.31%, NBIS +2.13%, VRT +1.24%, GEV +0.51%; and they rose together on Friday too (CRWV +5.68%, NBIS +7.48%, VRT +4.35%). ORCL is the largest name in the group, not a lone stock.
• Rates test (falsified): ORCL has widely been seen as trading like "a highly leveraged credit instrument sensitive to long-end rates" (its single-day −5.4% on 9/1 was attributed to the 30-year yield rising). If this morning were also rate-driven, the long end should be falling; but the actual readings are 10Y +2bp, 30Y +2bp and TLT −0.06%, the opposite direction. The rates explanation does not hold.
• At the same time, mega-cap tech and software broadly are down (MSFT −0.67%, AMZN −1.05%, GOOGL −0.70%, IGV −0.55%, DLR −0.73%, EQIX −0.35%) — so this is not a "growth stocks up across the board" beta either.

Conclusion: this is a genuinely existing sector line (AI infrastructure/neocloud) whose common catalyst this piece could not identify, honestly labeled attribution-unknown. But for ORCL as an individual name, attribution-unknown + already up 8.06% over two days + earnings after the close on 9/10 stack up, and the correct posture is still not to participate.

⑤ ORCL's "consensus" is an echo of guidance — do not treat it as an independent hurdle. The widely cited FY27Q1 consensus revenue of "US$19.13 billion" is exactly the midpoint of the guidance range the company itself gave on 6/10 (US$18.96 billion–US$19.26 billion) — that is not an independent third-party forecast, it is the company's guidance repeated back. More dangerous is EPS: another aggregated basis gives $1.299, whereas the company's own non-GAAP guidance is $1.72–$1.76, a 26% gap, so they cannot be the same basis (the former is very likely GAAP). Treating $1.299 as the non-GAAP hurdle would manufacture a "massive beat" out of thin air when the print lands. This piece therefore uses no single "consensus" number and lists only the company's original guidance.

---

2. Strongest Themes, Descending

---

3. Overall Single-Stock Strength Ranking

Sorted by total score, descending. Bullish grades: S / A+ / A / B+ / B / C. Pre-market moves and notional are the 08:00–08:10 ET snapshot.

3.1 Bullish Direction

3.2 Bearish Direction

---

4. Single-Stock Scoring Model (out of 100)

Components: source authority 0–15 · catalyst directness 0–20 · earnings elasticity 0–15 · moat and fundamentals 0–15 · expectation gap 0–10 · catalyst durability 0–10 · trading characteristics 0–10 · risk deductions 0 to −15

Notes on the scoring (why it was done this way):

• PHVS gets full marks on authority and directness, because it is a Phase 3 topline released by the company itself — the highest tier in this piece's evidence hierarchy; the deductions come mainly from "thin volume + already up 17.8% pre-market + single asset."
• INTC gets only 6 for authority: the sole source is a DIGITIMES supply-chain rumor, and SEC EDGAR shows no disclosure from the company since 8/12. It gets full marks on trading characteristics (the largest pre-market notional in the market at US$461 million), but 11 points of risk deduction (rumor + up 9.67% over two days + US$20 billion August offering dilution + share loss).
• ORCL gets only 4 for authority: attribution unknown, with no 8-K on EDGAR since 6/10. Its low score is not because the news is bearish but because nobody can say why it is rising, and earnings come in two days.
• BE takes a 13-point risk deduction: +14.07% over two days and +150% YTD, while SNDK, added to the S&P 100 in the same batch, rose 11.90% on 9/4 and is up only +0.11% this morning — a control group from the same week and the same mechanism has already shown how fast this kind of impulse fades.

---

5. Detailed Analysis of Top Names

5.1 PHVS (Pharvaris) — priority deep-dive | total score 70

• Related news (09-08 pre-market, company primary source): the CHAPTER-3 pivotal Phase 3 of deucrictibant XR 40mg for HAE attack prevention, n=85, adolescents and adults, covering all three HAE types, 2:1 randomized, double-blind placebo-controlled, 24 weeks. Monthly attack rate reduced 83% versus placebo (p<0.0001); a reduction of 87% among the 80 patients with types 1/2. The primary endpoint and all secondary efficacy endpoints reached statistical significance, with effect from the first week sustained through 24 weeks; well tolerated, with no treatment-related serious adverse events and only 1 discontinuation. Marketing applications and a US NDA are planned starting in the first half of 2027. Source⁶
• Catalyst logic: this is a valuation re-rating catalyst, not a current-earnings one — the company has no product revenue today. An 83% reduction in attack rate is strong data in HAE prevention, and oral dosing is an administration advantage over existing injectable prophylaxis. Order of impact: valuation > sentiment > revenue (revenue does not arrive until after 2027).
• Theme and stage: launch day. Today is the first trading session after the data release.
• Fundamental verification: a pre-revenue biotech, so revenue/EPS/gross margin metrics do not apply. Cash balance and runway cannot be confirmed in this piece (the latest quarterly cash figures were not obtained), and that is this item's evidence gap.
• Moat: an oral bradykinin B2 receptor antagonist which, if approved, would compete directly in HAE prevention with Takeda, BioCryst, KalVista and Ionis's donidalorsen. The moat comes from data strength and dosing convenience, not scale.
• Pre-market and technicals: +17.79%, $41.52; pre-market notional $12.8M (287,000 shares) — this is thin, and the price may reposition sharply after the open. The 52-week range is $20.65–$38.13, and the pre-market price is already through the 52-week high.
• Final judgment: priority deep-dive. The data itself is the hardest piece of good news in the market today, but it is already up 17.8% pre-market on thin volume, so chasing it directly has a poor risk/reward. The right approach is to wait for volume-and-price confirmation in the 30 minutes after the open, not to post orders pre-market.

5.2 AMGN (Amgen) — watch only | the single largest drag on the Dow today, but the extrapolation risk is real

• Related news (09-08 pre-market): ① passively hit — Novartis's pelacarsen (an antisense oligonucleotide, ASO) succeeded in lowering Lp(a) in Phase 3 yet failed to lower cardiovascular event risk, and the market immediately extrapolated to Amgen's olpasiran (siRNA); ② actively hit — BMO downgraded it to Market Perform on valuation (up about 34% YTD).
• Catalyst logic: this is space-level falsification risk. The question pelacarsen raises is far more serious than "one drug failed": can lowering Lp(a) itself translate into cardiovascular benefit?
• What matters is the "shape" of the failure, and this shape is the most extrapolable kind. The original Novartis and Ionis announcements are explicit: pelacarsen "achieved significantly lower Lp(a) levels" but this "did not translate into a reduction in cardiovascular risk." That is a translation failure, not a potency failure — the drug pushed Lp(a) down and events did not fall.
• So the "ASO vs siRNA" firewall is much thinner than the market imagines. pelacarsen is a GalNAc-conjugated antisense oligonucleotide (ASO) and olpasiran is a GalNAc-conjugated siRNA; but both act on the same biological node (hepatocyte APO(a) mRNA → less Lp(a) particle production) and use the same delivery route (GalNAc→ASGPR→hepatocyte). The difference between ASO and siRNA lies in degradation enzymology and dosing frequency, not in pathway. ⇒ This piece's judgment is that extrapolability is "high but not complete," rather than "undecided."
• olpasiran has three defenses left; graded one by one:

  Defense	Falsified?	Strength
  A. Deeper reduction, possibly crossing a threshold	Not falsified	The only one that could save the entire class; but it requires a threshold effect in the dose-response relationship, with pelacarsen sitting to the left of the plateau. There is currently no Phase 3 evidence at all
  B. A cleaner endpoint: olpasiran's primary endpoint is CHD death + MI + urgent revascularization, excluding stroke; pelacarsen used 4-point MACE (including non-fatal stroke)	Not falsified	Moderate. Mendelian randomization associations for Lp(a) are stronger for coronary disease than stroke, so including stroke does dilute
  C. A more enriched population: olpasiran's threshold is Lp(a)≥200 nmol/L and limited to coronary ASCVD; pelacarsen used ≥70 mg/dL and included stroke/PAD patients	Not falsified	Moderate. ⚠️ Conversion between nmol/L and mg/dL varies with assay method, so one cannot precisely assert that "olpasiran's threshold is higher"

  Table scrolls horizontally
• The time dimension matters enormously — the market is trading a 2028 event today: the OCEAN(a)-Outcomes registry entry (NCT05581303, 7,297 patients) gives an estimated primary completion date of 2028-03-31. olpasiran contributes zero revenue today, and AMGN's 2026–2027 EPS is completely unaffected by this news — this is purely a long-dated/terminal-value re-rating, and the 25.5x TTM P/E does not change mechanically because of it.
• A quantified imbalance: AMGN is losing about US$14.75 billion of market value this morning, while Novartis, which actually read out the failure, lost only about US$9.7 billion on the day it actually read out (Monday). The company being extrapolated to lost more than the company that actually failed. (Deduction required: BMO's same-day downgrade is also part of that US$14.75 billion for AMGN.)
• The falsifiable point most worth watching: the Novartis CMO's wording was "did not demonstrate a translation into reduced cardiovascular risk in the overall study population" — the qualifier "overall population" was left in, and the trial's primary endpoint was assessed in both the ≥70 mg/dL and ≥90 mg/dL populations. The result in the ≥90 mg/dL subgroup is entirely unknown for now and awaits a medical conference. If there is a signal in that subgroup, defense C strengthens greatly; if it is equally negative, olpasiran is left with only defense A.
• Dow attribution: AMGN is −$26.23 pre-market, which at a Dow divisor of ≈0.168 works out to −156.1 points, 40% of the Dow futures' −389 points, and 3.4 times the second-largest, GS (−45.3 points). 40% of the "Dow is weak" story today is this one stock.
• Pre-market and technicals: −6.24%, $409.95, pre-market notional $46.9M. It had already fallen 1.55% on 9/4, for a two-day cumulative −7.69%. The 52-week range is $269.77–$447.03, so it is still near the top of the range — i.e. "down 6% but nowhere near cheap."
• Final judgment: watch only. Unlike the first draft, this piece does not consider today's decline a mispricing — the failure's shape (Lp(a) came down but events did not) is the most extrapolable kind, ASO/siRNA is not an effective firewall, and −6.24% is reasonable in direction. But shorting is also not advised, for three reasons: ① the readout is not until 2028-03, so a short pays two years of time cost; ② the ≥90 mg/dL subgroup result is unknown, and any signal there would produce a violent rebound; ③ AMGN's current earnings are entirely unaffected (TTM P/E 25.5x, FCF yield 4.59%, dividend yield 2.46%, TTM operating cash flow US$12.478 billion, OCF/net income = 1.43x, good cash-flow quality). This is a stock whose long-dated story got worse while its current cash flow did not change, and neither direction has a path to realization within 12 months. ⚠️ One metric you cannot use: AMGN's ROE of about 91% and P/B of about 19x are meaningless numbers — goodwill from the Horizon acquisition plus continuous buybacks have compressed shareholders' equity to just US$11.688 billion. The valid profitability reading is ROA ≈ 9.1%. Net debt is US$43.315 billion (down US$11.16 billion from the 2023Q4 peak of US$54.479 billion), and net debt / TTM operating cash flow = 3.5x, still high but improving directionally.

5.3 IONS (Ionis) — watch only | hit twice today

• Related news: ① directly hit as pelacarsen's co-developer; ② competitive position damaged — PHVS's deucrictibant XR delivered an 83% attack-rate reduction in Phase 3 HAE prevention, directly threatening the position of Ionis's own donidalorsen in the same indication.
• This is one of this piece's independent findings: the market simply attributes IONS's −10.48% to pelacarsen, but the HAE leg is a second, independent negative. Corroboration: fellow HAE name BCRX is −2.46% pre-market. ⚠️ But that corroboration is very weak — BCRX's pre-market notional is only $0.8M; and in the same space TAK (+1.27%, 257 shares) and ARGX (−2.72%, 248 shares) have volumes so small they are pure noise and cannot serve as evidence. So this piece labels "HAE competitive spillover" as a hypothesis with directional support but insufficient evidence, not a conclusion.
• Pricing evidence: IONS fell only 0.07% on 9/4, while the 8-K was accepted only at 9/4 17:00 ET (after the close) — this is the core evidence for this piece's judgment that "pelacarsen is unpriced in US equities" (see §1.1①).
• Contract economics (10-K primary source, a layer the market generally only half-counts): Ionis can receive at most US$900 million on pelacarsen (US$75 million upfront + US$150 million license fee + US$25 million development milestones + up to US$290 million regulatory milestones + up to US$360 million sales milestones), plus tiered royalties in the mid-teens to low 20% range. As of the end of 2025 it had already received more than US$275 million. ⇒ Already banked and unrecoverable: >US$275 million; what went to zero today is at most US$650 million (regulatory + sales milestones).
• The overlooked second layer — Royalty Pharma: in the January 2023 transaction, RPRX takes 25% of pelacarsen's future royalties (i.e. Ionis retains only 75%), and that deal's up to US$625 million of additional milestone payments, conditioned on pelacarsen approval or commercial sales, is now essentially unreachable. The balance sheet still carries "liability related to sale of future royalties" of US$551.4 million (non-current) + US$11.7 million (current); the 10-K explicitly warns that a change in pelacarsen expectations would force the company to adjust the effective interest rate prospectively, marking down both non-cash royalty revenue and non-cash interest expense in coming quarters — purely accounting, with no cash effect, but it will contaminate year-over-year revenue readings for the next several quarters, something this morning's narrative has yet to mention. ⇒ Total exposure: at most US$650 million (Novartis) + US$625 million (RPRX) = US$1.275 billion of undiscounted milestones, plus 75% × (mid-teens to low 20%) of the royalty stream.
• Genuine shock absorbers: ① pelacarsen is 100% funded by Novartis, so Ionis has zero opex exposure and no sunk costs to write off; ② the company's narrative long ago shifted to its own commercialization — the 10-Q states it already has 7 marketed drugs (TRYNGOLZA, DAWNZERA, WAINUA, SPINRAZA, QALSODY, TEGSEDI, WAYLIVRA), and TRYNGOLZA obtained the sHTG indication in June 2026 (far larger than the original FCS label), which is the real current value driver.
• ⚠️ But "lots of cash" is an illusion: cash + short-term investments of US$2.055 billion (6/30) against total debt of about US$1.812 billion (as of 2025-12-31) ⇒ net cash of only about US$240 million. And H1'26 revenue was US$514.0 million, −11.9% year over year (the collaboration-milestone base is receding while product revenue is still ramping). On a net basis, "a long runway" does not hold.
• Pre-market: −10.48%, $52.00, notional $13.6M. The 52-week low is $50.01, so the pre-market price is only 4% above a 52-week low.
• Final judgment: watch only. Both negatives are real, and the Royalty Pharma layer makes the actual loss larger than the market is computing; but the stock is already down 10.5% and approaching 52-week lows, while the base business of 7 marketed drugs plus the TRYNGOLZA label expansion is untouched by today's news. Do not participate in either direction.

5.4 DYN (Dyne) / SRPT / PEPG — avoid | the three names caught by platform extrapolation

• Related news (Novartis 6-K, primary source): the HARBOR Phase 3 of del-desiran, about 150 patients, 54 weeks, primary endpoint vHOT missed statistical significance; but Novartis explicitly wrote that "clinical activity was observed in secondary endpoints and exploratory analyses," with safety consistent with prior experience.
• What is extrapolable (and should be extrapolated): the disease (DM1), the target (DMPK mRNA), the delivery receptor (TfR1), the route of administration (IV) and the observation window (HARBOR 54 weeks vs Dyne's Phase 3 at 49 weeks) are all homologous. The most extrapolable risk is not the molecule but this: can knocking down DMPK translate into measurable functional benefit within roughly a year in a slowly progressive disease — and is DM1 endpoint science itself mature? That applies 100% to DYNE-101.
• What is not extrapolable (three items, all substantive):
  1. The primary endpoints differ, and the failure happened exactly there. HARBOR broke on vHOT (video hand-opening time, hand myotonia); the primary endpoint of Dyne's Phase 3 (NCT07486934) is 5×STS (five-times sit-to-stand, lower-limb function), with vHOT only a secondary endpoint. Novartis says "clinical activity was observed" in secondary endpoints — if the final data show a signal on strength/function endpoints, that would actually support Dyne's endpoint choice. This cannot be judged before the data are released.
  2. The payload chemistry and antibody format differ (del-desiran is siRNA + a full monoclonal antibody; DYNE-101 is ASO + Fab). This piece gives that low weight, for the same reason as §5.2 (when the pathway is the same, a difference in carrier is not a firewall).
  3. The platform has not been falsified, and the evidence is in the very same 6-K: Novartis simultaneously disclosed that del-zota (DMD44) received FDA priority review, that del-brax (FSHD) will go to an FDA meeting on the strength of positive Phase 1/2 data, and reaffirmed its 2025–2030 five-year sales CAGR guidance of 5–6%. TfR1-mediated muscle delivery itself works; the problem lies with the DM1 indication and its endpoints. ⇒ This piece's judgment: extrapolability is "moderate." The disease-level translation risk in DM1 should be extrapolated; the AOC/FORCE delivery-platform risk should not.
• ⚠️ SRPT's −10.84% is very likely a misattribution. SRP-1003 is a 78-patient Phase 1/2 with "safety/TEAEs" as its primary endpoint (NCT06138743, still recruiting). By any reasonable method its weight in SRPT's valuation is close to zero. SRPT's real risk source is something else entirely (10-Q primary source): after two non-ambulatory patients died of acute liver failure, Elevidys was given an FDA black-box warning for acute liver injury/liver failure starting in November 2025 and the non-ambulatory population was removed from the indication; New York State has also restricted Medicaid coverage by age. Booking today's decline to a DM1 read-across obscures the leg that is actually deteriorating.
• DYN's real vulnerability is not "down 29%" but the mismatch between runway and catalyst timing: pro-forma liquidity of about US$1.304 billion (US$898.5 million at 6/30 + US$405.0 million net proceeds from the July offering) against US$199.3 million of debt ⇒ pro-forma net cash of about US$1.10 billion, 34% of the pre-market market cap. But the company's own 10-Q states runway only into "the second quarter of 2028," while its own Phase 3 has an estimated primary completion date of July 2028 — the two do not overlap. DYN will very likely have to raise again before its own Phase 3 reads out, and this morning's price has sharply raised the dilution cost of that raise. On an EV basis, this morning's decline is −38.2% (net cash does not re-price), steeper than the −28.95% in the share price.
• Pre-market: DYN −28.95% ($17.25, notional $8.7M), SRPT −10.84% ($20.06, $3.5M), PEPG −8.36% ($2.85, $0.4M, noise).
• Final judgment: avoid all three. Whether DYN's decline is excessive is a secondary question; the structural problem is "it must refinance before the readout," and that got more expensive today. SRPT is a case of questionable attribution + the real risk lying elsewhere. Do not catch the falling knife.

5.5 ROIV (Roivant) and the overlooked counterparty UTHR — watch closely | today's clearest structure with no market verdict yet

• First, a correction on the indication: mosliciguat treats PH-ILD (pulmonary hypertension associated with interstitial lung disease, Group 3 PH), not pulmonary arterial hypertension (PAH). That distinction determines its comparator and therefore who gets hurt.
• Related news (09-08 06:15 ET, company primary source + presented at the ERS congress): the PHocus Phase 2, 135 patients, 87 centers across 20 countries, 24 weeks, randomized double-blind placebo-controlled. The primary endpoint was met: placebo-corrected PVR at week 16 of −56.3% (p<0.0001); 6MWD at week 16 was +35.2 meters (a secondary endpoint, p=0.0027), and +52.7 meters at week 24 (exploratory). The molecule is an inhaled dry-powder sGC activator dosed once daily, originating from Bayer and developed by the subsidiary Pulmovant. The Phase 3 PHrontier (about 375 patients) is enrolling.
• ⚠️ That headline "56%" is not comparable within the same indication. The registrational trial of the existing standard of care, Tyvaso's INCREASE (PH-ILD, 326 patients), did no right-heart catheterization at all and has no PVR data; while Winrevair's PVR data come from a different disease (PAH). ⇒ "The largest PVR reduction ever" is a claim made on a surrogate endpoint the competitor never even measured, and it is unfalsifiable within PH-ILD.
• What is genuinely comparable is week-16 6MWD within PH-ILD: mosliciguat +35.2 meters vs Tyvaso +21.0 meters (registry Hodges-Lehmann basis) / about 31 meters (NEJM least-squares mean basis; the two are a difference of basis, not a data contradiction). ⇒ This is "in line to slightly better," not the blowout implied by 56%.
• The combination of "a huge hemodynamic effect + a moderate functional effect" is itself the classic PH-ILD pattern — exercise capacity in this population is limited by the underlying lung disease as well as by the pulmonary vasculature. This is precisely the core risk in the Phase 3 PHrontier: if its primary endpoint is 6MWD (not disclosed by the company, pending verification), then the correlation between the "56%" being priced today and Phase 3 success or failure is far lower than the market understands.
• A genuine differentiator that the headline may bury: the cough rate was 12.1%, below placebo's 18.2%. The real-world pain point of inhaled treprostinil is exactly tolerability and discontinuation rates, so "once-daily inhaled + a cough rate below placebo" has real commercial value.
• ⇒ Which leads to an observation worth watching: UTHR. mosliciguat threatens Tyvaso in exactly the same indication, and yet UTHR is +2.13% pre-market (the wrong direction), on just 1,992 shares and US$0.99 million of notional; fellow name LQDA (Yutrepia) traded just 1,435 shares. ⚠️ But one must be honest: the evidentiary strength of a 1,992-share pre-market is zero. It can neither prove the market "has not noticed" nor rule out "the market has already assessed it and correctly discounted it to zero" — at this volume the two explanations have identical likelihood. So this is not "a discovered mispricing" but a hypothesis to be tested: "competition exists mechanistically, while the pre-market price carries no information." Its evidence will only be generated after the open. ⚠️ And the 6MWD comparison above must itself be discounted: mosliciguat's +35.2 meters is a secondary endpoint from a Phase 2 (n=135), while Tyvaso's +21.0 meters is the primary endpoint of a Phase 3 (n=326) — different rulers; and point estimates from small Phase 2 trials are systematically inflated (winner's curse). The cough rate of 12.1% vs 18.2% at n=135 with 2:1 randomization is a difference of only 3–4 patients, and the company reported no p-value or confidence interval. Cross-trial comparison is not evidence of superiority — a rule this piece must also obey itself.
• ROIV financials (10-Q primary source): net cash of US$3.842 billion (cash US$1.2485 billion + marketable securities US$2.5936 billion, no material interest-bearing debt) = 15.2% of the 9/4 market cap. Note that the EV move is larger than the share-price move: the shares are +18.1%, but because net cash does not re-price, EV rose from US$21.4 billion to US$26.0 billion, +21.4% — the market's actual incremental valuation of mosliciguat is US$4.6 billion.
• ⚠️ Earnings quality must be read through operating cash flow, not net income: FY2026 net loss was −US$299.8 million but OCF was −US$750.3 million; the most recent quarter's net loss was −US$189.8 million vs OCF of −US$270.5 million. The income statement is systematically flattered by non-cash gains from subsidiaries/equity items, and the net loss understates the true burn by about 2.5x. At roughly US$270 million per quarter of burn, US$3.84 billion supports about 3.5 years.
• A catalyst that should not be double-counted today: LISRAYA (brepocitinib) was already FDA-approved on 2026-08-27 for dermatomyositis (the first and only targeted therapy), and has already traded for 5 full sessions; it is one of the main drivers of ROIV's move from the 52-week low of $12.53 to $34.93. That and this morning's +18% are two separate things — do not add them together.
• Final judgment: ROIV is a watch closely (the data are real, but the headline number is not comparable, the Phase 3 endpoint is unknown, and it is already up 18%); what is genuinely being overlooked is UTHR — clear logic, clear direction, and no market verdict yet.

5.6 Energy (MPC / XOM / VLO / PSX) — watch closely | what was hit is a refinery, not an export hub

• Related news (09-08, NPR/Al Jazeera): Houthi forces attacked Saudi energy facilities with ballistic missiles and drones, hitting the Jizan refinery, Abha, Najran, Economic City and the Khamis Mushait air base, injuring more than 70 people and starting multiple fires.
• Catalyst logic and this piece's key divergence: what was struck are southern facilities and refineries, not the Abqaiq processing center or the Ras Tanura export terminal in the east. The 2019 attack on Abqaiq cut about 5.7 million barrels per day at once — that was a supply shock; Jizan is a refinery (on the order of 400,000 barrels per day), and damage there affects product supply and refining margins, not crude export capacity. Therefore:
  • This piece judges this oil rally to be mainly a "risk premium," with an actual supply cut unconfirmed;
  • but precisely because a refinery was damaged, transmission into the refining group (MPC / VLO / PSX / DK) is more direct than into upstream — which is why this piece ranks refiners ahead of upstream.
• ⚠️ On "the equity side does not believe the oil move," this piece must first correct itself on one point: XLE +1.20% ÷ Brent +2.51% = 0.48, and the historical normal beta of energy equities to crude is already in the 0.4–0.6 range (upstream names have hedges, reserve costs and a capital-discipline discount, and never track 1:1). So 0.48 is normal, not "a clear signal of disbelief" — the first draft implicitly assumed beta=1, which is wrong. This piece did not obtain a rolling beta of XLE to Brent, so the observation about "which way the gap converges" is unfalsifiable without a beta benchmark, and it has been downgraded accordingly in §8.2. OIH's 0.00% comes on just 588 shares, which is noise, and this piece does not use it as evidence.
• Pre-market: MPC +1.99%, VLO +1.27%, PSX +1.09%, DK +2.41%; XOM +1.30% ($92.0M, the most liquid name in energy), CVX +1.12%, OXY +1.75%, COP +1.44%. ⚠️ Apart from XOM, energy single-stock pre-market notional is generally below US$10 million, and MPC has only $1.2M — pre-market quotes carry essentially no information for these names, so you must wait for the open.
• Final judgment: watch closely, do not chase. The verification point is falsifiable: if Saudi Aramco or the Saudi energy ministry confirms that crude exports and Abqaiq/Ras Tanura are unaffected, the oil risk premium should be given back within 1–2 sessions and energy equities weaken in step; conversely, if export damage is confirmed or the attacks continue, the 1.3pp gap between XLE and oil should converge upward.

5.7 BE (Bloom Energy) — watch only | valuation and crowding, not a "control group"

• Related news (09-04 after the close, S&P official): joining the S&P 500, effective before the open on 9/21, replacing Molson Coors. On fundamentals: Q2 revenue +166% year over year, and a YTD gain of more than 150%.
• Catalyst logic: a purely mechanical liquidity event — passive funds tracking the S&P 500 must buy, typically concentrated at the 9/18 close (the last session before it takes effect). That is certain buying, but it is one-off and has a definite expiry.
• ⚠️ A timing problem that must be corrected first (the first draft was wrong here): S&P's announcement was released after the close on 9/4, so all of 9/4's regular-session gains happened before the announcement and cannot possibly be "front-running the inclusion" — BE's +7.35% on 9/4 was the AI power trade and SNDK's +11.90% on 9/4 was the storage trade (see §5.9), and neither can be counted as part of the index event. The only part of BE genuinely attributable to the inclusion is this morning's +6.26% (plus the roughly +6% after the close on 9/4, which is the same move).
• Within the correct window (the first tradable session after the announcement = this morning's pre-market), the comparison looks like this:

  	9/4 regular session (before the announcement)	This morning's pre-market (first session after the announcement)
  BE (joining the S&P 500)	+7.35% (AI power, not index)	+6.26%
  ILMN (joining the S&P 500)	0.00%	+1.15%
  SNDK (joining the S&P 100)	+11.90% (storage, not index)	+0.11%
  DELL (joining the S&P 100)	+1.50%	−0.79%
  PANW (joining the S&P 100)	+0.40%	−0.89%
  ANET (joining the S&P 100)	+1.22%	+1.26%
  ⚠️ But the S&P 100 and the S&P 500 are not the same mechanism: passive assets tracking the S&P 100 (such as OEF) are only in the billions of dollars while the S&P 500 is in the trillions, so passive buying pressure differs by orders of magnitude and strictly speaking they cannot serve as controls for each other. So this piece keeps only one weak conclusion: the four S&P 100 names showed no inclusion impulse in the first session after the announcement, while BE showed a very large one — and how much of BE's +6.26% mechanical buying can explain is not clear.

  Table scrolls horizontally
• The real reason not to recommend it comes back to valuation and crowding: already up more than 150% YTD, while passive buying concentrates at the 9/18 close with only two weeks of window left. ILMN, also joining the S&P 500, is up only 1.15% this morning — a far lower degree of pricing than BE.
• Pre-market: +6.26%, $268.70, notional $290.7M (real volume). The 52-week range is $52.00–$351.28.
• Final judgment: watch only. The event is real, but it has already been front-run and only two weeks of window remain; a +150% YTD gain is too thin a valuation cushion, and support disappears once the passive buying lands. If you must play the index-inclusion line, ILMN (+1.15%) is far less priced than BE and has better risk/reward — see §7.3.

5.8 INTC (Intel) — watch only | the largest pre-market money in the market, paired with the weakest evidence chain

• Related news: ① DIGITIMES, citing the supply chain, says it plans to raise PC CPU prices by about 10% on 10/5, aiming to improve gross margin rather than win share, and that this would be the third round of increases since late 2025 (about +10% in 2026 Q1, with some product lines raised again in July); ② Northland upgraded it to Outperform with a $120 target.
• The problem with the evidence chain: SEC EDGAR shows Intel's most recent 8-K remains 8/12, with no company disclosure since — the price-increase story rests entirely on a single supply-chain outlet. More importantly, the same DIGITIMES piece also writes that Qualcomm and MediaTek are eroding Intel's IPC and IoT share. The combination of "a third consecutive price increase + share loss" can be read either as gross-margin repair or as a passive defense trading volume for price; reading only the first half is selective citation.
• The real reason for the Northland upgrade is Terafab, and there is a commonly confused name here: Terafab is not an Intel plant but a roughly US$25 billion chip joint venture by Tesla / SpaceX / xAI at Giga Texas; Intel announced on 2026-04-07 that it would be its foundry partner, with the counterparty adopting Intel 14A. ⚠️ But the contract's value, prepayments, capacity commitments and whether there is any take-or-pay cannot be found as quantified disclosure anywhere in Intel's 10-Q / 10-K — a "fact" capable of supporting tens of billions of dollars of market value currently exists only on a media basis.
• Dilution (primary source 424B5, fully worked out): offering price $95.00, base offering 210,526,315 shares = US$20.0 billion, greenshoe of 31,578,947 shares fully exercised on 8/11, for a total of 242,105,262 shares = US$23 billion gross / US$22.62 billion net. Pre-offering share count was 5,043M ⇒ 4.80% dilution, with a post-offering count of 5,285.07M shares. ⚠️ What deserves more attention is potential dilution, which is twice the dilution already incurred: the 424B5 footnotes disclose another 143 million shares escrowed to the US Department of Commerce (released alongside CHIPS "Secure Enclave" funding), 241 million share warrants to the Department of Commerce, and 123 million unvested RSUs/PSUs and options, totaling about 507 million shares ≈ 9.6% of the post-offering share count. The exercise condition on those 241 million warrants is that "Intel directly or indirectly holds less than 51% of the foundry business" — effectively attaching a penalty clause to the common bull argument that "spinning off the foundry unlocks value."
• Valuation: this is not a "cheap turnaround." GAAP net income carries no information under the current structure — Q2'26 GAAP net loss was US$11.033 billion, but operating profit was actually positive at US$1.796 billion; the loss came almost entirely from a −US$12.529 billion fair-value remeasurement of the escrowed-share derivative liability, and that liability is marked to Intel's own share price — the harder the stock rallies, the larger the GAAP loss. On a valid basis:

  Metric (at the 9/4 close of $95.80)	Value
  TTM revenue / gross margin	US$57.03 billion / 38.6%
  TTM operating profit	−US$77 million (essentially breakeven)
  TTM non-GAAP EPS	$1.09 → P/E 88x
  2027 consensus EPS $2.04	forward P/E 47x (⚠️ that estimate is a third-party aggregated secondary basis, not verified against a primary source)
  P/S 8.9x · P/B 4.6x · EV/EBITDA 41x	all at the top of Intel's historical valuation range (P/B was below 1x for long stretches in 2024–25)

  Table scrolls horizontally
• Operating leverage really is improving (this must be said fairly): revenue went from US$13.577 billion in Q1'26 to US$16.128 billion in Q2'26 (+18.8% QoQ), with an incremental gross margin of 45.6%, above the existing gross margin; Foundry's quarterly loss has narrowed for three consecutive quarters (−US$2.509 billion → −US$2.437 billion → −US$2.089 billion). But the slope is not enough: Foundry still lost US$9.356 billion on a TTM basis, an annualized improvement of only about 10%, so at this pace foundry breakeven is not within range in 2027; for now the product division is feeding the foundry.
• The magnitude of a 10% price increase (computed in this piece, with assumptions listed): assuming the increase flows fully into price, volume is unchanged and incremental cost is zero, on the CCPG segment's Q2 revenue base of US$8.877 billion — the effect on consolidated gross margin is +1.6 to +3.1pp, and annualized non-GAAP EPS about +$0.28 to +$0.60 (14%–29% of the 2027 estimate of $2.04). The magnitude is substantive rather than noise, but it also falls far short of independently explaining a +5% pre-market move. The revenue-neutral volume breakeven point is −9.1%. ⚠️ Two qualifiers: ① the original text says "another 10%" — this is the second/third round, and the effect of the previous round is already in Q2's 40.4% gross margin, so it is not pure incremental; ② Q3'26 guidance was issued on 7/23, before this rumor, so the guidance does not include it, and an effective date of 10/5 means Q3 will see channel pull-forward stocking and Q4 carries payback risk.
• Pre-market: +4.94%, $100.53, pre-market notional $461.3M — the largest in the market, on 4,588,355 shares. It was already +4.51% on 9/4, for a two-day cumulative +9.67%. The 52-week range is $23.75–$142.35.
• Final judgment: watch only. The size of the pre-market money is severely mismatched with the strength of the evidence: US$461 million of pre-market notional built on an unconfirmed supply-chain rumor, in a stock already up 9.67% over two days. This is the textbook case of the iron rule "do not chase a stock that has already gapped up sharply." If the price increase is formally confirmed by the company in October, entering then will carry far more certainty than today.

5.9 The storage chain (MU / SNDK / WDC / STX) — avoid chasing | no second confirmation as of 08:10 ET

• Related news: none. There is no new company-level or industry-level catalyst in the storage chain this morning.
• This piece's core observation: on Friday (9/4) the storage chain rallied together — SNDK +11.90%, KLAC +7.32%, MRVL +7.05%, MU +6.10%, STX +6.34%, WDC +5.86%. The previous pre-market list flagged "whether that big green candle can be confirmed by a second one" as the core open question. Today's answer is: it has not been confirmed, and the evidence against it is very clean —

  Ticker	9/4 move	This morning's pre-market	Pre-market notional
  MU	+6.10%	+0.48%	$1,174.0M (1st in the market)
  SNDK	+11.90%	+0.11%	$1,149.6M (2nd in the market)
  WDC	+5.86%	+0.65%	$35.8M
  STX	+6.34%	+0.88%	$12.2M
  MU and SNDK together account for US$2.3 billion of pre-market notional turnover, the two largest in the market this morning, and it bought price changes of +0.48% and +0.11%.
  ⚠️ But this piece does not elevate that into the conclusion of "distribution," for four reasons: ① "every trade has an equal buyer and seller" is a tautology and implies no direction; ② distinguishing distribution from accumulation requires trade-direction data, which this piece does not have; ③ this pre-market window spans a weekend plus Labor Day, so the accumulation period is far longer than a normal overnight and "largest in the market" contains a calendar factor; ④ the opposite reading is equally self-consistent — holding Friday's gains on heavy turnover without giving them back can also be read as the market accepting the 9/4 re-pricing.
  ⇒ The only safe statement is: as of 08:10 ET, Friday's big green candle has neither been confirmed by a second one nor been negated. That is enough to support "do not chase," but not enough to support "it has already topped."
  ⚠️ One overlap in attribution must also be declared: SNDK's +11.90% on 9/4 is treated as the storage trade in this section, while §5.7 also refers to that day in discussing index inclusion — the S&P announcement came after the close on 9/4, so that day's gain belongs to storage, not to the index event, and the two must not be double-counted.

  Table scrolls horizontally
• The correct reading of SOXX +1.28%: this morning's semiconductor ETF gain is not contributed by the storage chain but carried by three stocks that each have their own independent news — INTC (+4.94%, price-increase rumor), ASML (+3.33%, High-NA intent) and TSM (+1.54%). Strip out those three and semiconductor breadth is flat.
• One piece of contrary evidence, honestly labeled attribution-unknown: Japanese semiconductor equipment names fell hard today (Disco −5.70%, Tokyo Electron −3.30%, Advantest −1.97%), the opposite direction from European and US equipment names this morning (ASML +3.33%, AMAT +1.60%, LRCX +1.74%, KLAC +1.49%). This piece found no clean reason that explains both sides at once and will not invent one.
• Final judgment: avoid chasing, downgrade to watch. This is not a bearish industrial view on storage but a statement that there is no executable edge at today's price.

5.10 ORCL (Oracle) — avoid | it is a sector line, but the company-level driver is still unknown and earnings are imminent

• Company-level driver: not found. SEC EDGAR shows Oracle's most recent 8-K remains 2026-06-10, with no company disclosure since; searches for rating or order news dated 9/8 also returned nothing.
• But its sector membership is clear (see §1.1④): ORCL belongs to the "AI infrastructure/neocloud" group, and that group rose together this morning (CRWV +2.31%, NBIS +2.13%, VRT +1.24%, GEV +0.51%), while mega-cap tech and software broadly fell at the same time (MSFT −0.67%, AMZN −1.05%, GOOGL −0.70%, IGV −0.55%). The most popular explanation, "rate-driven," found no support this morning (long end each about +2bp, TLT −0.06%, the opposite direction; ⚠️ but both numbers are within noise, and credit spreads were not obtained, so one can only say "unsupported" rather than "falsified"). ⇒ It is not a lone stock, but this piece could not find the group's common catalyst.
• One point where search results can easily mislead: items retrieved for "Oracle OpenAI US$300 billion," "backlog US$523 billion" and "shares +43% in a day" correspond to the September 2025 earnings print, not this year's. Explaining today's pre-market with last year's narrative is wrong.
• ⚠️ And do not talk only about a "54% drawdown": ORCL's 52-week low is $114.50 (this July), and the 9/4 close of $158.78 is already +38.7% off that low (or +45.6% measured against this morning's pre-market price of $166.71). Talking only about the drawdown from the high describes a stock that has already bounced nearly 40% as if it were in a one-way decline.
• How it fell from $345.72 — there was no "blow-up earnings print"; the trigger was the cash flow statement: the peak of $345.72 on 2025-09-10 was exactly the day after the FY26Q1 print (when RPO first exploded higher). The real turning point was FY26Q4 on 2026-06-10: RPO of US$638 billion (+363%) and EPS +24% were both beautiful, but the same statements showed FY26 capex of US$55.66 billion (+162%), capex/revenue of 82.6%, free cash flow of −US$23.69 billion and total debt of about US$130 billion. It then fell 19% in the single week of 6/26 (its worst since 2001), with the 5-year CDS spread rising to a record high of about 198bp. The market switched from an "AI revenue story" to an "AI credit story" — what is moving is the CDS, not EPS. The multiple has compressed from about 45x back to 20.8x TTM non-GAAP, back into the 18–25x range ORCL occupied for years before the AI re-pricing.
• Two sides of the fundamentals (10-K primary source): FY26 revenue was US$67.36 billion (+17%), with OCI at US$18.1 billion for the year, +77%, and Q4 alone +93% — the growth is real. But only about 12% of the US$638 billion of RPO will be recognized in the next 12 months, with 34% in months 13–36 and 34% in months 37–60. The backlog's duration is extremely long, while the capex supporting it has to be spent now. Net debt is about US$98.2 billion (plus US$26.6 billion+ of operating leases not included), net debt/EBITDA is 3.28x, and interest coverage has already fallen from 4.48x in FY26 to 4.26x in Q4, with roughly US$40 billion more to refinance in FY27. Cloud cost growth (+52%) continues to outrun cloud revenue growth (+39%), so scale has yet to deliver unit-economics improvement.
• Pre-market: +4.99%, $166.71, notional $242.8M (an 08:23 ET reading). It was already +3.08% on 9/4, for a two-day cumulative of about +8.2%.
• Event risk and a trap that must be avoided: FY27Q1 is released after the close on 9/10 (Thursday). The company's own 6/10 guidance is revenue +27%~29% (about US$18.96 billion–US$19.26 billion) and non-GAAP EPS of $1.72–$1.76. The widely cited "consensus revenue of US$19.13 billion" is exactly the midpoint of that range — an echo of guidance, not an independent estimate; while the other circulating figure, "consensus EPS $1.299," differs from the company's non-GAAP guidance by 26% and is almost certainly a GAAP basis. Treating $1.299 as the non-GAAP hurdle would manufacture a "massive beat" out of thin air when the print lands. This piece therefore lists only the company's original guidance and uses no single "consensus" number.
• Final judgment: avoid. The sector membership is now clear, but the company-level driver is still unknown + it is already up about 8.2% over two days + earnings are two days away + the market is pricing it on credit rather than software. In front of a balance sheet that is already stretched and a backlog with extremely long duration, chasing an 8% move into earnings for reasons you cannot articulate means taking unknown news risk and known event risk at the same time.

5.11 BSX (Boston Scientific) — short watch | the decline does not match the severity

• Related news (09-08, SEC 8-K primary source): unauthorized network activity was discovered on 8/25, causing disruption to manufacturing, order processing and shipping globally. The company says it is "unlikely to achieve" the Q3 and full-year net sales growth and adjusted EPS guidance ranges given on July 29, 2026, and explicitly states the incident will have a "material impact" on Q3 and the full year. The distribution network and manufacturing/sterilization facilities have largely been restored, but the timeline for full recovery remains uncertain; updated guidance will not come until the 10/28 third-quarter earnings call. The company says it expects no material impact on its long-term financial position.
• Catalyst logic: under this piece's iron rules, a guidance withdrawal is more serious than any single-quarter beat/miss. Here we have a complete guidance withdrawal + an admission of material impact + a seven-week information vacuum (today to 10/28). The information vacuum is itself a source of risk premium: before 10/28, the market cannot know how large the shortfall is.
• The mismatch in the market's reaction: only −2.38% pre-market ($46.66, notional $10.1M). It actually rose 1.81% on 9/4, for a two-day total of just −0.76%. An explicit full-year guidance withdrawal has bought less than a 2.4% decline — a possible explanation is that the 8/25 incident was already partly priced, but "the incident is known" and "the guidance withdrawal is known" are two different things, and the latter is new information that only arrived this morning.
• Final judgment: short watch (not a short recommendation). The logic is "the gap between the severity of the news and the price reaction," not a long-term view on the company — the company itself says it expects no long-term impact. Verification point: if the decline still converges within 3% after the open and volume does not expand, the market really has digested it and this watch is void.

---

6. Bearish / Avoid List

---

7. Rankings Within Themes

7.1 Clinical-Data Re-pricing (today's strongest theme)

7.2 Energy / Geopolitics

7.3 Index Rebalancing (effective 9/21)

7.4 Semiconductors

---

8. Open-Verification Signals

8.1 Pre-market (now to 09:30 ET)

• Price discovery in the mispriced pharma names is today's main battlefield: the pre-market notional of PHVS, ROIV, DYN, SRPT and IONS is all below US$18 million, so pre-market prices carry almost no information. Do not make any decision off pre-market prices; wait for real matching after 09:30.
• AMGN ($46.9M) and NVS ($53.5M) are the only two pharma names with real volume, and their opening prices are the pricing benchmark for this event.
• For the storage chain, watch whether huge volume can convert into gains: MU and SNDK already have more than US$1.1 billion of pre-market notional each. If volume is still high and gains still low 30 minutes after the open, the "distribution" judgment in §5.9 holds; if they break Friday's highs on expanding volume, that judgment needs to be downgraded.

8.2 Intraday (09:30–10:00 ET)

• Whether the Dow–S&P spread converges: this piece's attribution says 40% of the Dow's weakness comes from AMGN. If AMGN's decline narrows after the open and the Dow spread narrows in step, the attribution is confirmed; if AMGN rebounds while the Dow still materially underperforms the S&P, there is a second macro factor this piece has not identified — that is a falsifiable test.
• Sector ETF linkage: the declines in XLV (−1.10%) and XBI (−0.99%) are far smaller than the carnage at the single-stock level (NVS −12.5%, DYN −29%), which says this is a single-stock event, not systematic sector risk. If XBI's intraday decline widens beyond −2.5%, the character upgrades to a space-level contraction in risk appetite.
• Energy: do not treat the 1.3pp gap as a signal yet. XLE/Brent = 0.48 sits inside the normal beta of energy equities to oil (0.4–0.6), and this piece did not obtain a rolling beta benchmark, so that gap is not falsifiable. The genuinely falsifiable observation is the fundamental one: whether Saudi Arabia confirms that crude exports and Abqaiq/Ras Tanura are unaffected (see §5.6).

8.3 Options and Volatility

• VIX at 15.66 is trading live again (the previous piece could only use 14.53 from 9/4 because of the Labor Day closure). ⚠️ The +2.35% shown by the quote API is computed against that phantom 15.30 print from Labor Day with no trading; against the real 9/4 close of 14.53 it is actually +7.8%. The absolute level of 15.66 is still mild (not panic), but the increase is not mild — do not use the +2.35% figure to support a "the market is calm" narrative.
• IV in PHVS / ROIV / DYN will be extremely high today: the data are already out, so IV crush risk is immediate, rather than waiting for after hours as with earnings. Expressing these directions through options has very poor value for money.
• IV in ORCL and ADBE will keep climbing into the 9/10 earnings, so buying directional options today means bearing both time decay and IV crush.

8.4 Risks

• Reversal after a gap: INTC (+9.67% over two days), BE (+14.07% over two days), ORCL (+8.06% over two days) and PHVS (+17.79%) are all in a "already gapped sharply" state, and a gap-fill after a higher pre-market open is the most common path for this pattern.
• A lone riser with no sector follow-through: INTC is +4.94% while AMD is +1.26% and NVDA +0.57% — INTC is a lone name, not a semiconductor sector move. Likewise BE's rise has no comparable-company follow-through.
• The thin-volume trap: a large number of key names today have pre-market notional below US$10 million (MPC $1.2M, DK $0.05M, BCRX $0.8M, TAK 257 shares, ARGX 248 shares). This piece has labeled these as noise, and readers should draw no conclusions from them.
• The wait-and-see effect of a macro vacuum: there is no important data from today through Thursday, and August CPI on Friday 09/11 at 08:30 ET is the decisive vote on the hiking path. Trends during a vacuum usually lack durability, so today's sector direction should not be extrapolated into the back half of the week.
• There is a 3-year Treasury auction at 13:00 ET today; a weak result would push up short-end yields and add pressure on high-multiple growth stocks.

---

9. Final Conclusions

① The 5 Names Most Worth Watching Today

This list deliberately spans 5 mutually unrelated driver variables (single-company clinical data / mechanism extrapolation across a space / mechanical index buying / geopolitical supply / the semiconductor industry cycle), so that five positions are not all betting on the same thing.

② The 3 Strongest Themes Today

③ Directions to Avoid Today

1. Chasing the storage chain (MU / SNDK / WDC / STX) — Friday's big green candle was not confirmed this morning, and it was not confirmed in the most adverse way possible: "the two largest pre-market money flows in the market (US$2.3 billion) bought +0.48% / +0.11%." Huge volume with no gain is absorption, not accumulation for a move.
2. Front-run index-inclusion names (BE) — ⚠️ the first draft was wrong to treat 9/4's gain as "front-running" (the S&P announcement came after the close on 9/4, see §5.7); after correction, what is attributable to the inclusion for BE is this morning's +6.26%. The reason to avoid comes back to valuation and crowding: already up more than 150% YTD, with the passive-buying window running only to the 9/18 close. ILMN, also joining the S&P 500, is up only +1.15% this morning, a far lower degree of pricing.
3. Rumor-driven large gaps (INTC) — US$461 million of pre-market money built on an unverified rumor from a single supply-chain outlet, with no company disclosure since 8/12, in a stock already +9.67% over two days.
4. Attribution unknown + ahead of earnings (ORCL / ADBE) — stocks that nobody can explain the rise of, facing an earnings test in two days.
5. Thin, small-cap DM1 platform extrapolation names (DYN / SRPT / PEPG) — the direction may be right, but judging "it has fallen too far" on US$0.4M–US$8.7M of pre-market notional is itself unreliable. Do not catch the falling knife.
6. Anything with pre-market notional below US$10 million — there are unusually many of these today, and their pre-market quotes are noise, not prices.

④ Final One-Sentence Judgment

Today is a bottom-up day: the most information is concentrated in primary clinical data on a handful of single stocks, not at the index level — the Dow looks 0.75pp weaker than the Nasdaq, but bottom-up attribution shows 40% of that is one stock, AMGN, and reading it as a change in the macro wind would be badly wrong; the three things that actually require execution discipline are: do not chase the rumor stock that has already gapped (INTC) or the crowded inclusion name (BE), do not price pharma stocks on thin pre-market volume, and do not assume Friday's big semiconductor green candle is still the running leadership theme — as of 08:10 ET it has been neither confirmed nor negated, and the US$2.3 billion of pre-market turnover in MU and SNDK says precisely that there is disagreement at this level. The real macro test is CPI at 08:30 ET on Friday 09/11, and no sector direction before then should be extrapolated.

---

Operations Notes (not sent to clients)

Data collection and tooling status this run:

• Every WebFetch to CNBC.com returned 403 (tried both the TSMC/ASML and the Novartis pieces), the Benzinga premarket page likewise 403, and the Reuters energy section refused. The content of these pieces was obtained instead via search summaries plus accessible mirrors (Techzine, Investrade, StockTitan, Business Standard), and all key conclusions were traced back to primary sources (Novartis via the SEC 6-K, BSX via the SEC 8-K, the index changes via the S&P official press release).
• Both Novartis newsroom URLs 404 (/news/media-releases and /newsroom/media-releases). Used the SEC EDGAR 6-K (f6k_090826.htm) instead, which is actually more authoritative than the corporate site; HARBOR's enrollment count, endpoint definitions and the "reaffirmed five-year CAGR guidance" all come from that document.
• stockanalysis's screener JSON API is dead (/api/screener/... all 404). Switched to: ① parsing the HTML <table> on /markets/premarket/{gainers,losers}/ directly for the pre-market gainers/losers lists; ② https://stockanalysis.com/api/quotes/s/<TICKER> for single stocks (this endpoint works, returning p/ep/ecp/epv/h52/l52 with a td trading-day field). . This quotes endpoint performed very well this run and should be locked in as the primary channel for the pre-market slot (the alternative to the memory note "CNBC extended-hours fields are unavailable pre-market" is now empirically confirmed).
• US Treasury yield CSV: field_tdr_date_value=2026&all returns January data; you must use field_tdr_date_value_month=202609 to get September. Already fixed in the script. Also confirmed that the absence of a 9/7 data row = the bond market was closed, which is very clean machine evidence of a holiday and is reusable.
• CNBC Korean symbols cannot be fetched: 000660.KS-KR / 005930.KS-KR (SK Hynix, Samsung) return all None, so Korea was covered only at the KOSPI index level and no constituent-level storage verification was possible. The Japan (8035.T-JP etc.) and Taiwan (2330-TW) formats are fine. Another channel is needed next time Korean single stocks are required.
• The three tickers HES, KALV and CSLLY 404 on the quotes endpoint (HES may have been delisted following its merger). No effect on conclusions.
• yfinance was not used in the main flow this run, and both fundamentals-analyst sub-agents were warned about rate-limit risk and the stockanalysis backup channel.

Known evidence gaps in this piece (left for the recap to fill): Two substantive corrections brought back by the sub-agents (both changed conclusions, worth remembering):

• The "unpriced" conclusion for pelacarsen holds only for US equities and is wrong for NVS itself — the SIX Swiss exchange was open as usual on US Labor Day and already priced it at −3.18% on Monday. The first draft nearly wrote NVS's arithmetic ADR catch-up up as "an unpriced opportunity." The implied FX rate from ADR ÷ Swiss line differs by only 2bp over two days; this check is very handy and should be locked in.
• The first draft treated ASO vs siRNA as a firewall for olpasiran and wrote the conclusion as "undecided," which is wrong. The failure shape is "Lp(a) came down but events did not" = a translation failure, the most extrapolable kind; the two drugs act on the same node via the same delivery route. Changed to "extrapolability high but not complete."
• The first draft wrote mosliciguat's indication as PAH when it is actually PH-ILD. That error would directly lead to benchmarking the wrong company (looking at Winrevair instead of Tyvaso) and therefore missing UTHR, today's cleanest unpriced lead.
• Attributing SRPT's −10.84% to a DM1 read-across is very likely wrong (SRP-1003 is only a 78-patient Phase 1/2 with safety as its primary endpoint).

Evidence gaps that remain in this piece (left for the recap to fill):

1. The ≥90 mg/dL subgroup result from Lp(a)HORIZON plus the Lp(a) reduction pelacarsen actually achieved — the Novartis CMO's wording left in the qualifier "in the overall study population." Those two numbers alone very nearly determine olpasiran's survival probability, and they await a medical conference.
2. The quantified terms of the Terafab contract for Intel (value/prepayment/capacity/take-or-pay) cannot be found in the 10-Q or 10-K — yet it is the core reason for Northland's upgrade and the pillar of the 47x forward valuation.
3. No disclosure of capitalized interest could be found in the filings for either ORCL or INTC. ORCL has construction in progress on the order of US$100 billion and Q4 interest annualizing at US$5.75 billion; if a large amount of interest is being capitalized, the 4.26x coverage ratio is flattered.
4. INTC's 2027 consensus EPS of $2.04 is a third-party aggregated secondary basis, not verified against a primary source (⚠️ already flagged in §5.8).
5. PHVS's cash balance and runway were not obtained; §5.1 states "cannot be confirmed."
6. Whether the primary endpoint of the PHrontier Phase 3 is 6MWD or PVR has not been disclosed by the company — that determines the correlation between the "56%" being priced today and Phase 3 success or failure.
7. BSX's 52-week high of $109.50 vs the current price of $47.80 is questionable (a stock split during the period is suspected, and h52 may not be adjusted). This piece therefore deliberately avoided phrasing like "down 57% from the high," so as not to make an erroneous drawdown claim without verifying the split.
8. Japanese semiconductor equipment names fell hard today (Disco −5.70%) in the opposite direction from strength in European and US equipment names; attribution unknown; the common catalyst for the "AI infrastructure/neocloud" group rising together was also not found. §0.4, §0.9 and §5.9 all label this honestly, with no fabricated explanation.

risk-auditor QC results: substantive errors already corrected (these were wrong in the first draft — note them and do not repeat)

• The index-inclusion timing error (most serious): the S&P announcement came after the close on 9/4, so BE's +7.35% on 9/4 and SNDK's +11.90% on 9/4 both happened before the announcement and cannot possibly be "front-running the inclusion." The whole argument the first draft built on that ("a control group from the same week and mechanism already proved the impulse fades in a day") is void and has been rewritten to use only data from the first session after the announcement, with a note added that the passive AUM of the S&P 100 and the S&P 500 differ by orders of magnitude and they cannot serve as controls for each other.
• "A 2bp residual" is circular: the implied-FX residual is algebraically identical to the difference between the two percentage moves and introduces no external information; and the ADR pre-market was quite possibly quoted mechanically by market makers off the Swiss line. It has been downgraded, and the "unpriced" conclusion is now explicitly carried by three independent pieces of evidence: the 8-K timestamp, IONS at −0.07%, and the market holiday.
• "The rates explanation has been falsified" used the wrong variable: this piece itself characterizes the group as "sensitive to credit spreads," yet used Treasury yields to falsify it, and both ±2bp and −0.06% are within noise. All changed to "unsupported," with an explicit statement that credit spreads were not obtained.
• The XLE beta assumption was wrong: XLE/Brent = 0.48 already sits within the normal beta range for energy equities (0.4–0.6), and the first draft's treatment of it as "a signal of disbelief" implicitly assumed beta=1. Corrected, and the unfalsifiable "gap convergence" observation in §8.2 downgraded.
• VIX's +2.35% is computed against the Labor Day phantom print; against the real 9/4 close of 14.53 it is actually +7.8%. This one nearly supported an erroneous "the market is calm" narrative.
• MPC's Chinese name was written as "Marathon Oil" — that is the upstream company acquired by COP; MPC is Marathon Petroleum (the refiner), and the entire argument depends precisely on "it is a refiner."
• Dow attribution: AMGN's price triple was taken from different read times in different sections (−6.24%/−$27.28 vs −6.00%/−$26.23), now declared at the top of the piece; "96% agreement" has been deleted because futures and cash are different rulers; and the divisor of 0.168, having no primary source, is flagged pending verification.
• UTHR self-contradiction: this piece set an iron rule that "notional below US$10 million is noise," and then built the No. 2 recommendation slot on 1,992 shares. Rewritten as "a hypothesis to be tested, with zero pre-market evidentiary strength," with the added discount that the 6MWD comparison is "a Phase 2 secondary endpoint vs a Phase 3 primary endpoint, different rulers."

Parts confirmed by QC as requiring no change: the full INTC arithmetic (back-computation, 4.80% dilution, 9.6% potential dilution, price-increase sensitivity, the −9.1% volume breakeven, the 9% annualized Foundry improvement), the treatment of AMGN's distorted ROE/PB and the switch to ROA, the EV-basis conversions for DYN/ROIV, the three pieces of hard evidence for pelacarsen being "unpriced in US equities," the three primary-source checks in §1.1②③⑤, the two corrections on SRPT and mosliciguat, section and table numbering with no duplicates or skips, and no fabricated numbers.

Still unclosed, left for the recap: ① the precise release time of the S&P announcement; ② the two names Everpure and HONA still have no ticker, which is a status-type fact and needs word-by-word checking against the S&P official press release (transcription risk); ③ ORCL's 198bp CDS and the "−19% in the week of 6/26" both lack a primary source, and the former is the only quantitative pillar of the "priced on credit" narrative; ④ DYN's 10-Q filing date and the as-of date of its runway statement were not obtained, and if that 10-Q was filed after the July offering then "runway does not overlap with the Phase 3" may not hold; ⑤ the gap between IONS's "already received >US$275 million" and "upfront + license fee + development milestones = US$250 million" implies the amount received includes regulatory milestones, in which case "at most US$650 million going to zero" involves double counting; ⑥ the entire options discussion in §8.3 has no IV/skew figures behind it and is inference written as statement.

---

⚠️ Risk disclaimer: this list is only a pre-market information review and watchlist and does not constitute investment advice. US equities carry high volatility and pre-market gap risk, and post-earnings IV crush and guidance reversals are possible; automatically generated content may contain stale information or factual errors, so please rely on company disclosures and SEC filings, and do not use this directly as a basis for trading.