U.S. Pre-Market Brief | 2026-09-10 (ET) Thursday

Coverage window: 2026-09-09 16:00 ET regular-session close → 2026-09-10 08:37 ET. Includes 9/9 after-hours earnings, the overnight Asia/Europe sessions, this morning's pre-market, and the August PPI and initial jobless claims released at 08:30 ET.

Price conventions (a read timestamp is given at every point — do not compare across timestamps):

• Single-stock/ETF pre-market prices, percent changes and pre-market volume come from the Nasdaq official quote API (api.nasdaq.com/api/quote). Previous closes are reverse-computed locally (pre-market price − net change) and cross-checked name by name against the percent change the API returns; all are self-consistent. ⚠️ The API's secondaryData field cannot be used as the previous close — in this run USO / XLK had secondaryData sharing the same baseline as the pre-market price, so using it directly produces wrong numbers.
• Index futures, live Treasury yields, crude, precious metals and industrial metals come from CNBC, with read timestamps marked at each point; these are last electronic-session trades, not settlement prices.
• Treasury levels and direction: the benchmark is the official Treasury yield-curve CSV (15:30 ET fixing); live pre-market values come from CNBC. ⚠️ CNBC's 2Y change_pct field again contradicts itself today (last 4.487% is above its own previous close of 4.427%, yet it shows a negative change), so this brief does not trust that field anywhere; everything is computed locally as "live value vs the prior-day Treasury fixing."
• Crack spreads are computed locally: same-month contracts, ×42-gallon conversion, 3-2-1 weighting, both legs read from the same source at the same moment.
• Company financials come from primary SEC 8-K / 10-Q / 10-K filings, marked at each point; anything tagged "secondary source," "to be verified" or "not obtained" has not been primary-verified.

⚠️ Pre-market volume and price are cumulative; a single snapshot is not a fact. For the key quantities that feed conclusions, this brief took two readings — before PPI (08:07–08:19) and after PPI (08:32–08:37) — and uses the difference between them for the attribution breakdown in §1.3. Values will still change before the open.

---

0. One-Line Summary

1. Every asset today sits on the same causal chain, and the chain starts at the Strait of Hormuz. The U.S. military destroyed 5 Iranian oil tankers on 9/8 and the Houthis struck Saudi Arabia's Jazan refinery (400,000 bpd); WTI rose to $99.31 (+3.39%) and Brent to $104.45 (+3.20%) (08:24 ET). That supply shock showed up directly in this morning's August PPI: diesel +24.1%, energy +4.2%, accounting for more than three-quarters of the rise in the goods component.

2. ⭐ Today two central banks/statistical agencies said the same thing twice. At 08:15 ET the ECB raised all three of its key rates by 25bp (deposit facility 2.50%, main refinancing 2.65%, marginal lending 2.90%, effective 9/16), and the stated reason in the statement is: "the conflict in the Middle East continues to generate inflation pressures, and inflation is set to remain well above target for an extended period" — the ECB itself attributes the hike to the Middle East conflict. Fifteen minutes later at 08:30 ET, U.S. August PPI delivered the American version of the same chain: +0.4% month over month (in line with expectations), +5.4% year over year; core (ex food/energy/trade services) +0.3% m/m, +4.7% y/y; and more than three-quarters of the gain in the PPI goods component came from energy, with diesel alone +24.1%. At the same time initial claims came in at 206,000 and the insured unemployment rate at 1.2% — high inflation plus a still-tight labor market. The 10-year yield was 4.893% pre-market (the Treasury's 9/9 fixing of 4.83% was already the year's high), the 2Y 4.487%, and VIX 17.46 (+6.08%). Next Tuesday, 9/16, is the FOMC, and the market is pricing a hike, not a cut.

3. Driver type: geopolitics → macro (PPI/CPI) → rates, and at the single-stock level, earnings that failed to deliver. There is no significant new catalyst today beyond the macro data; the real events come tonight: ORCL and ADBE both report after the close.

4. Money is flowing "sell AI capex, buy physical energy," and it is an unusually clean mirror image: among the energy names covered here, all 10 for which a valid previous close was obtained are up, and COP / CVX / MPC / VLO have pre-market prices already above their respective 52-week highs (pre-market prices do not count toward the official 52-week high/low calculation and need confirmation after the open); meanwhile semis SMH −2.02% and AI power (OKLO −2.37% / SMR −2.50% / VRT −2.31% / GEV −1.87%) are down across the board. But this is not a "broad commodity reflation trade" — copper −4.79%, silver −4.06%, gold −1.15%. Only oil is up. This is a supply shock, not a demand recovery.

5. Pre-market state (08:32–08:37 ET): S&P futures −0.21%, Nasdaq futures −0.76%, Dow futures +0.02% (the divergence among the three is itself today's conclusion); SPY −0.36% / QQQ −0.98% / IWM −0.51% / DIA −0.17%; dollar index 99.14 (+0.33%); the overnight Asia/Europe sessions did not follow tech lower — Nikkei +0.20%, KOSPI −0.25%, Hang Seng −1.27%, the three major European indices −0.10% to −0.34%. Asian semis did not crash, which means this morning's selloff in U.S. semis originated in the U.S. session itself; it was not transmitted overnight.

---

1. News Overview

1.1 Main Headlines (by impact level)

1.2 On the attribution "memory stocks fell because Apple was cleared to buy Chinese memory" — this brief does not adopt it

Several articles attribute the drop in memory stocks to "the Trump administration possibly allowing Apple to source memory chips from CXMT / YMTC." After tracing publication dates item by item, this brief concludes that attribution cannot explain this morning's pre-market, for three reasons, each independently verifiable:

Only one exclusion reason actually holds up: the date anchor. The timestamps on the relevant reports fall on 7/24 (Apple and Micron argued at the White House), 7/30 (senators applied pressure), 8/14 (Commerce Secretary Lutnick said he "does not support it"), plus one leak-style paraphrase claiming it "will be cleared after the September Trump-Xi meeting." Several secondary articles write that "memory stocks fell sharply on Monday" — but 9/7 was the Labor Day holiday (the Treasury yield-curve CSV has no 9/7 record between 9/4 and 9/8, which cross-verifies this). This is the classic pattern of writing a prior session's move as today's move.

⚠️ The first draft listed two additional "exclusion reasons" here; on review one of them was judged wrong, and it is withdrawn and explained below:

The draft wrote "the transmission direction is backwards — if Chinese fabs expand capacity, semiconductor equipment should benefit, yet equipment fell the hardest this morning, so it doesn't add up." That reasoning is wrong: CXMT and YMTC are subject to Entity List restrictions and cannot buy U.S. equipment in the first place. If Apple's orders shift from "Micron/Samsung/Hynix, who do buy LRCX/AMAT/KLAC" to "Chinese fabs, who do not buy U.S. equipment," then equipment falling harder than memory is exactly consistent with that narrative, not contradictory. The reason is withdrawn.

The draft also wrote "the official status of this event is still rumor" — that confuses "is the news true" with "can the news move prices": a rumor can absolutely be priced. Downgraded to background context, not used as an exclusion reason.

⇒ Revised conclusion (weaker than the draft): every publicly verifiable date anchor for that attribution is old news, and this brief did not obtain evidence of a new leak or official development this morning; but it also cannot be ruled out. So on this morning's semiconductor decline, this brief's conclusion is "attribution unconfirmed," not "that attribution has been ruled out." See §1.3.

1.3 Breakdown using the two readings around 08:30 (downgraded because of the ECB — read together with the warnings below)

⚠️ ⚠️ The flaws in this table must be stated first, or it will be over-used. (a) The window is contaminated by the ECB. The ECB hiked at 08:15 ET, which falls squarely inside the read window for the "before PPI" column. The actual read timestamps in that column are: XLK/IGV 08:03, SPY/QQQ/IWM 08:06, TLT/SLV 08:08 (all before the ECB), but SMH and XLE at 08:18–08:19 (after the ECB). This column is neither a single moment in time nor a clean "pre-event" reading. (b) PPI and initial claims were released simultaneously at 08:30 and cannot be separated; moreover, PPI at +0.4% m/m was in line with expectations and does not necessarily constitute a "surprise." ⇒ The table below can therefore only be read as "what happened between two readings," not as "what PPI caused."

The same set of instruments, read once at 08:03–08:19 ET and once at 08:32–08:33 ET:

What this table can support is far weaker than what the first draft wrote. Point by point:

• What can be said: between the two readings, risk assets broadly drifted lower (SPY −0.19pp, QQQ −0.30pp), tech fell more (IGV −0.56pp, SMH −0.51pp, XLK −0.48pp), and energy was the only sector that strengthened (+0.33pp), while silver fell the most (−1.34pp). The direction is consistent with "rising inflation/rate expectations plus rising oil."

• ⚠️ You cannot say "the longer the duration, the bigger the drop" — this table refutes that itself. TLT is the longest-duration pure rate instrument in the table, yet its increment was the smallest at −0.15pp. And reverse-computing the yield: TLT's −0.67% on the day corresponds roughly to 10Y +4bp, while 4.893% versus the Treasury's 9/9 fixing of 4.83% is +6.3bp — most of those 6bp had already happened before 08:30 (part of it should be attributed to the 08:15 ECB). The chain "PPI → rates → kill long duration" gets no clean support from this brief's own data.

• ⚠️ On "semiconductors carry a rate-independent idiosyncratic discount" — that first-draft conclusion does not hold and is withdrawn. The SMH − IGV spread went from 1.23pp to 1.18pp, a change of only 0.05pp, within the noise of thin pre-market liquidity; "basically unchanged" is not itself strong evidence; and more importantly, "it existed before 08:30" does not equal "it is unrelated to rates" — the 08:15 ECB hike and that 5bp of yield increase both happened before then. SMH and IGV also have different betas to begin with, so the same rate shock should produce different magnitudes. ⇒ After revision, only the observable phenomenon remains: semiconductors show roughly 1.2pp of weakness relative to software, cause unconfirmed.

⚠️ This section was overturned by its own author twice during drafting; recorded faithfully: first, before PPI I was about to write "software is fine ⇒ this is not a duration problem," and the 08:32 re-read showed IGV falling from −0.28% to −0.84% (the largest increment in the table), which killed that claim on the spot; second, the rewritten "semiconductor-specific discount is unrelated to rates" was flagged in QA as a non sequitur (treating sequence as causation) and withdrawn again. All that survives is description of the phenomenon, with no causal assertion.

1.4 Overnight Asia/Europe (close/latest, CNBC, read 08:12–08:13 ET)

Key reading: Nikkei +0.20%, KOSPI −0.25%. Korea is the global home of memory (Samsung / SK Hynix) and Japan is the home of semiconductor equipment (Tokyo Electron). Both markets closed essentially flat today while U.S. semis are down 2% pre-market — this contrast directly rules out the explanation that "an overnight crash in Asian semis was transmitted here."

---

2. Strongest Themes, Descending

⭐ Themes 1 and 2 are not two things; they are two ends of the same thing. Oil up → diesel PPI +24.1% → headline PPI +5.4% y/y → rate-hike odds up → 10Y 4.893% → long-duration assets killed. So "long energy + short long duration" is not two independent trades today, it is two legs of the same trade, with highly correlated exposure. If there is a ceasefire in the Middle East, both legs fail at once. This must be recognized at the position-sizing level.

---

3. Overall Single-Stock Strength Table (split by direction, descending within each group)

⚠️ Read this first: the pre-market liquidity threshold must be a notional amount, not a share count. In the table below, "Pre-Mkt Notional" = pre-market price × pre-market volume. This brief uses $5 million of notional as the threshold for "is this quote credible" — quotes below it (e.g. defense name NOC at 76 shares = $39,000, GD at 45 shares) are never used as evidence in this brief; see §6.2.

3.1 Bullish Direction

3.2 Bearish Direction

---

4. Single-Stock Scoring Model (100 points total)

⚠️ In the first draft three totals did not match the sum of their components (XOM misrecorded as 74, OXY as 71, AAPL missing its risk deduction); corrected. Total = sum of the seven components + risk deduction. After correction the ranking changes from XOM > OXY > AAPL to XOM (64) > AAPL (60) > OXY (57), and the ordering in §9① has been updated to match.

---

5. Top Single-Stock Deep Dives

5.1 XOM / OXY — the two of six large-cap energy names that have still not made new highs

Related news: the U.S. military destroyed 5 Iranian tankers on 9/8; the Houthis struck Saudi Arabia's Jazan refinery (400,000 bpd). WTI $99.31 +3.39%, Brent $104.45 +3.20% (CNBC, 08:24 ET). CNBC⁴

Catalyst logic: a pure supply-side shock that directly lifts upstream realizations and operating cash flow, without requiring any improvement in demand.

Theme-stage judgment: day 2–3 of the move, not day 1. The tankers were destroyed on 9/8 (Tuesday), WTI was already up 3.99% on 9/9, and it added 3.39% today. Any description of this as an "overnight surprise" is wrong.

⭐ The key divergence inside the sector (computed locally, read 08:34–08:35 ET):

⚠️ Convention warning: "new high" in the table above is judged entirely on pre-market prices, yet pre-market trades do not count toward the official 52-week high/low calculation, and the breakouts are only +0.81% to +1.37%, entirely within pre-market noise. They may not hold at all after the open. So "the premise has been consumed" as a downgrade reason is conditional — if COP/CVX fall back below their 52-week highs after the open, their watch status should be restored.

Judgment: yesterday this list recommended COP/OXY/XOM on the grounds that "they are the group that has risen least within energy." Today COP and CVX have already reached 52-week highs — for them that premise has been consumed. Only XOM (−5.34%) and OXY (−7.77%) still qualify.

⚠️ But "has risen least" must be kept separate from "is anyone buying" — rising least can also mean nobody wants it. XOM's pre-market notional of $21.5M is the highest among the 13 energy names covered here (total $72.4M); OXY's $7.2M is middling. ⚠️ But that metric is not normalized by market cap or 20-day average volume, so large caps rank high mechanically, and this brief does not use it as a ranking basis (see §9①). And while FANG ($0.6M) and BKR ($0.4M) are also up, their quotes are not credible and this brief does not use them.

Fundamental verification: ⚠️ yfinance was rate-limited by Yahoo throughout this run, so financial data for the energy group was not obtained and no revenue/EPS/valuation percentiles are given. The judgments above rest entirely on price, volume and 52-week position, with no fundamental verification — that is the honest cost of this section; please weight accordingly.

Final judgment: XOM watch closely / OXY watch closely / COP, CVX, MPC, VLO downgraded to watch only.

---

5.2 ⭐ AVAV — "+4.33% pre-market" is a number that will fool you

On the surface: FY27Q1 was reported after the close on 9/9, revenue $480.49M and adjusted EPS $0.59, both beats; pre-market +4.33% on $47.1M of notional (real).

⚠️ But switch the baseline to the moment just before the release and the conclusion flips entirely:

• 9/8 close = $148.78
• 9/9 regular-session close $140.80 (−5.36%, which happened before the release)
• 9/10 pre-market $146.90
• ⇒ Relative to the 9/8 close, i.e. before the release, the net change is −1.26%.

That +4.33% is essentially just giving back 9/9's intraday decline; the two-day net change is close to zero or slightly negative. And 9/9's intraday decline was sector-wide (KTOS −3.03% and RCAT −6.50% the same day, versus SPY at only −0.46%), not specific to AVAV.

Four hard problems in the report itself (all from primary SEC filings):

1. Guidance unchanged word for word. The 9/9 press release says "continues to expect," not "raises." FY27 revenue $2.125–2.225B, adjusted EBITDA $305–325M, adjusted EPS $3.02–3.34 — identical to what was first given on 6/29. More telling: the company won the $464.8 million Army E-HEL laser contract on 9/2 and still raised no guidance seven days later.
2. Adjusted EBITDA −5.6% y/y ($53.4M vs $56.6M) while revenue was +5.7% — negative operating leverage. The EPS improvement comes from the natural run-off of acquisition amortization ($79.7M→$43.4M), interest swinging from −$17.4M to +$4.1M, and R&D being cut 27.6% ($33.1M→$24.0M). This is not an improvement in operating efficiency.
3. Free cash flow −$36.0M; the +$13.5M of operating cash flow includes +$128.3M of receivables collection, and excluding that, operating cash flow is −$114.9M; inventory over the same period was +31.3%.
4. Material weaknesses in internal control remain unremediated. The 10-Q filed this morning (Item 4) states that management concluded disclosure controls and procedures were "not effective as of 2026-08-01", with two material weaknesses (BlueHalo's IT general controls and the goodwill-impairment review control) still not remediated; the company just went through an Item 4.02 non-reliance announcement plus a 10-Q/A restatement in June over a miscalculated goodwill impairment.

Valuation: at the pre-market price, FY27E adjusted P/E is about 46x and EV/EBITDA about 24x — against guidance whose midpoint implies only +10% revenue growth and which requires EBITDA over the remaining three quarters (Q2–Q4) to jump from $53.4M per quarter to $87.2M per quarter (+63%) to be met.

Final judgment: watch only. The 42% year-to-date decline has compressed the thematic premium, not pushed the valuation into cheap territory. Grade lowered from A to C.

---

5.3 AEO — a $179 million tariff refund is hiding inside "gross margin +980bp"

Facts (company press release): FY26Q2 revenue about $1.38B (+8%), comps +6%, Aerie revenue +25% / comps +19%, gross margin 48.7%, +980bp y/y; FY26 operating income guidance $540–550M, "including the net tariff-refund benefit"; Q3 guidance calls for mid-to-high-single-digit comps, flat gross margin y/y, and operating income of $110–115M.

⭐ Computed locally (the arithmetic is shown; please verify yourself):

• Net tariff-refund benefit $179 million ÷ revenue $1.38 billion ≈ 13.0 percentage points of gross margin
• Reported gross margin 48.7%, prior-year period = 48.7 − 9.8 = 38.9%
• Excluding the refund, this quarter's gross margin ≈ 48.7 − 13.0 = 35.7%
• ⇒ Year over year that is not +980bp but roughly −320bp

The same restatement on the guidance side: at Q1, FY26 operating income guidance was $390–410M (midpoint $400M); it is now $540–550M (midpoint $545M), and the $145M increase is smaller than the $179M refund. ⇒ Excluding the refund, the implied figure is about $366M, below the bottom of the original guidance range.

⚠️ A convention uncertainty that must be flagged: the $179 million is the net benefit within this quarter's gross profit, and whether the "net tariff-refund benefit" cited in the FY guidance is the same amount or includes other quarters was not broken out by the company and was not obtained locally. So the second restatement above is directionally reliable but its exact value is to be verified. The first (the gross-margin restatement) uses only this quarter's data and is unaffected.

Final judgment: avoid. Pre-market −14.80% to $14.39, approaching but not yet below the 52-week low of $14.06 (the first draft wrongly wrote "already below"; corrected). Real operating gross margin is falling, and the "record gross margin" headline was manufactured by a one-time refund.

---

5.4 COO — why a −1.45% guidance cut corresponds to a −17.66% share price

Earnings facts (primary SEC 8-K): Q3 revenue $1,066.2M (+1%), about 3% below consensus; non-GAAP EPS $1.15, a 2.7%–3.6% beat — but the call disclosed that roughly $0.03 of it came from a tariff refund (secondary source, to be verified); excluding that, $1.12 ≈ consensus and the beat essentially disappears.

Size of the guidance cut (comparing the two 8-Ks; magnitudes computed locally):

⭐ The most important thing in this table is not the magnitude but the attribution: 96.8% of the revenue cut comes from CooperVision alone, while CooperSurgical barely moved.

⚠️ First, set the price baseline straight (using the same yardstick applied to AVAV in §5.2; the first draft applied it only to AVAV, and it is now completed here):

⚠️ The key detail for COO: the −6.22% in the 9/9 regular session happened before the release, on volume 4.0x the 20-day average, while peers ALC were −1.00%, BLCO +0.34% and XLV −0.33% — this was an idiosyncratic decline, and no public intraday catalyst was found locally. It may have been front-running/a leak, or it may be a detection gap. So the "magnitude mismatch" argument — "a −1.45% guidance cut corresponding to a −23% two-day decline" — must be discounted: the −6.22% portion has nothing to do with the earnings content that was published. On a net-earnings-effect basis the mismatch is −1.45% against −22.8%, still significant, but the chain of reasoning must be stated clearly.

So why is it down 17.66%? Three reasons; the last two are the real ones:

1. Quarterly EPS growth flipped from positive to negative. The newly added Q4 guidance of non-GAAP EPS $1.05–1.09 (midpoint $1.07) is −6.1% y/y against FY25Q4's $1.14, whereas the first three quarters were +16.1%. This number is not in the headline.
2. ⭐ The strategic review ended with "no sale." The same-day EX-99.2 announced that the board unanimously decided to retain CooperSurgical, with the wording that "the offers received were not in the best interests of shareholders." The spin-off/sale optionality that had supported the stock since December 2025 was removed in one stroke after the close on 9/9 — and it was falsified by real external bids, not by sentiment. The 52-week high of $89.83 happens to sit near the window when the strategic review was launched on 2025-12-05.
3. The fourth guidance cut in two years (secondary source, to be verified) — credibility discounts are cumulative.

Valuation: the pre-market $52.27 corresponds to a forward P/E of about 11.5x (using the freshly cut FY26 guidance), below all 21 quarter-end readings of the past 5 years (range 13.43–29.67); P/B of 1.22x is likewise outside the historical range of 1.47–2.96. The buyback authorization has been expanded from $2.0 billion to $3.0 billion, with about $1.5 billion remaining ≈ 14.8% of the pre-market market cap; Q3 FCF was $273M (+66%).

⚠️ But that "extremely low percentile" needs two discounts: (a) the denominator is guidance that was just cut, management explicitly declined to give FY27, there is a known headwind of about 2pp from the tax rate, and consensus FY27 EPS of $5.00 will most likely be revised down further; (b) the P/B denominator contains $3.876 billion of goodwill, and CSI was just repudiated by market pricing — a low P/B at a moment when asset realizability is in question is a weak signal, not a strong one.

⭐ One crack management did not explain: the company attributes CVI's weakness entirely to "proactive destocking in the U.S. channel" and says U.S. end consumption is stable in the mid-single digits all year. But Asia-Pacific organic was −5% and accounts for 17.6% of CVI revenue, which that explanation does not cover at all.

Final judgment: watch only (not buy). The magnitude mismatch (a −1.45% guidance cut against a −23% two-day decline) does make it a candidate for mean reversion, but the disappearance of the M&A optionality is permanent, not sentiment.

---

5.5 MU — the highest pre-market notional in the field, and a premise almost everyone gets wrong

Pre-market notional $726.0M, the highest in the field; MU+SNDK+WDC+STX together are $1,278M — memory is where the money actually is this morning, even though equipment names top the decliners list.

⚠️ First, correct a popular premise: several reports claim "memory stocks have fallen more than 20% from their highs and entered a technical bear market." That holds for peers but not for MU right now (a sub-agent computed this from Nasdaq historical prices, through the 9/9 close): MU is −18.1% from its 52-week intraday high and −15.3% from its peak close, the shallowest drawdown in the group; WDC is −39.7%, SNDK −25.1%, STX −22.6%. MU's current price is above both the 50-day ($933) and 200-day ($614) moving averages.

Fundamentals (primary SEC 8-K/10-Q): FQ3-26 revenue $41.456 billion (+345.7% y/y), GAAP gross margin 84.6%, GAAP EPS $24.67; FQ4 guidance of revenue $50.0 billion ± $1.0 billion, gross margin about 86%, GAAP EPS $30.73 ± 1.00.

⭐ Three calibrations that are easy to miss:

1. FQ4 has 14 weeks; FQ3 has 13. The apparent sequential gain is +20.6%, but normalized per week it is only +12.0% — the extra week contributes about 8.6pp. Sequential momentum falls from +73.8% in FQ3 to +12.0%, an order-of-magnitude deceleration.
2. This round of profit is 100% from pricing, not shipments. From the 10-Q: within DRAM's +67% sequential gain, ASP contributed low-60% and bit shipments grew only low-single-digit; within NAND's +99%, ASP contributed mid-80% and bits only mid-single-digit. With an 84.6% gross margin and a nearly rigid COGS structure, every 1% drop in ASP cuts gross profit by about 1.18% and operating profit by about 1.24% (the first draft wrongly wrote 6.5%, which is actually 1/(1−84.6%) — the wrong denominator; corrected). The real fragility is not the point-elasticity multiple but the fact that all of the incremental profit is bet on ASP, a single variable.
3. The valuation shows the classic cycle-top pattern: forward P/E 6.6–7.2x (about the 12.5th percentile over 5 years), while P/B is 11.5x (98.6th percentile over 10 years) and P/S 12.9x (97.9th). The 86-percentage-point gap between them is the textbook shape of "the market paying an extremely high P/B for an E it believes is unsustainable."

   ⚠️ But a counter-note is required, or the warning will be mistaken for a conclusion: MU adds about $28 billion to net assets every quarter, and BVPS went from $45.11 to $87.97 in a year. P/B percentiles are systematically inflated during periods of rapid earnings accumulation and cannot be placed side by side with the 2018/2022 readings. In addition, this cycle has take-or-pay long-term contracts (the company says that even at the price floor, gross margin would be far above the peak of any prior cycle) providing partial support. This is a red flag, not a verdict.

Next earnings: 9/30 after the close (FQ4 + full-year FY26); that quarter ended on 9/3 and the company is currently in its quiet period.

Final judgment: watch only.

---

5.6 ORCL — tonight's event, and a positioning fact almost everyone gets backwards

The date is primary-verified: the Oracle IR press release explicitly says results will be "released on Thursday, September 10th, after the close of the market," with the call at 16:00 CT (= 17:00 ET). The most recent 8-K on EDGAR is still the one from 6/10, so as of 08:36 ET the results have not been released.

⚠️ The narrative "it's already halved from the 52-week high, so it's cheap" needs one counter-fact: ORCL is indeed −53.9% from its 52-week high of $345.72, but it is running hot into the print, not lying at the lows waiting for it — over the six trading days from 9/1 to 9/8 it gained +14.4%, it is +40.6% off the 7/24 low of $114.99, and RSI is 62.4. Both facts have to be presented together.

⭐ Consensus is just an echo of guidance: consensus revenue is $19.13B / non-GAAP EPS $1.74, while the company's own Q1 guidance from 6/10 translates to revenue of $18.96–19.25B and EPS of $1.72–1.76. Consensus lands precisely in the middle of the guidance range. ⇒ "Landing inside the guidance range" is neutral or even slightly negative; only exceeding the top of guidance constitutes a positive surprise.

The arithmetic most worth doing tonight: FY27 full-year guidance is $90 billion. FY26's Q2+Q3+Q4 totaled $52.432 billion; if Q1 lands at the guidance midpoint of $19.10 billion, then Q2–Q4 must deliver $70.90 billion, +35.2% y/y — yet Q1 is only guided to +27%–29% and FY26Q4 actually came in at +21%. The $90 billion target requires growth to re-accelerate to 35% for three consecutive quarters after Q1, and the burden of proof falls on the Q2 guidance given tonight.

Four verifiable cracks (all primary):

• Free cash flow: FY26 operating cash flow $31.977 billion − capex $55.663 billion = −$23.686 billion; consensus FY27 FCF is −$47.77 billion.
• Leverage: interest-bearing debt $129.541 billion (+40%), plus non-current operating lease liabilities of $26.648 billion (+131%); Q4 interest expense alone was $1.438 billion (+47% y/y), equal to 23.4% of that quarter's operating income. S&P cut the rating to BBB− on 7/9 (secondary source, to be verified). FY27 plans about $40 billion of refinancing, including the announced $20 billion ATM offering (i.e. dilution); FY26 buybacks were only $95 million and share count rose 1.68% y/y.
• Gross margin: consolidated gross margin went from 70.51% in FY25 to 65.82% in FY26, and 65.22% in Q4 alone; "cloud and software" cost growth (FY26 +52%) continues to outpace cloud revenue growth (+39%). This is the OCI-dilution thesis projected onto the GAAP statements, and it requires no reference to any non-public document.
• RPO of $63.8 billion × 10 = $638 billion (+363%) is accurate, but $75 billion of it is "customer-prepaid GPU funds + customer-supplied GPUs", whose revenue quality is nothing like software subscriptions. ⚠️ The 10-K only discloses that "no single customer accounts for ≥10% on a revenue basis"; it has never disclosed RPO customer concentration; the two third-party estimates circulating in the market — "OpenAI is half the backlog" and "about 66%" — differ by 16pp, and no primary evidence was obtained locally, so both are marked unverified.

Final judgment: watch only. A directional position ahead of an event is gambling, not research. Options imply a one-day move of about ±11% (secondary source).

---

6. Negative/Avoid List

6.1 Avoids with primary evidence

6.2 ⚠️ Avoided for insufficient evidence (which is not the same as bearish)

1. All defense-name pre-market quotes this morning. ITA +0.50% looks like a beneficiary of the Iran situation, but the constituent pre-market notionals are: NOC 76 shares ($39,000), GD 45 shares ($16,000), HII 45 shares, LHX $340,000, RTX $270,000, LMT $1.83 million — all below the $5 million threshold. These "gains" are not market consensus, they are a handful of scattered orders. This brief does not list defense as a beneficiary theme. The only name with real pre-market notional, AVAV ($47.1M), was separately rejected in §5.2.
2. Tanker stocks. FRO +0.21%, TNK +0.20%, ASC +0.03%, STNG +1.13% (1,461 shares = $120,000). Oil is +3.4% while tankers barely move, and several are already near 52-week highs. This is the same pattern as the 9/9 brief's "oil cleared verification point 1 with +3.71% of headroom while tanker stocks did not rise at all," now recurring for a second consecutive day.
3. BHVN (−11.20%, $7.9M) and RXT (+8.64%, $14.7M): the notionals are real, but no same-day primary catalyst was confirmed locally (the RBC downgrade report that turned up corresponds to a previous close that does not match today's, so it is judged to be prior-period news). Attribution unknown; no directional participation.
4. CULP (+18.34%), SHOE (−21.50%), FLWS (−15.85%), LOVE (−12.79%): micro-cap/low-priced names with pre-market notionals of $0.2M–$5.0M, liquidity insufficient to support institutional execution.
5. Refiners (MPC/PSX/VLO): see §7.2 — the crack spread barely moved this morning; what they are rallying on is crude beta, not refining margin.

---

7. Within-Theme Rankings

7.1 Theme 1: Crude (direction confirmed, but clear internal divergence)

7.2 ⭐ A quantitative test that downgrades the refiners (computed locally)

The Jazan refinery in Saudi Arabia (400,000 bpd) was attacked, which intuitively should help U.S. refining margins. But the commodity side does not support that transmission:

The crack spread is −$0.10 (−0.16%) — essentially unchanged.

⚠️ Two qualifications that must be stated together: (a) the comparison window covers only 1 day, while this brief itself judges crude to be on pricing day 2–3 — whether cracks widened first on 9/8–9/9 was not checked locally, so the evidence window for "refining margins did not expand" is insufficient; (b) a 3-2-1 level of $61/bbl is historically very high, and "unchanged on the day" does not mean "refining profitability is poor." This item therefore supports only "no incremental refining-margin expansion this morning," not "refining fundamentals are bad." ⇒ Products and crude rose in lockstep and refining margins did not expand. The 52-week highs MPC/VLO/PSX printed this morning are buying crude beta, not refining margin. Oil is a refiner's cost, not its revenue.

7.3 Theme 3: Semiconductors (bearish across the board, ranked internally by "decline vs attribution")

---

8. Open-Bell Verification Signals

8.1 Pre-market (as of 08:37 ET)

• Direction and volume in the leaders of the strong theme: XOM +1.68%/$21.5M, OXY +1.95%/$7.2M — the gaps are moderate (<2%) and do not qualify as "large gaps", leaving room for both gap-and-go and gap-fill.
• ⚠️ Already gapped large, do not chase: AVAV (+4.33%), CULP (+18.34%), RXT (+8.64%); on the downside, COO (−17.66%) and SHOE (−21.50%) should likewise not be chased short.
• XLE/WTI elasticity ratio (computed locally, same-moment readings): 9/9 close 0.21 → 08:19 ET 0.36 → 08:33 ET 0.42. It is improving, but it is still far below 1 — the equity market is still treating this oil rally as a geopolitical premium that will mean-revert, not a durable repricing.

8.2 Intraday (after 09:30)

• First 30 minutes: watch above all whether XOM can hold its gain provided WTI does not pull back. Against a broad-market backdrop of −0.36% (SPY) pre-market, energy has to earn it with its own alpha.
• Sector ETF linkage: the XLE-to-USO ratio; the SMH-minus-IGV spread (if the 1.2pp SMH − IGV gap converges after the open, that "idiosyncratic discount" was just an artifact of thin pre-market liquidity).
• Volume confirmation: MU's $726M pre-market notional is already the highest in the field; if it falls on heavy volume after the open, the memory pullback enters a second phase; if it recovers on heavy volume, the pre-market move was panic.

8.3 Options and events

• ORCL implied one-day move about ±11%–11.5% (secondary source). Post-earnings IV crush risk is very high: even if you get the direction right, buying options can still lose money as implied volatility collapses.
• ADBE reports the same evening. Two heavyweight names reporting the same night will substantially amplify after-hours Nasdaq futures volatility.

8.4 ⚠️ Two scheduled hard events during today's session (omitted from the first draft, added here)

⇒ Each of the two main themes has an intraday adjudication point, and both are in the afternoon. The morning's direction does not determine the closing direction.

8.5 ⚠️ Scenarios that would invalidate this brief's conclusions (ordered by probability)

1. A ceasefire/diplomatic breakthrough in the Middle East → Themes 1 and 2 fail simultaneously (see the correlation note in §2), and both legs — XOM/OXY and "avoid long duration" — reverse together.
2. Tomorrow's CPI comes in clearly below expectations → the rate-hike pricing is given back, 10Y falls, today's beaten-down IGV/IWM/SLV bounce hardest, and today's energy longs face an unwind of the inflation trade.
3. A primary catalyst for the semiconductor discount emerges after the open → if bad news is confirmed, "avoid" in §7.3 is not enough; if it was only thin pre-market order flow, SMH recovers quickly and the 1.2pp weakness in §1.3 is falsified.
4. WTI touches the round $100 level and pulls back, or 12:00 EIA inventories show a large build → energy names open high and fade.
5. Strong demand at the 13:00 30-year auction → yields fall back and today's "avoid long duration" call fails within the session.
6. Institutional bottom-fishing appears in COO → the "watch only" in §5.4 was too conservative.

---

9. Final Conclusions

① The 5 stocks most worth watching today

Design principles and flaws that must be disclosed: 0. ⚠️ The ordering has been adjusted to the corrected §4 totals: XOM (64) > AAPL (60) > OXY (57). The first draft placed OXY second because of the summation error in the scoring table.

1. The five names cover four different driver variables — oil price (XOM/OXY), product cycle and sector divergence (AAPL), post-earnings oversold mean reversion (COO), and event (ORCL). This is deliberate: one of yesterday's lessons was that "all six verification points were copies of the same variable."
2. ⚠️ A concentration flaw remains, disclosed faithfully: XOM and OXY are two expressions of the same oil trade and would fail together on a ceasefire. And as §2 notes, oil and rates are the same causal chain today, so "long energy" and "avoid long duration" do not hedge each other.
3. Every verification point is set on "the thing most likely to make this trade lose money," and the three for XOM/OXY/AAPL are all set on relative performance — because the risk is not "does oil/the market go up" but "if it does, does the name follow."
4. ⚠️ The biggest flaw in this section: no fundamental data was obtained for the energy group this run (yfinance was rate-limited throughout), so the XOM/OXY recommendations rest entirely on price, volume and 52-week position, with no valuation or cash-flow verification. Compared with yesterday's brief (which had P/B and EV/EBITDA percentile support) this is a downgrade in evidence grade; please reduce the weight accordingly.
5. Five categories of "good-looking" names were deliberately excluded: COP/CVX/MPC/VLO at new 52-week highs (premise consumed), all defense and oil-service names with pre-market notional below $5 million, tanker stocks unconfirmed for a second consecutive day, RXT/BHVN with unknown attribution, and AVAV, which beat but did not raise guidance.

② The 3 strongest themes today

③ What to avoid today

1. Semi equipment (LRCX/TER/KLAC/AMAT) — the biggest declines with the murkiest attribution; and the popular "Apple buying Chinese memory" attribution does not fit on transmission direction (it should help equipment).
2. AI power/nuclear (OKLO/SMR/VRT/GEV) — OKLO fell 2.37% on the day it received an Overweight initiation; good news was rejected by the market, which is a stronger signal than the decline itself.
3. Refiners (MPC/PSX/VLO) — the crack spread was −0.16% this morning, essentially unchanged; the new highs are buying crude beta, not refining margin.
4. All pre-market signals from defense names — NOC at 76 shares, GD at 45 shares, all with pre-market notional < $5 million, which does not constitute evidence (the only one with real volume, AVAV, was rejected separately).
5. Tanker stocks — oil is +3.4% while tankers barely move, unconfirmed for a second consecutive day.
6. Large moves with unknown attribution (RXT +8.64%, BHVN −11.20%) — the notionals are real but no primary catalyst was found locally, so no directional participation.
7. Directional positions in ORCL / ADBE — double earnings tonight, high IV crush risk.
8. Automatically buying "beats" as such — within today's window, AVAV (beat but no guidance raise), NAVN (beat and raise yet −16.5%), COO (EPS beat yet −17.7%) and AEO (record gross margin yet −14.8%) all four fell. ⚠️ No full-sample statistics were run locally, so this is not a general rule, but four cases pointing the same way is worth noting.

④ Final one-line judgment

Today's loudest headline is "Brent $104, WTI approaching $100," but the real event this morning is that two institutions stated the same causal chain twice within 15 minutes: at 08:15 the ECB hiked 25bp, with the reason written down in black and white as "the conflict in the Middle East continues to generate inflation pressures"; at 08:30 U.S. August PPI came in at +5.4% y/y, with more than three-quarters of the goods-component gain coming from energy and diesel alone +24.1%. The supply shock in the Strait of Hormuz is no longer just an energy-sector story — it has walked into a central bank's policy statement and into official inflation statistics, and that is precisely why "long energy" and "avoid long duration" cannot hedge each other today: they are two legs of the same trade. As for that extra 1.2 percentage points of semiconductor weakness, this brief tried three attributions and overturned two of them; all that survives is the phenomenon itself, cause unconfirmed. This afternoon's 12:00 EIA inventories and 13:00 30-year auction will each deliver an intraday verdict on one of those two legs.

---

⚠️ Risk disclosure: this list is a pre-market information review and watch list only and does not constitute investment advice. U.S. equities carry high volatility and pre-market gap risk, and post-earnings IV crush and guidance reversals occur; automatically generated content may contain stale information or factual errors. Rely on company disclosures/SEC filings, and do not use this directly as a basis for trading.

Ops Notes (not sent to clients)

Data-fetch and tooling status this run:

1. yfinance unavailable throughout: all three sub-agents reported YFRateLimitError / HTTP 429, and the main flow did not use it. The consequence has been carried into the body: the energy group (XOM/OXY etc.) has no fundamental data at all, and §9① discloses this faithfully as "a downgrade in evidence grade." Recommendation: the energy group's valuation percentiles should be restored in the next brief.
2. ⚠️ The biggest trap this run: the Nasdaq quote API's secondaryData cannot be used as the previous close.
   • The initial mapping treated secondaryData.lastSalePrice as the prior close and produced a batch of completely wrong-direction readings such as MU +2.75%, AVAV −5.36%, AEO −1.92% (AEO was actually −14.5%).
   • Root cause: the field means different things for different instruments. TLT's secondaryData really is the 9/9 close (baseline 82.20); but USO / XLK have secondaryData sharing the same baseline as primaryData (for USO both reverse-compute to a prior close of 149.97) — it is a "combined last trade," not the prior close.
   • The fix (now hard-coded into work_uspre0910/nq.py): always reverse-compute the previous close as primaryData.lastSalePrice − primaryData.netChange, cross-check it against percentageChange, and flag <<对不上 on any mismatch. Zero alerts across 40+ instruments this run.
   • How it was found: the AEO news said −11.2% while the API computed −1.92% → located by reverse computation. The "price-triple-must-be-reverse-computed" memory hit directly; recommend adding the secondaryData trap to that same entry.
3. PPI was fetched with a setsid background script polling until 08:31 before pulling BLS, which avoids the old trap of "the Bash tool kills background processes." But note: a foreground while poll will consume the Bash tool's 10-minute limit and be killed (this happened once this run); always use setsid plus an output file.
4. CNBC's 2Y change_pct broke again today (last 4.487% > its own previous close of 4.427% yet showing −0.1133%), the second consecutive trading day it has hit. Everything was switched to computing from the Treasury CSV fixing. Recommendation: upgrade "never use CNBC Treasury change_pct" from memory to a hard filter at the script level.
5. ⚠️ An attribution that was nearly written wrong: several outlets attributed the memory selloff to "Apple cleared to buy CXMT/YMTC," with the wording "fell sharply on Monday." But 9/7 was the Labor Day holiday (the Treasury CSV has no record between 9/4 and 9/8, which cross-verifies it), so the date anchor of that attribution does not hold; and the transmission direction is backwards (helping Chinese capacity expansion ⇒ should help equipment, yet equipment fell the most this morning). It was excluded in §1.2 on three independent grounds and the conclusion is written as "attribution unknown." This is a combined hit of the "recurring-column-headlines-hide-date" and "right-entity-wrong-channel-and-clock" memories.
6. ⚠️ Self-overturned once: before PPI came out I was about to write "software is fine ⇒ this is not a duration problem"; the 08:32 re-read showed IGV falling from −0.28% to −0.84% (the largest increment in the field), and my own second reading overturned the claim. It was rewritten as a two-layer breakdown (§1.3), with the reversal explicitly noted in the body. "Always recompute once before filing" proved its worth again.
7. WebFetch failure list: Benzinga 403, Schwab 403 (returned an authorization error page), CNBC article body 403, fastcompany 403, alphaspread 403, Yahoo 429. CNBC/Benzinga article pages have now 403'd across several briefs in a row; recommend abandoning them outright and switching to search summaries plus primary SEC/IR sources.
8. ⚠️ The wrong quarter was pulled for AEO's 8-K: the EDGAR link aeo-ex99_1.htm returns FY26 Q1 (published 2026-05-28), not this Q2. The document itself is complete and self-consistent with no tell; the only way to catch it is the quarter-end date of 5/2. It was dropped and the Q2 numbers were taken from the company press release instead. This is another form of "stale-source-silent-lag": EDGAR links will also hand you the previous period.
9. All three sub-agents succeeded and were high quality, and each corrected a substantive error in the main flow:
   • AVAV: corrected "+4.33% = the market rewarding the print" to "−1.26% relative to the 9/8 close," and dug out guidance unchanged word for word + material weaknesses in internal control unremediated — which took AVAV from the top candidate down to "watch only."
   • ORCL: corrected the "halved and lying at the lows" positioning narrative (it is actually +14.4% over six days running into the print) and supplied the self-consistency arithmetic for the $90 billion full-year target.
   • MU: corrected "MU is still in a technical bear market," a media premise I was about to adopt directly (it actually has the shallowest drawdown in the group), and supplied the gap between a 12.5th-percentile forward PE and a 98.6th-percentile PB.
   • Conclusion: the "research sub-agents are worth waiting for" memory is validated again. All three changed conclusions in the body; they did not merely add detail.
10. risk-auditor was run (about 8 minutes) and caught 6 red flags that changed conclusions, all corrected in the body with a trail left:
    • ① The ECB's 08:15 25bp hike was missed — zero mentions in the whole draft. This is the most serious one this run: it is both an S-level primary event and falls exactly inside §1.3's "before PPI" read window, contaminating that breakdown, and the ECB statement attributes the hike to the Middle East conflict, which is the strongest external evidence for this brief's core causal chain, and it was nearly missed entirely. ⇒ Added to §0-2, §1.1 (row 0), the §1.3 warnings and §9④.
    • ② §1.3's "the longer the duration, the bigger the drop" is refuted by TLT in its own table (TLT has the smallest increment at −0.15pp), and most of the 6bp yield rise happened before 08:30 ⇒ conclusion A downgraded.
    • ③ §1.3's "the semiconductor discount is unrelated to rates" is a non sequitur ("happened before" ≠ "no causation"), and the 1.23→1.18pp change of 0.05pp is noise ⇒ conclusion B withdrawn, only the phenomenon retained.
    • ④ Three totals in the scoring table did not match their components (XOM 74→64, OXY 71→57, AAPL missing its risk deduction 64→60) ⇒ the top two in §9① change from XOM/OXY to XOM/AAPL.
    • ⑤ §1.2's exclusion reason "the transmission direction is backwards" is itself wrong — CXMT/YMTC are Entity-List constrained and cannot buy U.S. equipment, so equipment falling harder is exactly consistent with that narrative ⇒ the reason was withdrawn and the conclusion downgraded from "ruled out" to "attribution unconfirmed."
    • ⑥ Two scheduled events today were missed: EIA inventories pushed to 12:00 ET by Labor Day, and the 30-year auction at 13:00 ET ⇒ new §8.4 added.
    • Also corrected: MU operating leverage 6.5%→1.2% (wrongly used 1/(1−GM)), AEO's "already below the 52-week low" was a hard error ($14.39 > $14.06), COO's breach depth 11.8%→11.24%, COO's verification points changed from AND to independent tests (the original branch was unreachable), ORCL's three inconsistent thresholds unified at 35.2%, qualifying domains added to several universal quantifiers, and all 52-week highs annotated as "judged on pre-market prices, needs confirmation after the open."
    • ⚠️ Lesson: the main flow did plenty of self-checking this run (date verification, arithmetic review, second readings, notional thresholds), yet still missed an S-level primary macro event and an entire table's worth of summation errors. "Checking yourself" cannot replace a QA agent. The next brief should start the fundamentals agent before 07:40.