U.S. Market Recap | Thursday, 2026-09-10 (ET)

Coverage window: 2026-09-10 09:30 ET open → 16:00 ET close, plus the 16:00–17:03 ET after-hours session.

Price basis (every figure is tagged with its read moment — do not compare across moments):

• Open/high/low/close and volume for individual stocks/ETFs come from the CNBC quote API (regular-session 16:00 ET close basis), with independent cross-checks on key names via the Nasdaq official quote API. Vol ratio = day's volume ÷ 10-day average volume (CNBC pcttendayvol).
• Treasury yield levels and daily changes all come from the U.S. Treasury official yield-curve CSV (15:30 ET fixing); after-hours live values are taken from CNBC and separately tagged with their read moment. ⚠️ CNBC's 2Y change_pct field contradicted itself for the third consecutive trading day today (last 4.588% is above CNBC's own prior close of 4.427%, yet it shows −0.3008%); this report does not rely on that field anywhere.
• Crude/precious metals/industrial metals: the settlement price (14:30 ET) and the electronic-session last trade (16:50 ET) are two different numbers; this report gives both and tags each occurrence — do not mix them.
• The 30-year Treasury auction results come from the TreasuryDirect official API as primary data (not a news paraphrase).
• Company financials come from SEC 8-K / EX-99.1 / EX-99.2 primary filings, tagged at each point; items marked "consensus" or "secondary source" have not been primary-verified.

⚠️ The read moment of the after-hours quotes is the single most important qualification in this report. ORCL / ADBE / RH — all three held their earnings calls starting at 17:00 ET, while the after-hours quotes here were read at 16:10 and 17:03 ET — they reflect the 8-K numbers themselves and contain almost nothing from the calls. These prices can still move substantially before the open.

---

0. One-Line Recap

Today was risk-off, and specifically a day where the pre-market brief "got oil right and energy equities wrong."

WTI settled at $102.48 (+6.7%) and Brent at $107.63 (+5.9%), both the highest closes since May 19 — the pre-market list's first theme delivered completely on the commodity side. But the equity side inverted: XLE gapped up +1.27% and closed −0.58%, almost at the day's low. Of the 13 energy names covered here, 12 gapped up (BKR was flat at −0.05%), and on an open→close basis 12 of them fell; the only one that did not give it back was DVN, while XOM / OXY / COP / CVX were all at a loss relative to their own pre-market prices. The number the pre-market list computed itself — the XLE/WTI elasticity ratio of 0.42 — was the single most informative line in that report, and today it collapsed to −0.09.

The three major indices fell for the fourth consecutive trading day: S&P 7,591.70 (−0.58%), Dow 52,064.10 (−0.60%), Nasdaq Composite 26,081.72 (−0.65%), Nasdaq 100 −1.08%, Russell 2000 −1.04%. Breadth was far uglier than the indices: NYSE 690 advancing / 2,069 declining, 223 new lows (99 the prior day); Nasdaq 1,263 advancing / 3,657 declining, 308 new lows. VIX 17.84 (+8.38%). The 10-year Treasury at 4.95% (+12bp, a year-to-date high) and the 2-year at 4.56% (+13bp) — a bear-flattening curve; rate-hike repricing was today's real main line.

Pre-market list hit overview: of the three "watch closely" names only AAPL delivered, but it delivered emphatically — AAPL +3.56%, closing at the day's high on a vol ratio of 1.70x, outperforming by 4.62pp against a QQQ at −1.06%; its verification point (set on excess return versus QQQ) passed beyond dispute. XOM (+0.61%) and OXY (+0.23%) were positive in absolute terms, but relative to their own pre-market prices they were −1.05% and −1.69% — executing at the pre-market price and holding to the close was a loss. Of the 8 items on the avoid list, 6 delivered and delivered cleanly (semiconductor equipment −3.87%, AI power −5.08%, refining −1.18%, RXT/BHVN −9.07%), and the one wrongly avoided item was the tankers: FRO +2.52% / TNK +2.01% / STNG +1.75% / ASC +1.26% — the only stock group today that kept a positive elasticity to oil, and the pre-market brief had tagged it as "unconfirmed for a second consecutive day."

Next-session tone: cautious, with directional authority handed entirely to the 08:30 ET August CPI. This is the last inflation print before the 9/16 FOMC, and market-implied odds of a hike have risen to 56%–63% (sources differ; all secondary sources). With the 10Y already at 4.95% and breadth already this poor, CPI is a two-sided event with asymmetric magnitude. After hours, ORCL +6.3%~6.5% (RPO of $664 billion, far above consensus), ADBE −2.15% (beat and raised full-year guidance and still fell), and RH +6.64% are the three set catalysts for the next open.

---

1. Market Overview

1.1 Indices and Volume

The three major index levels have been cross-confirmed as consistent between CNBC and an independent news source.

1.2 ⭐ Market Breadth — This Is Where Today Was Genuinely Ugly

(WSJ Markets Diary, 16:00 ET close basis)

⭐ Three readings that must be spelled out:

1. The index fell 0.6%, while the number of decliners was 3x the number of advancers. This is not a "mild pullback" — it is a broad-based decline in breadth masked by a handful of index heavyweights.
2. The count of new lows more than doubled in a single day (NYSE 99→223, Nasdaq 193→308, Arca 72→248). The expansion in new lows says more than the index drawdown does about this decline having entered its second stage.
3. ⚠️ But TRIN gives the opposite signal, and it has to be said alongside: Nasdaq TRIN 0.63 and Arca 0.56 are both below 1, meaning declining stocks' share of volume is smaller than their share of issue count — i.e. what fell hardest were the small caps; selling pressure in the large caps was relatively tame. That is self-consistent with "the index only fell 0.6%," and it also means today was not panic selling but a broad yet unaggressive reduction in exposure. NYSE TRIN at 1.08 is slightly above 1 and is the only one of the three leaning toward "selling pressure with volume behind it."

1.3 Rates, Volatility and the Dollar

Treasuries (official Treasury fixing, 15:30 ET):

⇒ Bear flattening: 5Y/2Y rose more than the 30Y. That is the shape of rate-hike expectations, not the shape of inflation expectations or a term premium. The after-hours live value (CNBC, 17:00 ET) put the 10Y at 4.965%, still rising after the close.

Other:

1.4 ⭐ The 13:00 ET 30-Year Auction: Demand Was Strong, the Long End Fell Anyway

The pre-market list named this auction as one of the day's two "intraday adjudication points," writing that "if demand is strong, the long-duration assets being killed today could rebound quickly." The result: demand was indeed strong, and long-duration assets did not rebound.

(TreasuryDirect official API primary data, CUSIP 912810UW6, 29-year-11-month reopening)

Share basis: the denominator is total competitive awards across "indirect + direct + primary dealers" ($21.96 billion this time, $24.90 billion last time), excluding non-competitive awards and SOMA add-ons.

⭐ Primary dealers were forced to take down only 2.2% — structurally a very clean auction. Moreover, the 5.308% awarded yield is below the 30Y close of 5.37% that day, showing the market level at the 13:00 mark was lower than at the close — the yield rise happened mostly after the auction.

⇒ Conclusion: today's long-end selloff was not caused by supply that could not be absorbed, but by demand-side rate-hike pricing. The pre-market brief got the direction of this adjudication point backwards (it expected strong demand → rebound), but precisely because it was backwards, it ruled out the "supply shock" explanation and confirmed the main line as rate-hike expectations by elimination. That is more useful than if it had been right.

1.5 Sentiment Call

Risk-off, but of the "repricing" type rather than the "panic" type. Evidence: (a) VIX only reached 17.84, nowhere near panic territory; (b) Nasdaq and Arca TRIN were both well below 1, selling pressure concentrated in small caps; (c) defensives did not strengthen as a group — Consumer Staples XLP +0.05% was the only traditional defensive sector to rise, while Utilities XLU −0.98% and Real Estate XLRE −0.83% actually fell more than the broad market. Utilities and real estate falling together is direct evidence that rates, not a flight to safety, drove flows — on a genuine risk-aversion day XLU should lead.

---

2. Pre-Market List Reconciliation

2.1 §9① The Five "Most Worth Watching Today," Checked One by One

⚠️ The core of this table is the "vs pre-market price" column, not "today's chg%." The pre-market list was written at 08:32–08:37 ET; the executable price available to readers was the pre-market price, not the prior close. Reconciling on "today's chg%" alone systematically overstates the hit rate.

2.2 ③ The 8 "Avoid Today" Items, Checked

Also: the memory group (MU −4.90% / SNDK −4.06% / WDC −4.43% / STX −2.66%, average −4.01%) was listed as "watch only / avoid," and also delivered.

2.3 Hit Rate

This must be given on three bases; a single number would mislead:

On basis (B), the "+0.10% mean" beat SPY (−0.60%) by 0.70pp — but that positive return is 100% attributable to AAPL alone. The median is −1.05%. Strip out AAPL and the watch-closely group is −1.37%, underperforming the market.

2.4 ⭐ One-Sentence Critique: The Most Expensive Error Was Not Misjudging Oil, It Was Translating "Oil Was Right" Directly Into "Energy Equity Positions"

The pre-market list's first theme, "Middle East supply shock → crude," was S-grade correct on the commodity side — WTI settled +6.7%, the highest close since May 19. But it was wrong on the equity side, and the pre-market list had already computed the disconfirming evidence itself and written it into §8.1:

Original text: "XLE/WTI elasticity ratio (self-computed in-house): 9/9 close 0.21 → 08:19 ET 0.36 → 08:33 ET 0.42. Improving, but still far below 1 — the equity market still treats this leg of the oil rally as a mean-reverting geopolitical premium rather than a durable repricing."

That line was the most informative judgment in the whole report, and it was placed inside "opening verification signals" as a footnote, never entering the position decision in §9①. Measured at today's close:

⇒ The pre-market brief read 0.21→0.36→0.42 as "improving" and raised its confidence in energy positioning on that basis, but all three readings fell inside the thin-liquidity pre-market session, in a window under 30 minutes. The full-session reading collapsed straight to negative. Three pre-market snapshots do not constitute a trend.

A harder point: the energy sector today was uniformly gap-up-and-fade, with exactly one exception.

⇒ The gap belonged to last night and this morning; the segment after the open is what carries forward. Of the 13 energy names, 12 were negative from open to close (the sole exception being DVN's +0.04%), meaning almost no incremental buying took over after 09:30.

⚠️ Basis completion (to stop the table above from being read as "the whole group was uniform"): BKR was the only name without a gap up (open −0.05%), and its −6.62% open-to-close decline was a single-stock event (§3.4), not part of the "energy equities won't follow oil" phenomenon; it should be excluded separately in the statistics. Excluding BKR, the average open-to-close for the other 12 is −1.76%.

2.5 ⭐ Two Criteria: One Saved the Portfolio, One Screened Out the Best Name

The pre-market list used two yardsticks to downgrade energy names; today one was right and one was wrong:

(a) "Already at a 52-week high ⇒ premise already consumed ⇒ downgrade to watch only" — right. The names downgraded by this rule, COP (+0.37%), CVX (−0.49%), MPC (−1.76%), VLO (−0.91%), all underperformed the retained XOM (+0.61%) today. Three of the four closed negative. This criterion is clean and reusable.

(b) "Pre-market notional < $5 million ⇒ quote untrustworthy ⇒ downgrade to watch only" — screened out today's two best names.

⇒ The $5 million threshold answers the question "can this pre-market quote be used as evidence," and it answered it entirely correctly — DVN's pre-market +1.74% really was backed by only $1.9M. But the pre-market list used it as a ranking criterion, and on the question of "how will this name trade after the open" it carries zero information. DVN's vol ratio that day was 1.69x (higher than XOM's 1.18x); there was real money in it intraday, and thin pre-market volume only tells you its holders don't trade pre-market.

This forms an ironic contrast with what the pre-market list itself did with XOM in §9①-2: it had already recognized that "highest pre-market notional in the group" was unusable for ranking because it was not normalized by market cap, and actively downgraded that reason to "pending verification" — yet the opposite side of the same indicator (low pre-market notional) was still used as grounds for downgrading, without a symmetric demotion. Both directions of a criterion should be held to the same standard.

---

3. Theme Verification

3.1 ⭐ Did the Pre-Market Brief Identify the Strongest Theme? — It Got the Direction Right and the Ranking Wrong

The pre-market brief ranked "crude" as theme 1 and "rate-hike repricing" as theme 2. Today's actual strength ordering should be the reverse:

• Rate-hike repricing is the only chain that was self-consistent all the way from commodities to bonds to equities to the internals of sectors: 10Y +12bp → TLT −1.16% → small caps −1.04% → unprofitable growth (OKLO −6.32%, SMR −5.55%) → mortgage REITs (RWT −14.79%) → utilities and real estate falling together. Not one link was a counterexample.
• The crude chain broke at the "commodity → equity" step.

The pre-market brief had in fact already written out the correct causal ordering (§2, original text: "oil up → diesel PPI +24.1% → headline PPI +5.4% y/y → rate-hike odds rise → 10Y 4.893% → long-duration assets killed") — it treated oil as the chain's starting point, but took "going long the starting point" as the optimal expression. Today's measurement: the end of the chain (rates) traded cleanly and decisively, while the starting point (energy equities) found no bid. The market was trading this chain's transmission result, not its cause.

3.2 Semiconductors: Day Two, and the Discount Widened Instead of Converging

The verification set in pre-market §8.2 was: "if the 1.2pp SMH − IGV gap converges after the open, then that layer of 'specific discount' was merely an artifact of thin pre-market liquidity."

⇒ The gap did not converge; it widened to 1.82pp. The "artifact of thin pre-market liquidity" explanation is ruled out.

⚠️ But the IGV reading must be discounted, or this evidence gets overused: IGV's volume today was only 0.59x its 10-day average — less than 60% of normal. The "resilience" of an ETF almost nobody traded is weak evidence; it may simply reflect no sellers rather than the presence of buyers. SMH's vol ratio was 1.05x (normal), so that side of the reading is solid.

⭐ One piece of evidence added today that is independent of liquidity: if the semiconductor decline were purely rate duration, then software — equally long-duration — should have fallen as much or more, while pure rate proxies like utilities/real estate fell 0.98% and 0.83% respectively today — semiconductors fell 2.44%, 2.5~3x as much. Restating it with the lesson the pre-market list itself learned in §1.3: "falling more than the rate proxies" can only show that a non-rate component exists; it cannot show what that component is. Attribution remains unconfirmed — and this is already the second trading day.

⚠️ But an internal contradiction that the pre-market list itself missed showed up today and must be named: QCOM +0.27%, outperforming by 2.71pp inside a sector where SMH was −2.44%, and it was ranked #23 on the pre-market avoid list with the reason given as "sector." And line 11 of the pre-market list's own §1.1 says "QCOM upgraded to Strong Buy (CFRA)." QCOM's intraday range today was $171.00–$182.40 (6.5% amplitude), driven by the CFO's comments at a conference on the Amazon AI chip partnership (roughly $60 billion in scale, revenue contribution starting in the December quarter, mapping to the FY27 data center target of $5 billion; secondary source). The evidence was in the report at the time; it just was never used to carve out an exception to "sector beta." This is an internal contradiction, not missing information.

3.3 ⭐ The Theme the Pre-Market Brief Missed/Misjudged: Tankers, and Why "Withdrawal Confirmed" Is Not a Permanent Label

Pre-market §6.2, item 2, original text: "Oil +3.4% while tankers barely moved … this is the same pattern as in the 9/9 report, where 'oil passed the verification point with +3.71% of headroom while tanker equities failed to rise across the board,' recurring for a second consecutive day**." On that basis they went onto the avoid list.**

Today's actual:

⇒ Tankers were the only stock group today that kept a positive elasticity to oil, and the only group that was "weak pre-market, strong at the close" — exactly the inverse of energy equities' "strong pre-market, weak at the close."

Two lessons:

1. "Unconfirmed for a second consecutive day" was treated as a permanent label. On day three it delivered. Freight-rate logic is inherently different from oil-price logic (Strait of Hormuz transit disrupted → rerouting → ton-mile demand rises → freight rates rise), and its transmission is slower than "oil price → upstream selling price" but also sturdier: it benefits from the shipping disruption itself, not from the level of oil prices. A day when oil rises and energy equities fall is precisely the day that distinction becomes visible.
2. ⚠️ But this must not be written up too strongly either: +1.26%+2.52% against WTI's +6.7% is an elasticity of only 0.190.38, still far below 1. Calling it "delivered" is right; calling it "launched" is insufficiently evidenced. This report's conclusion: tankers are upgraded from "avoid" to "watch," not to "recommended."

3.4 Other Single Stocks Worth Recording

• BKR −6.66% (vol ratio 1.71x) — the largest single drag on today's energy sector, and it was an idiosyncratic single-stock event, not the sector. At the Barclays conference the CEO said integration costs from the Chart Industries acquisition and an initial margin of roughly 17% will suppress near-term results, and cut the 2026 free-cash-flow conversion target to 40%–45% (secondary source). The pre-market list did list BKR as avoid, but the reason given was "pre-market notional $0.4M, quote untrustworthy" — avoided correctly, with a completely wrong reason. This was a case of "doing the right thing for the wrong reason."
• AVAV: the analysis in pre-market §5.2 was precisely validated by the intraday path. Open $146.77 → intraday spike to $159.24 (+13.1%) → close $147.07, −7.64% off the high and only +0.12% versus the pre-market price, vol ratio 4.87x. The pre-market line "the so-called +4.33% is basically just giving back the 9/9 daytime decline; the two-day net move is close to zero or even slightly negative" still held at the close (−1.15% versus the 9/8 close of $148.78). This is the finest item in this reconciliation: it made a contrarian call on a gap day and was fully confirmed.
• SHOE: closed $12.28 (−5.03%), a +20.98% rebound from the pre-market price of $10.15. It was listed as "avoid" pre-market with the explicit note that "names down too much should not be chased short" — conclusion correct. This is the textbook V-shaped fill of a micro cap after an extreme gap, and it again shows that shorting a −21% gap is negative expectancy.
• COO's vol ratio was 7.25x, recovering intraday from a low of $51.01 to $54.17. The pre-market panic was absorbed (see §2.1 and §6④).

---

4. After-Hours Earnings Moves (Next-Session Catalysts)

⚠️ Read-moment qualification for this entire section: the after-hours prices below were read at 16:10 and 17:03 ET. ORCL / ADBE / RH all began their calls at 17:00 ET — these prices reflect the 8-K numbers and contain almost nothing from the calls. By the next pre-market session they may be entirely different.

4.1 ⭐ ORCL (Oracle / 甲骨文) — After Hours +6.3%~6.5%, While the Pre-Market Verification Point Said "Negative"

Regular-session close $152.94 (−5.38%, vol ratio 2.08x); after hours $162.60–$163.17 (+6.3%~+6.5%, two independent channels read at 16:10 and 17:03 ET respectively).

Earnings facts (SEC 8-K / EX-99.1 primary):

⭐ The pre-market verification point's verdict — it has to be reported honestly that it read negative:

Pre-market §9①, original text: "Guidance of ≥ +35% is required for consistency with the full-year target; < +32% means the full-year $90 billion has to be cut."

Recomputing with tonight's actual numbers (FY26 Q2+Q3+Q4 = $52.432 billion):

• Q1 actual $19.3 billion ⇒ Q2–Q4 needs ≥ $70.7 billion, i.e. +34.8% y/y
• Q2 guidance is +30%~+34%, midpoint +32%, and even the top of the range at +34% is below 34.8%
• ⇒ By the pre-market criterion, this falls on the "must be cut" side. Yet Oracle explicitly maintained ≥$90 billion and gave EPS of $8.10. The burden of proof has been pushed onto Q3/Q4 — they now have to grow more than 35% y/y to make up the gap.

⇒ Conclusion: the verification point's arithmetic was right, and the contradiction was not resolved tonight. But the market did not trade it. The +6.4% after hours was trading two other things: (a) RPO of $664 billion beating consensus by $33.4 billion — the pre-market list had written that "if RPO is flat or declines, it damages both the growth line and the concentration line"; RPO not only didn't decline, it added roughly another $26 billion sequentially; (b) EPS beat by 10.3%, far more than the 0.9% revenue beat.

⭐ This was a failure of the "criterion correctly constructed, but not measuring the pricing variable" type. The pre-market brief set its verification point on the internal consistency of revenue guidance with the full-year target — a correct and important analytical question; but for a company whose RPO equals 33x its quarterly revenue, what the market prices in the current period is the order book, not the current quarter's revenue guidance. The criterion should have been set on RPO and capex.

⚠️ Three cracks the pre-market list had already flagged, further cemented by tonight's numbers (they have not disappeared; they just weren't priced tonight):

1. SaaS only +10%, while IaaS was +121%. Growth is driven entirely by capital-intensive, low-margin IaaS — exactly the continuation of what pre-market §5.6 flagged as "blended gross margin FY25 70.51% → FY26 65.82%."
2. Quarterly capex of $28.499 billion. Annualized at that rate, roughly $114 billion, more than 2x all of FY26's $55.663 billion. FCF was −$5 billion this quarter, and the +184% surge in operating cash flow itself needs explaining (the pre-market brief noted that RPO includes roughly $75 billion of "customer prepayments for GPUs" — prepayments lift operating cash flow and RPO simultaneously, and tonight's two brightest numbers may come from the same source). ⚠️ This machine did not obtain a breakdown of this quarter's deferred revenue/customer prepayments from the 8-K; that linkage is an inference and is unproven.
3. RPO customer concentration is still undisclosed. The pre-market brief had verified that "the 10-K only discloses on a revenue basis that no single customer is ≥10%, and has never disclosed RPO customer concentration," and tonight's EX-99.1 likewise gives no breakdown. How many customers account for the $209 billion y/y increase in RPO remains unknown.

4.2 ADBE (Adobe / 奥多比) — Beat + Raised Full-Year Guidance, Still −2.15% After Hours

Regular-session close $248.83 (−2.37%, vol ratio 2.11x); after hours $243.41–$243.48 (−2.15%~−2.18%, 17:03 ET).

⭐ Why did beat + raise still fall? Three reasons; the latter two are the main ones:

1. The beat was too small to constitute a surprise. Revenue beat consensus by 1.0%, EPS by 0.8% — for a software name with non-trivial implied volatility, that magnitude sits inside the noise. And the consensus itself is an echo of company guidance (consensus of $6.69 billion falls right inside guidance of $6.67–6.72 billion, and the EPS consensus of $6.08 sits dead center in guidance of $6.05–6.10). "Exceeding the top of your own guidance by a hair" is routine, not new information.
2. ⭐ The repricing happened before the print, not at the print. ADBE is −32.9% from its 52-week high, having fallen 25% in 2024, 21% in 2025 and roughly 18% year-to-date in 2026 (secondary source) — the market has repriced the "AI disrupts software" worry repeatedly over the past two years. An in-line quarter cannot refute a narrative about three years from now.
3. ⚠️ Leadership uncertainty layered on top. On 9/3 the company announced that Anil Chakravarthy will become President and CEO on 2026-12-01, with Shantanu Narayen moving to Executive Chairman; the stock fell about 2% after hours that day. At the same time the CFO position is in an open search (secondary source). During a CEO handover window, a "steady" quarter is not enough to make anyone add.

⇒ The pre-market list's placement of ADBE on "avoid (ahead of the event)" was correct, and correct in both directions: it avoided the −2.37% regular-session decline and it avoided the after-hours trap of falling despite a beat+raise.

4.3 RH — After Hours +6.64%, and a Tariff Mechanism Identical to AEO's

Regular-session close $134.02 (−3.84%); after hours $142.92 (+6.64%, 16:10 ET).

(SEC 8-K / EX-99.2 shareholder letter, primary. ⚠️ Note: RH's EX-99.1 is merely a notice pointing to the shareholder letter and contains no numbers at all; the numbers are all in EX-99.2.)

⭐ This is the mirror image of AEO and is worth reading side by side with pre-market §5.3:

The pre-market list spent a whole section unpacking AEO's "+980bp of gross margin conceals a $179 million tariff refund; restated, roughly −320bp." RH tonight had the same mechanism — a $55.1 million tariff benefit, 600bp — but the opposite outcome, because the disclosure approach differed:

⇒ A one-time item by itself does not determine the direction of the stock; "what's left after you strip it out" does. Strip it from AEO and you get a negative, and the company wouldn't strip it; strip it from RH and it's still positive, and the company stripped it first. The pre-market list's restatement method for AEO got a clean control test on RH today — the same yardstick, two opposite conclusions, which is far stronger than having AEO as the only sample.

4.4 ⚠️ Note on Sourcing the After-Hours Moves

This report only confirmed three key names with material after-hours moves: ORCL, ADBE and RH. The "biggest after-hours movers of the day" type summaries retrieved did not match this machine's measured quotes name by name (those summaries claimed TTAN −21.2% and CHYM +9.6%, whereas the measured figures were TTAN regular session −1.94%/after hours +0.46% and CHYM −5.44%/after hours −0.14%), so they were judged to be prior-period or mismatched content, discarded wholesale, and excluded from this report. Coverage of after-hours moves in this section may therefore be incomplete.

---

5. Flows and Sentiment

5.1 Sector Rotation (SPDR sector ETFs, 16:00 ET close)

5.2 ⭐ Three Control Tests That Can Falsify the Attribution

(a) "Today was a risk-aversion day" — negated by the defensives themselves. On a genuine risk-aversion day XLU should lead; today XLU −0.98% and XLRE −0.83% both underperformed the market's −0.58%; the only sector to rise, XLP, was up just +0.05%. ⇒ Money was not buying safety, it was cutting duration.

(b) "This is a reflation trade" — negated by the commodities themselves. Only oil rose: silver −6.61%, copper −5.23%, gold −2.33%. Copper is the purest demand proxy, and it fell 5.23%. ⇒ This is a supply shock, not a demand recovery. The pre-market brief's characterization here was entirely correct, and today it replayed at larger magnitude.

(c) "This is pure duration de-grossing" — negated by the semiconductor/software divergence. If it were purely the discount rate, software (IGV −0.62%) should not have been 1.82pp more resilient than semiconductors (SMH −2.44%), and utilities and real estate should not have fallen only a third as much. ⇒ There is a non-rate component in semiconductors; for a second consecutive trading day, the attribution remains unconfirmed.

5.3 Risk-On / Risk-Off Characterization

Risk-off, of the "rate-driven broad de-grossing" type, not the "event-driven panic" type.

---

6. Next-Session Outlook (Friday, 2026-09-11)

① Theme Continuity

② Tomorrow's Earnings and Macro Calendar

⭐ Today's EIA inventories (pushed to today because of Labor Day; data week ended 9/4) — a bearish number that oil ignored:

⇒ An overall bearish inventory report, and oil still rose 6.7% that day. This shows pricing power in oil right now lies entirely on the geopolitical supply side, not on U.S. inventories. Next-session inference: only Middle East news can turn oil around, not data.

⭐ This inventory report also fills in the evidence gap in pre-market §7.2: gasoline +1.27 million barrels and distillates +2.09 million barrels mean refined products are building. That is a second independent piece of evidence pointing the same way as "crack spreads did not expand this morning, −0.16%." When the pre-market list downgraded refiners it flagged itself that "the comparison window only covers 1 day, evidence insufficient" — today's inventory data upgrades that downgrade reason from "insufficient evidence" to "sufficient evidence," and MPC −1.76% / VLO −0.91% / PSX −0.87% have already confirmed it with price.

③ Key Names to Watch (Ticker + Falsifiable Verification Point)

⚠️ Any directional judgment before tomorrow's 08:30 CPI is betting on a single data point. All the verification points below are designed as "how to judge whether to keep holding after the data prints," not "whether to buy before the data."

Design notes and defects that must be disclosed:

1. The five names cover four different drivers — product cycle (AAPL), order book and capex (ORCL), crude (DVN), shipping disruption (FRO), within-sector exception (QCOM). This is deliberate: one of today's lessons is that two of the three watch-closely names pre-market (XOM/OXY) were two expressions of the same oil trade, and they failed simultaneously today.
2. ⚠️ Concentration defect disclosed honestly: DVN and FRO both depend on the Middle East situation not easing, and a ceasefire would break both at once. — but their transmission channels differ (oil price vs freight rates), so they are not identical exposures.
3. ⚠️ The biggest defect in this section: apart from ORCL, the other four have no fundamental/valuation verification and rest entirely on price, vol ratio and relative performance. That is a drop in evidence grade; weight them down accordingly.
4. ⚠️ All five verification points are set on relative performance or structural conditions, and not one is on absolute moves — because tomorrow's 08:30 CPI will hit every name at once, so absolute moves will carry almost no single-stock information tomorrow.
5. Deliberately excluded: XOM/OXY (today proved energy equities don't follow oil, and the pre-market recommendation is already at a loss), all refiners (crack spreads + refined product builds, doubly confirmed), AI power (worst group of the session), semiconductor equipment (day 2 and still no attribution), ADBE (beat+raise and still fell, CEO/CFO handover window).

④ Areas to Avoid

1. Refiners (MPC / VLO / PSX) — the evidence is now twofold. Crack spreads did not expand this morning, plus today's EIA gasoline +1.27 million barrels and distillates +2.09 million barrels (refined products building). Oil is a refiner's cost; oil rose 6.7% today and all three fell, which is price confirmation already.
2. AI power / nuclear (OKLO / SMR / VRT / GEV) — average −5.08% today, worst group of the session. Squeezed simultaneously by rates (unprofitable, long duration) and the receding AI capex narrative. ⚠️ But note that ORCL's after-hours quarterly capex of $28.499 billion is a counter-signal — if the call reinforces capex, this group could reverse tomorrow, so don't short it.
3. Semiconductor equipment (LRCX / TER / KLAC / AMAT) — day 2, attribution still unconfirmed. Don't short a decline you don't understand, and don't bottom-fish it either.
4. Establishing any directional position ahead of CPI. Tomorrow's 08:30 print is the last inflation reading before the FOMC, and market hike pricing is already elevated at 56%–63% — which means "in line" is not necessarily good news, and the upside elasticity of a "below expectations" print may exceed the downside elasticity of an "above expectations" one. The direction is asymmetric and the magnitude is unpredictable.
5. Chasing names that closed at the day's high today (AAPL +3.56% closed at the high, DVN closed at the high). Today AVAV fell −7.64% from its high and RXT −11.83% from its high, already demonstrating how this market currently handles gap-ups and spikes.
6. Chasing shorts on an extreme gap. Today SHOE was +20.98% versus its pre-market price, COO +3.64%, FLWS +3.51%, LOVE +2.00%, RWT +1.12%, AEO +0.97% — 6 of the 6 names that blew up on 9/9 after-hours earnings closed above their pre-market prices. The pre-market extreme on a gap day is the worst selling point of the session.
7. Chasing shorts in precious metals after the oversold move. Silver has fallen hard two days running (−6.61% today), so technical bounce risk is rising.

⑤ Input Notes for the Next Pre-Market List

1. ⭐ Promote the "XLE/WTI elasticity ratio" from a footnote to a veto item for the energy branch. Today it collapsed from 0.42 to −0.09. Suggested rule: if that ratio is < 0.2 for two consecutive trading days, energy names may not occupy any recommendation slot, no matter how much oil rises. And the ratio may only be computed close-to-close; three pre-market snapshots do not constitute a trend — that is today's most expensive lesson.
2. ⭐ The liquidity threshold and the ranking must be decoupled. The pre-market notional threshold ($5 million) answers only "can this pre-market quote serve as evidence"; it must not be used for downgrading or ranking. DVN ($1.9M, +2.12%, best in the group, vol ratio 1.69x) and FANG ($0.6M, +1.36%, second in the group) proved that today. Suggestion: names with low pre-market notional should be labeled "pre-market signal unusable, needs price/volume confirmation in the first 30 minutes after the open," not "watch only."
3. ⭐ The "already at a 52-week high ⇒ premise consumed" criterion went a perfect run today (COP/CVX/MPC/VLO all underperformed XOM); recommend codifying it as a standard criterion.
4. ⭐ Verification points must be set on the pricing variable, not on the most important analytical question. The internal consistency of ORCL's revenue guidance is a correct and sharp analytical question, but what the market prices is RPO. Suggestion: before writing a verification point, ask "if this reads negative and the stock soars, would I be surprised?" — if not, it isn't measuring the pricing variable.
5. ⚠️ Criteria of the same kind must be demoted symmetrically in both directions. The pre-market brief had already recognized that "high pre-market notional" cannot be used for ranking and actively demoted it, yet still used "low pre-market notional" to downgrade DVN/FANG. An indicator cannot be invalid in only one direction.
6. ⭐ "Unconfirmed for N consecutive days" is not a permanent label. Tankers delivered on day 3. Suggestion: for any branch tagged "receding/unconfirmed," re-examine it once every day and state explicitly "is it still unconfirmed today," never carrying over yesterday's conclusion.
7. Three things that must be re-checked tomorrow: (a) the ORCL/ADBE/RH call content — this report's after-hours prices were all read at 17:00–17:03 ET and contain no call content; (b) the attribution of the semiconductor discount — already day 2, and if there is still no attribution on day 3, "attribution unclear" itself should be upgraded to a standalone risk criterion; (c) VLCC/Suezmax freight rate indices — the tanker conclusion currently lacks this primary evidence.
8. Cross-verification discipline: today there were 3 instances where a search summary conflicted with primary data and was overturned by it (see the operations notes). The pre-market list's primary-source-first principle was again proven necessary today, not a formality.

---

⚠️ Risk disclaimer: this recap is an after-the-close review of information and observations only and does not constitute investment advice. Data may differ in timeliness or basis of measurement; please rely on company disclosures/SEC filings, and do not use this directly as a basis for trading.

Operations Notes (not sent to clients)

Data sourcing and tooling status for this run:

1. yfinance was not used at all this run; zero rate-limiting problems. The primary channel was the CNBC quote API (quote.cnbc.com/quote-html-webservice), which ran through 60+ tickers in one pass, covering open/high/low/close, volume, 10-day average volume, 52-week high/low, after-hours price and after-hours timestamp, without a single failure. Suggestion: codify the CNBC API as the default primary channel for U.S. recaps and demote yfinance to backup. The scripts are saved as work_usrecap0910/cnbc.py (batches of 12 tickers) and vol.py (vol ratio).

2. ⚠️ New pitfall: CNBC's symbol for the Nasdaq Composite is .IXIC, not .COMP. Writing .COMP silently returns all None (no error raised), while writing COMP returns an individual stock called COMP ($10.13, −5.15%) — neither mistake raises an error, and the second one even hands back a number that looks perfectly normal. This is the same family as "cboe-stale-symbol-silent-trap." .IXIC's 26,081.72 / −0.65% has been cross-confirmed with an independent news source. Recommend adding to memory.

3. ⚠️ CNBC's Treasury change_pct has been broken for three consecutive trading days. Today for US2Y: last 4.588% > CNBC's own previous_day_closing of 4.427%, yet change_pct showed −0.3008%. Per memory, everything was switched to self-computation from the Treasury CSV fixing. This has now hit 3 days in a row; recommend upgrading it from a memory note to a hard filter at the script layer (drop the change_pct field whenever reading US*Y).

4. ⭐ The biggest takeaway this run: three search summaries were overturned by primary data, and all three would have changed conclusions in the body.

   • (a) The 30-year auction. WebSearch explicitly answered that "there was no 30-year auction on 9/10; the next one is 10/8." That is wrong. Querying the TreasuryDirect API directly confirmed that 9/10 did have a $22 billion 29-year-11-month reopening (CUSIP 912810UW6, high yield 5.308%, BTC 2.61, primary dealers awarded just 2.2%). Had the search result been trusted, all of §1.4 would have been deleted — and that section is exactly what rules out the "supply shock" explanation and confirms the "rate-hike pricing" main line.
   • (b) The after-hours movers ranking. The search summary gave TTAN −21.2%, CHYM +9.6%, MIND −15.6%, which did not match the measured quotes name by name (TTAN was actually −1.94% in the regular session/+0.46% after hours; CHYM −5.44%/−0.14%). The content was complete and self-consistent with no visible flaw; only pulling actual quotes exposed it. It was discarded wholesale, and §4.4 tells clients honestly that coverage may be incomplete.
   • (c) AAPL's gain. The search summary said "up 1.1% on Thursday"; the actual close was +3.56%. The root cause is that the piece was written intraday and the summary carried no read moment. Since AAPL is the only fully delivered recommendation in this report and the pivot of the entire narrative, $326.57 / +3.56% was independently confirmed through two channels, CNBC and the Nasdaq official API, and corroborated with AAPL's own intraday range (low 316.51, high 326.74, close 326.57) as a trend candle closing at the high rather than a small 1.1% gain. This is a superposition of "premarket-quotes-accumulate-snapshot-is-not-fact" and "stale-source-silent-lag."
   • ⇒ All three were cases of "the search summary reads perfectly normal and is only exposed against a primary source." The primary-source-first discipline directly rescued three passages this run.

5. ⚠️ RH's numbers are not in EX-99.1. RH's 8-K EX-99.1 is only 7KB and its content is "results have been published in the shareholder letter, see ir.rh.com" — not a single number. The actual numbers are in EX-99.2 (a 1.06MB shareholder letter). Grabbing only EX-99.1 would lead to the wrong conclusion that "the company disclosed no numbers." This is a variant of the same class of silent failure as memory's "edgar-cik-leading-zero-empty-file": HTTP 200, document intact, just without what you need. Suggestion: the 8-K scraping script should automatically try EX-99.2 whenever EX-99.1 is under 15KB.

6. ⚠️ The read moment of after-hours quotes is this report's biggest basis risk, and it is flagged explicitly in two places, at the top of the body and in §4. ORCL / ADBE / RH all started their calls at 17:00 ET, while this machine's after-hours reads were at 16:10 and 17:03 ET — containing almost nothing from the calls. Memory's "after-hours-quote-needs-call-timestamp" is a direct hit. Suggestion: if the U.S. recap slot runs around 17:00 ET, write "after-hours prices were read before the earnings calls" into the template as a standing disclaimer rather than adding it ad hoc each time.

7. ORCL's closing price differs by 0.15% between two channels: CNBC's 16:10 read was $152.94, while the Nasdaq API's back-computed baseline was $153.17. The magnitude is consistent with closing-auction/last-trade differences and does not affect any conclusion; the body uses $152.94 throughout and gives the after-hours gain as a range (+6.3%~+6.5%) in §4.1, without elaborating separately.

8. Market breadth comes from the WSJ Markets Diary mdc_marketsdiary JSON API (the HTML page is JS-rendered, so scraping the page directly yields no data). One call returns four groups — NYSE / NASDAQ / NYSE American / NYSE Arca — including advance/decline counts, new highs/lows, TRIN and up/down volume. This is a newly opened channel this run and the quality is high (TRIN and Arca's ETF breadth are both hard to get elsewhere); recommend codifying it. The script invocation is in the curl commands under work_usrecap0910/.

9. ⚠️ No sub-agents were run this time (per the explicit constraint of this session). Against memory's "研报子 agent 值得等": this report's ORCL capex/RPO linkage, AAPL's relative-strength attribution, and the day-two attribution of the semiconductor discount are all exactly where fundamentals-analyst and risk-auditor would usually add reinforcement. Item 3 of the design notes in §6③ honestly discloses to clients that "apart from ORCL there is no fundamental verification."

10. Self-check checklist for this run (reusable for the next report): (a) cross-confirm the three major index levels with an independent news source; (b) confirm AAPL's closing price through two channels; (c) use the official API rather than news for the 30Y auction; (d) tag every after-hours price with its read moment and compare against call times; (e) force a "vs pre-market price" column into the reconciliation table (memory "recap-vs-premarket-price-column"); (f) add a "High→Close" column to the reconciliation table to identify spike-and-fade (today it caught AVAV −7.64% and RXT −11.83%); (g) use the vol ratio column to distinguish "a real decline" from "nobody traded it" (today it caught IGV at 0.59x, which directly discounted the "software was resilient" evidence). (f) and (g) are the two columns added this run; recommend codifying them into the template.