US Pre-Market Brief | 2026-09-21 (ET) Monday

Coverage window: 2026-09-18 (Fri) 16:00 ET regular-session close → 2026-09-21 (Mon) 08:27 ET Quote read-timestamp: unless otherwise noted, all pre-market quotes are taken from 2026-09-21 08:00–08:15 ET (Nasdaq official API, isRealTime=true, including pre-market volume), and were re-checked once at 08:27 ET before finalization (see "Pre-Finalization Re-Check" below); index/sector closing data are on the 09-18 16:00 ET close basis. Pre-market change baseline: uniformly the 09-18 close, verified name by name by back-computing last price − change amount.

---

0. Today in One Sentence

1. There is only one trade today: Saudi crude exports have recovered (back above 4 million bpd in September, versus a low of just 2.4 million bpd in August), and WTI fell for a 4th consecutive session to $97.57 (−2.72%) — and it was precisely oil-driven inflation that forced the Fed's first rate hike in over three years on 9/16. Oil eases → inflation expectations and the term premium fall back → yields decline across the curve → duration and high-beta assets rebound broadly. (⚠️ Note: the shape of the yield move is that 2Y moved least and 10Y moved most, which looks more like a term-premium shift down than a "repricing of the hiking path"; this report has no direct witness for the latter — see verification point ③ in Section 8.)
2. Strongest catalyst: falling oil (macro) + bitcoin rose overnight from 81,185 to about 85,055 (+4.77%, dispersion across three cross-verified sources < 0.06%), a seven-month high, accompanied by $262–648 million of short liquidations. Strongest themes: crypto stocks > semiconductors/optical modules > AI power. ⚠️ Note: there is almost no genuinely same-day single-stock catalyst today (see Section 1) — this is a pure macro-beta day.
3. Flow direction: the only sector down in the pre-market as a whole is energy — XLE −0.93%, XOP −1.08%, OIH −0.81%, with refiners and some E&Ps leading the declines (MPC −1.98%, COP −1.62%, VLO −1.40%, PSX −1.40%); everything else is broadly higher, SMH +2.02% > ARKK +1.89% > XLK +1.43% > QQQ +1.14% > IWM +0.82% > SPY +0.72% (⚠️ IWM and SPY differ by only 10bp, within the pre-market bid-ask spread, and this does not constitute evidence of small-cap strength).
4. Driver type: primarily macro (oil → rates). ⚠️ There is almost no genuinely same-day single-stock catalyst today: JPMorgan's IREN upgrade was 9/14, the SEC tokenization exemption was 9/18, and the INTC/SK Hynix report was 9/16. This week has no mega-cap tech earnings, but that is not the same as "no major earnings": Costco COST reports after the close on 9/24 (Thursday), and Accenture ACN, Nike NKE, Darden Restaurants DRI and others are on this week's calendar (ACN is a bellwether for AI services demand, NKE for the consumer). It is also a heavy week for Fed speakers (10 appearances in total).
5. Pre-market state: S&P futures +0.72%, Nasdaq futures +1.10%, Dow futures +0.86%; 10Y Treasury 4.947% (−4.9bp), 2Y 4.705% (−3.8bp), 30Y 5.288% (−3.9bp), with TLT +0.70% independently confirming the decline in yields; the dollar index is flat to slightly soft (EUR/USD 1.1490 +0.06%, USD/JPY 157.20 +0.22%); VIX closed Friday at 14.81 and printed 14.85 (+0.27%) at 08:16 ET this morning — with futures sharply higher, a VIX that ticks up rather than down is a mild non-confirmation signal; October VIX futures 17.75 (−0.29).

⚠️ The single most important sentence in this report: almost every long name on today's list is a different expression of the same variable (oil → rates). 20 tickers is not 20 independent bets — it is 20x correlated exposure to 1 bet. See Section 8.

---

🔴 Important Correction (08:40 ET, before the open)

After completing primary-source verification against SEC/EIA, this report overturned two core judgments from the first draft. Because the correction occurred before the open, it is written directly into the body rather than appended afterwards:

⚠️ The IREN error deserves separate comment: in this very report I assigned INTC a full risk deduction for "treating 9/16 stale news as a same-day catalyst", and yet in the same report I made exactly the same error with IREN. The test was applied to others but not to myself. This is the single item most worth remembering this session.

---

Pre-Finalization Re-Check (08:27 ET)

The body data basis is 08:00–08:15 ET. One more read was taken before finalization; all directions were unchanged, but two readings are worth recording:

⚠️ A confession about NVDA: the verification point set for NVDA is "downgrade if open→close underperforms AMD by more than 1pp". Before the open, its pre-market performance was already moving in the unfavorable direction (+1.06% → +0.67%). This does not trigger the verification point (the test only looks at open→close after 09:30), but readers have a right to know this signal before the open. It is also a reminder: this NVDA call was supported by the single dimension of "position" to begin with, and it is the only recommended slot in this report whose fundamental verification was incomplete.

---

1. News Overview

Basis-correction note: multiple secondary sources (including some TipRanks/Investing.com pages) say Micron reports on "September 23" — that is Micron's fiscal 2025 Q4 date. Micron's own 8/26 press release explicitly says September 30. There is no Micron earnings report this week; do not position on that basis.

---

2. Strongest Themes, Descending

---

3. Single-Stock Strength Master List (partitioned by direction, descending by total score within each group)

Pre-market change baseline = 09-18 close (verified by back-computing last price − change amount, and matched name by name against the Nasdaq official secondaryData closing prices).

🟢 Bullish direction

🔴 Bearish direction

---

4. Single-Stock Scoring Model (100 points)

⚠️ NVDA's "expectation gap" has been revised down from 9 pts to 4 pts, dropping its rank from #1 to #2: the first draft put "still negative return since 9/3" into the expectation-gap column, but that is price position/chip structure, not an expectation gap (an expectation gap should mean the shortfall between fundamentals or consensus and reality). And that same "position" factor had already been counted once in "trading characteristics 10 pts" and once in "risk deduction of only −3" — the same variable was counted three times. This is the same error as the criticism in Section 8 that "20 tickers is not 20 independent bets", only occurring inside a single name's model. After correction: NVDA has no same-day primary information and consensus has not changed, so the expectation gap is scored 4 pts.

Scoring note (why the biggest gainers got the lowest scores): INTC, AMD, CIEN and MSTR are among the top pre-market gainers, but "expectation gap" and "risk deduction" pushed their totals down — their common problem is that the catalyst is already priced and the position is already high. INTC is the clearest case: it has the heaviest pre-market notional turnover in the field (about $530 million), yet not one piece of same-day primary news can be found.

⚠️ Scope of applicability of this scoring model (found in QA, noted here): this table is used only for internal ranking of long candidates and does not produce the final list in Section 9. Of the five watch slots in Section 9, VLO (short), IBIT and ANET never entered this table — short candidates and ETF instruments do not fit this model's "catalyst directness/expectation gap" basis, while ANET was selected on "position within the sector + volatility" rather than a score. Conversely, MRVL (64), which scores higher than they do, did not make the final five. Readers should not read Section 9 as the output of this table.

⚠️ Screening basis for the Section 3 master list: this report actually pulled data on 102 individual stocks + 28 ETFs/indices, and the 10+6 names in Section 3 are the result of manual selection, based on "top pre-market movers + an attributable, discussable theme" — not a mechanical sort by percentage change. As a result, names with larger gaps than AMD/AMAT, such as ARM (+4.76%) and CRCL (+5.71%), appear only in Section 7 and never entered the master list or got scored. This is a subjective screening step and readers should be aware of it.

⚠️ This table was revised down once after completing primary SEC financial verification: COIN went from 67 to 57 (revenue is actually −18.5% year over year, and after stripping out all crypto fair-value gains/losses and one-off restructuring, core operations still lost money), and HOOD went from 63 to 60 (its "expectation gap" is negative — the market prices it as crypto beta, while crypto is only 7.6% of its revenue). Both labels were simultaneously downgraded from "watch closely" to "watch only". See the COIN and HOOD subsections below.

---

5. Detailed Analysis of Top Names

Note: this section is ordered by writing sequence, which does not equal the Section 3 master-list ranking (the master list was adjusted through two rounds of verification: IREN fell from #1 to #10, NVDA returned to #1, COIN and HOOD were revised down). Each subsection heading notes the name's actual master-list rank.

NVDA 英伟达 (NVIDIA) (Overall #1) — the "least repaired" mega-cap in the field

• Related news: no primary company news today. The background is that Anthropic CEO Dario Amodei wrote on 9/12 (Saturday) that AI companies "must slow the pace of improving model capability", with Sam Altman and Musk subsequently agreeing; the piece was published over the weekend, and the market's pricing occurred on 9/14 (Monday, NVDA −3.36% that day), driving a selloff in the AI buildout chain (compute/chips) while money rotated into GOOGL/MSFT/META, which are less capex-dependent (24/7WallSt¹⁴).

• Catalyst logic: falling rates → lower discount rate for long-duration growth stocks, benefiting valuation; at the same time the "AI deceleration" narrative is ebbing. What is affected is valuation and sentiment, not current revenue/EPS.

• Theme stage: repair phase (not crowded). This is its biggest difference from peers —

  	9/03 close	9/18 close	Period change	Today pre-market
  NVDA	228.45	222.27	−2.7%	+1.06%
  AMD	456.16	559.82	+22.7%	+2.84%
  INTC	91.67	108.60	+18.5%	+5.48%

  Table scrolls horizontally

  Among semis, NVDA is still below its 9/3 level while AMD/INTC have far exceeded theirs. If the "AI deceleration" narrative ebbs further, NVDA has the most repair room and the least gap-fill risk.

• Fundamental verification (SEC EDGAR primary, completed after the first draft — the conclusion is upgraded from "position only" to "hard evidence"):

  • Q2 FY2027 (ended 2026-07-26): revenue $96.221 billion (+106% year over year); data center $89 billion (+117% year over year); gross margin 75.0% on both a GAAP and non-GAAP basis; operating margin 66.2%.
  • Q3 guidance of $108.0 billion ±2% (+89.5% year over year), gross margin 74.0%, and the company explicitly states the guidance includes no data center revenue from China — a free option not yet in the price.
  • 🔴 Here is the real fulcrum of this recommendation: TTM GAAP P/E = 222.27 / 7.91 = 28.1x (cross-verified against CNBC's 28.10), and a self-built 3-year TTM P/E series (n=643) shows this valuation sits in the 1.7th percentile of the past 3 years (median 51.6x, 4.4th percentile over the past 1 year). This is the cheapest zone since the ChatGPT era began. Even using the non-GAAP 31.7x, it still lands around the 7th percentile of the past 3 years.
  • ⚠️ Basis warning: NVDA's GAAP EPS is actually higher than non-GAAP ($2.46 vs $2.22), because it includes portfolio fair-value changes in GAAP while excluding them from non-GAAP. About 12.8% of the TTM GAAP $7.91 is investment-gain water, so on an operating basis one should look at 31.7x. Also: the commonly cited "forward P/E of 18.4x" most likely corresponds to FY2028 rather than FY2027 (self-computed FY27 is about 23.6x from Q3 guidance), and showing the two side by side makes the valuation look one notch cheaper than it is.
  • Counter-evidence that must be stated at the same time (it explains why the discount may be justified rather than a mispricing): in a single quarter long-term debt rose from $8.47 billion to $33.37 billion (a net increase of about $24.9 billion), while H1 spending on equity securities was $42.4 billion, with equity and non-marketable investments totaling about $94.0 billion, heavily overlapping with customers; days sales outstanding rose from 50.8 days to 59.6 days; H1 operating cash flow/GAAP net income was only 63.1%, below its historical level. NVDA has gone from "zero leverage, pure operations" to "borrowing + large-scale equity deployment into customers".

• Moat: the CUDA ecosystem + system-level solutions, the absolute leader in AI compute.

• ⚠️ One wording correction: the first draft said NVDA had "the cleanest position in the field". By 52-week range position, NVDA is at 80.3%, actually higher than INTC's 70.3%. The accurate statement is: NVDA's 52-week high was set 4 months ago (5/14) and it has traded sideways without a new high since — that is "clean in time", not "lowest in magnitude".

• Pre-market and technicals: +1.06%, 1.55 million shares, a small gap, so there is no "chasing" problem. Volume on 9/18 was 190 million shares (quadruple witching).

• Final judgment: priority deep-dive. It is not the biggest gainer today, but it has the best "gain/crowding" trade-off.

IREN (Overall #10, sharply downgraded from #1) — 🔴 the most serious self-correction in this report: I treated a 9/14 stale item as a same-day catalyst

• 🔴 Correction (discovered after the first draft through primary SEC verification, overturning all core arguments of the draft): the first draft wrote that "JPMorgan's two-notch upgrade" happened this morning pre-market, and on that basis listed IREN as "the only primary-source new catalyst today" and overall #1. That is wrong.
  • The rating action actually occurred on 2026-09-14 (Monday), analyst Richard Choe, Underweight → Overweight, PT $46 → $65. By today it has been digested for 5 trading sessions.
  • IREN opened at $43.83 on the upgrade day and closed at $46.68 on 9/18, a gain of 6.5% in between — the catalyst entered the price long ago.
  • Another detail: the new $65 target corresponds to December 2027 while the old $46 target corresponded to December 2026, so the reference year changed and the increase is not directly comparable.
  • This is the same error as the "treating 9/16 stale news as a same-day catalyst" that I criticized INTC for in this very report, and I committed it in the same report in which I criticized it. IREN has been downgraded from overall #1 to #10, its label changed from "priority deep-dive" to "watch only", and its total score cut from 69 to 30.
• 🔴 Fundamental verification also overturned the "AI pivot" narrative (IREN FY2026 10-K, primary):
  • Of FY2026 revenue of $707 million, bitcoin mining was $578.2 million (81.8%) and AI cloud services only $128.8 million (18.2%).
  • Not one cent of that $9.7 billion Microsoft contract has been recognized on the income statement to date. From the 10-K: "No lease revenue was recognized during the periods presented" — it is treated as an ASC 842 operating lease (with IREN as lessor), not ASC 606 revenue. In accounting terms, this company is still a bitcoin miner today.
  • 87.7% of the "$2.10 billion of operating cash flow" was not earned from operations: $1.842 billion of it is customer prepayments (deferred revenue). That is real cash, and prepayment from a Microsoft-caliber counterparty is a strong endorsement of the contract's authenticity, but it is a liability, not profit.
  • Adjusted EBITDA is −8.9% year over year ($245.7 million vs $269.7 million a year earlier) while revenue was +41.1% — the reason is SG&A exploding from $136.5 million to $449.1 million (+229%). FY26 net loss was $702.6 million.
  • Dilution path: FY24 weighted 99.64 million shares → FY26 actual 394.06 million shares (2026-08-14 10-K cover), 4.0x dilution in two years; and about $3.5 billion of the $6.0 billion ATM program remains unused.
  • Valuation must be recomputed in full: third-party terminals' IREN share-count field (224.5 million shares) is stale and lower than its own FY26 diluted weighted share count, understating market cap by 43%. Using the 10-K cover's 394.1 million shares, market cap is about $18.39 billion, P/S 26.0x (not the 14.8x shown on terminals); on the AI cloud segment alone it is 143x.
  • Contracted backlog: unperformed RPO of $5.1 billion, but only $900 million of that is recognized in the next 12 months.
• What still holds (not negated by the above corrections): both the Microsoft ($9.7 billion / 5 years) and NVIDIA ($3.4 billion / 5 years) contracts have SEC filings; Horizon 1 was delivered in August 2026 and accepted by Microsoft; the $1.84 billion of customer prepayments is a strong endorsement. This is an option not yet falsified, not a pivot already delivered.
• Final judgment: watch only. The catalyst is old, the pivot has not begun on the income statement, and dilution is ongoing.

COIN Coinbase (Overall #6) — the catalyst is real, but the fundamentals are worse than the market believes (downgraded after verification)

• Related news: (a) BTC +4.77% overnight, breaking $85,000 (the only new information today); (b) the SEC's five-year exemption allowing certain platforms to conduct tokenized securities trading — this one acts directly on the scope of an exchange's business, but it was released on 9/18 (Friday), when COIN was already +11.66%, making it already-priced stale news.
• ⚠️ The first draft listed a third item, "analyst price target raised today", but because the institution's name, target price and link could not be obtained, and per this report's consistent standard for unattributed rating news (the most common reality is a repackaged old rating), that item has been deleted, and COIN's catalyst directness and total score have been recomputed accordingly.
• 🔴 Fundamental verification overturned the first draft's judgment (SEC 10-Q, primary): I originally gave it "medium-strong: the leading US-compliant exchange / 67 pts / watch closely"; after verification it is downgraded to 58 pts / watch only. Reasons:
  • Q2'26 total revenue −18.5% year over year ($1.220 billion); retail transaction revenue −30.5%.
  • The narrative that "subscription services can hedge the trading decline" did not hold this quarter — subscription services were themselves −12.2%, with staking rewards halved, −42.3%. The only thing growing is institutional trading (+64.6%, Deribit consolidation).
  • After stripping out all crypto asset fair-value gains/losses ($209.5 million) and one-off restructuring charges ($52.41 million), core operations still lost about $75.30 million. The apparent $1.429 billion of net income in the year-ago quarter included $1.507 billion from investment gains, so core operations were losing money then too. Core operations lost money in both quarters; the entire difference is investment gains and losses.
  • A single counterparty is 26% of total revenue (22% in the year-ago quarter, and rising) — that is the USDC/Circle stablecoin revenue share, and the pricing power does not sit with Coinbase.
  • Valuation: P/S (TTM) of 8.62x looks undemanding, but a low multiple here is the result of a shrinking denominator (revenue), not a margin of safety.
• ⚠️ Evidence from last earnings: on 2026-07-31 it closed −10.59%, gapping −6.41% and then falling another 4.47% intraday — gap and intraday in the same direction is a genuine repricing, not a give-back of front-running.
• Judgment: watch only. Today's catalyst (coin price + regulation) is real, but it lands on a fundamental base whose revenue is shrinking.

HOOD Robinhood (Overall #4) — the market is buying it with the wrong label

• 🔴 I had its risk backwards; correcting here: my first draft wrote "high crypto revenue share, strongly cyclical". Primary SEC 8-K data show exactly the opposite —
  • Crypto is only 7.6% of HOOD's total revenue ($100 million), and it is the only declining segment (−38% year over year).
  • The real pillars are: options 26.1% ($342 million) + event contracts/prediction markets 11.9% ($156 million, more than 10x year over year) = 38% combined, 5x the size of crypto; net interest income is 29.7%.
• This creates a "negative expectation gap": HOOD is +5.17% pre-market today, highly synchronized with BTC's +4.77% — the market is pricing it as "crypto beta" while its revenue mix no longer supports that label. A gain built on a false premise is more prone to give-back.
• Fundamentals are actually fine: Q2 revenue +32% (also +23% sequentially, so not a single-quarter anomaly), net margin 43.8%, adjusted EBITDA margin 56.7%, interest-bearing debt/equity of only 22.7% (note: CNBC's 239.51% counts customer payables as debt and is wrong).
• But the profit line has water in it: GAAP net income includes a $129 million one-off gain from deconsolidating the RVI fund ($0.14/share); excluding it, net income is +15.0% year over year, not the reported +48%.
• Expensive: P/E (TTM) 63.1x, P/S 23.0x, P/B 11.9x, and the 69.4th percentile of its 52-week range.
• Neither of its two legs is in the company's own hands: net interest income (29.7%) is pure rate beta, directly compressed by a cutting cycle; event contracts (11.9%) are pure regulatory beta, which a single CFTC document could zero out. Also: credit loss provisions +100% year over year — a broker is turning into a consumer lender.
• ⚠️ Evidence from last earnings: on 2026-07-30 it gapped +1.19% but went −4.74% open→close — anyone who bought off the pre-market headline lost 4.74% that day. This is exactly the gap-fill pattern this report keeps emphasizing.
• Judgment: watch only.

MRVL Marvell (Overall #2) — the middle choice with relatively solid fundamentals within the sector

• No same-day primary catalyst (a same-day Benzinga article confirms pre-market +2.89% to 251.30, consistent with this report's +2.85%/251.20; it is sector beta).
• Custom ASICs and optical/electrical interconnect both benefit from AI data centers. Per Benzinga, the average target price of 50 analysts is $295.25.
• Judgment: watch closely.

MSTR Strategy (Overall #3) — a high-beta coin-holding vehicle, not "cheap bitcoin"

• Up +16.39% in a single day on 9/18 (54.67 million shares traded, about 3x normal), and another +6.80% pre-market today.
• Key calculation: BTC went from the 9/15 low of 74,913 → about 85,000 today, a gain of about +13.5%; over the same span MSTR went from 129.60 → 164.38 pre-market, a gain of about +26.8%, implying roughly 2.0x BTC beta. That means if BTC falls 10%, this relationship points to about −20% for MSTR. (This beta is measured over a single interval, so treat it as an order-of-magnitude reference, not a stable coefficient.)
• 🔴 "mNAV has fallen below 1 so it is cheap" is the most dangerous misreading today (SEC 10-Q / this morning's 8-K, primary):
  • The 8-K filed this morning (09-21) shows holdings of 846,000 BTC as of 09-20, at a cumulative cost of $63.80 billion, average price $75,416; at BTC ≈ $85,055, the coin holdings are worth $71.96 billion (unrealized gain +12.8%).
  • On "common equity market cap / coin market value", mNAV = 0.897 < 1, which looks like a discount. But the preferred liquidation amount of $15.462 billion + convertibles of $6.714 billion (together 30.8% of the coin market value) all rank ahead of the common. Net of that layer, the common still trades at a premium of about 15%.
  • Perpetual preferreds are a cash outflow with no maturity date: the annual dividend obligation is about $1.675 billion (of which STRC alone is $1.206 billion, coupon 11.5%). Convertibles have put dates that can be negotiated; preferreds do not. Existing dollar reserves of about $6.09 billion can cover roughly 3.5 years assuming no new issuance and no coin sales.
  • "BTC per share" has already stalled out: year to date, share count +25.6% and coin holdings +25.8%, so BTC per share is a net +0.1%; and since 6/30 holdings have been completely frozen at 846,000 coins while the share count keeps rising → BTC per share has grown negatively over the past three months (−5.2%).
  • The single most telling fact: the company has sold bitcoin at $60,773 (below its own $75,416 cost) and used cash to buy back STRC preferreds. Capital allocation has shifted from "issue stock to buy coins" to "shrink high-cost liabilities" — that is a defensive posture, not an offensive one.
• Judgment: watch only. It offers no information beyond bitcoin, only leverage plus a layer of preferred claims ranking ahead of you. To express a bullish BTC view, IBIT (+4.74%) is the cleaner path.

AMAT 应用材料 (Applied Materials) (Overall #5) — note that it speaks to supply, not demand

• +2.57% pre-market. Equipment names benefit from memory/logic capacity-expansion expectations.
• ⚠️ Cognitive reminder: semiconductor equipment revenue reflects the formation of future supply, and cannot be taken directly as confirmation of current memory market conditions. Capacity expansion becomes supply pressure 12–18 months later.
• Judgment: watch closely.

INTC 英特尔 (Intel) (Overall #7) — the heaviest money in the field, buying a piece of stale news

• Pre-market +5.48%, 4.59 million shares, about $530 million notional — the heaviest pre-market money within this report's data coverage.
• But no primary new catalyst can be found for today. The SK Hynix partnership the market widely cites is 9/16 news, and moreover:
  • That news was fully priced over the two sessions of 9/16 (+4.03%) and 9/17 (+7.67%), and 9/18 was already −0.18%;
  • SK Hynix's official line is only that it is "exploring various options and has not finalized any specific plan or arrangement"; other reports say it denied an HBM4E foundry partnership (Bloomberg⁹ / MarketScreener¹⁵).
• Position: 9/3 close 91.67 → 9/18 close 108.60, +18.5% in 10 trading sessions (9/7 was Labor Day, market closed); pre-market today 114.55, already +25% versus 9/3. Volatility over that span was violent (9/10 −5.57%, 9/14 −5.59%).
• Fundamental verification (SEC EDGAR, primary):
  • Q2'26 (2026-07-23 after the close) revenue $16.128 billion, +25.4% year over year, with operating margin turning from −24.7% to +11.1%, and Data Center and AI (DCAI) +59.0% year over year — the operating-level inflection is real.
  • But the foundation of the bull case is almost empty: Intel Foundry posted a single-quarter operating loss of $2.089 billion, and back-computed from intersegment eliminations, its revenue from external customers has a floor of only about $288 million (about 5% of that segment's revenue). In other words, the market is paying for a revenue line that barely exists yet.
  • The company itself guides Q3 non-GAAP EPS down sequentially: $0.42 → $0.38 (−9.5%), even as revenue and gross margin guidance both point up.
  • Valuation: P/S (TTM) 10.01x, in the 98th percentile of 5 years; non-GAAP TTM P/E 99.6x; ROIC 3.42% vs WACC 15.32% — still destroying value.
  • Overlooked large dilution: on 2026-08-10 it issued 242.1 million shares at $95.00/share, raising about $23.0 billion (8-K), diluting about 4.8%; net debt consequently improved from −$20.8 billion to pro forma net cash of about +$2.0 billion.
  • A counterintuitive mechanism: the Escrowed Shares placed with the Commerce Department under the CHIPS Act must be marked to market, so the higher the stock price, the larger the GAAP loss (that item cost $12.529 billion in Q2 alone). Its GAAP profit is therefore structurally unusable for the foreseeable period.
  • Also: an additional roughly $700 million of restructuring charges, taking total FY2026 restructuring to about $2.8 billion, plus a new round of layoffs.
• ⚠️ Evidence from last earnings: on 2026-07-24, with revenue +25%, a beat versus its own guidance, and operating margin turning positive, the stock closed down 7.89% that day.
• ⚠️ A qualification on the "no catalyst" judgment itself: this report ran three rounds of targeted search (same-day dedicated coverage, general search, SK Hynix-specific), all of which returned only the 9/16 stale news; at the same time the three companies' most recent SEC EDGAR filings stop at 8/12, 9/3 and 8/19 respectively, so there are no new company filings today. But "I did not find it" does not equal "it does not exist" — 4.59 million shares of pre-market volume is real money, and it may well correspond to news this report failed to capture. Please read this item as "this report could not find a catalyst", not "there is definitely no catalyst".
• Judgment: watch only. This is the only name in this report with a full −15 risk deduction — the largest money inflow, paired with the weakest same-day evidence, paired with a valuation in the 98th percentile of 5 years.

CIEN Ciena (Overall #8) — the sector rose together, but the stock-level catalyst is missing

• Pre-market +6.24% (52,000 shares, about $19.4 million notional).
• The 9/17 investor forum released three-year targets (2029 revenue target of about $14 billion, backlog of about $10 billion), but the stock was only +1.10% that day — that is, the market gave no strong reaction at the time, so using it to explain today's +6.24% does not hold.
• The entire optical communications sector rose in sync today (CRDO +4.21%, ALAB +3.13%, COHR +3.04%, LITE +2.43%, ANET +1.81%), indicating this is sector-level money, not a CIEN company event.
• ⚠️ Sector volatility is extreme: over the last 10 trading sessions LITE has swung +11.04%, −9.92%, +9.59%, +4.17% in succession; and Lumentum/Ciena/Corning are reported to have already doubled in 2026, making this a crowded trade.
• ⚠️ Pre-market volume is extremely thin, so the quote is not a fact: CIEN's pre-market volume is only about 52,000 shares (another read at 08:14 ET was 54,170 shares), less than $20 million notional. Until volume builds, the +6.24% figure is far less representative than INTC's 4.59 million shares.
• Fundamental verification (SEC EDGAR, primary) — it is actually the most solid fundamentally and the least stretched on valuation in this group:
  • FQ3'26 (2026-09-03 pre-market) revenue $1.6711 billion, +37.0% year over year; operating expenses only +6.6% → operating profit +310%, adjusted EPS +215% year over year. The operating leverage is real.
  • Its non-GAAP basis is the cleanest of the three: it assumes a 20% tax rate, above its actual GAAP tax rate of 14.2% (i.e. the adjustment runs against the company, which is conservative).
  • Valuation: forward P/E 33.58x, PEG 0.47, ROIC 21.97% vs WACC 10.85% (the only one in this group clearly creating value); P/S is in only the 42nd percentile of the past 1 year.
  • But the company's own FQ4 guidance implies peaking margins: adjusted gross margin 46.4% → 45%, adjusted operating margin 22.5% → 20%; back-computing from the guidance midpoint, operating profit is $1.750 billion × 20% = $350 million, below FQ3's actual $375.7 million — revenue +4.7% sequentially while profit is −6.8% sequentially.
  • Concentration risk: two customers together are 41.7% of revenue; FQ3 free cash flow was only 44% of net income (receivables/inventory/prepayments consumed $481 million combined, and capex doubled year over year).
• ⚠️ Evidence from last earnings: on 2026-09-03, with revenue +37%, adjusted EPS +215% and full-year guidance raised, the stock closed down 10.36% that day, having already fallen a cumulative 17.1% in the three days before the report.
• Judgment: watch only. The fundamental quality itself is fine; the problem is that today's gain has neither a catalyst nor volume support, and the company has just lowered its own profitability outlook.

AMD (Overall #9) — impeccable fundamentals, but the least friendly position

• Up +22.7% from 9/3 → 9/18, and another +2.84% pre-market today. No same-day primary catalyst.
• The comparison with NVDA (see the NVDA subsection above) shows: on the same AI logic, AMD has already fully expressed it, NVDA has not.
• Fundamental verification (SEC EDGAR, primary) — the strongest fundamentals in this group, but the price has already used them up:
  • Q2'26 revenue $11.536 billion, +50.1% year over year; data center +107% year over year, now 58% of the total; single-quarter operating cash flow of $2.366 billion, achieved despite a net working-capital drag of $730 million (not by stretching suppliers); net cash +$9.885 billion, the cleanest balance sheet in this group.
  • Its non-GAAP treatment is conservative: $483 million of long-term investment gains ($0.29/share) within GAAP net income was proactively excluded by the company, a direction that works against itself. Conversely it is a reminder: $0.29 of the GAAP EPS of $1.38 is investment gains, so the denominator behind the TTM GAAP P/E of 142.89x has water in it.
  • 🔴 The most critical item — the market is paying for a quarter management has never guided: FY26 consensus revenue is $50.9 billion; subtracting H1 actual of $21.789 billion and the Q3 guidance midpoint of $13.0 billion, the implied Q4 must reach about $16.111 billion (+23.9% sequentially, +56.9% year over year), and the company has only guided to Q3 and has never given any Q4 number. As soon as Q4 guidance comes in below that, the expectation gap is negative.
  • Valuation: P/S 22.13x (95th percentile of 5 years), EV/EBITDA 94.65x, P/TBV 35x, ROIC 9.77% vs WACC 17.80%; the stock sits in the 94.3rd percentile of its 52-week range, the only one in this group hugging its 52-week high.
  • The gaming segment is still bleeding at −31% year over year; GAAP and non-GAAP operating margins differ by 10pp, coming from $544 million per quarter of acquisition intangible amortization, and that part will not disappear on its own.
• ⚠️ Evidence from last earnings: on 2026-08-05, with revenue +50%, data center doubling and an across-the-board beat, the stock closed down 7.04% that day, with almost the entire decline occurring in the overnight gap.
• Judgment: watch only.

---

6. Bearish/Avoid List

Why refiners deserve more attention than integrated oil & gas companies (the hardest piece of evidence in this report)

Many people read this energy move as "oil rises → energy stocks rise". A cross-sectional comparison directly refutes that:

If the driver were the crude price itself, XOM and CVX could not have stayed motionless. What actually rose is refining margins (crack spreads).

Direct readings (not inferred backwards from share prices — the driving variable itself) — a self-computed 3-2-1 crack spread from NYMEX futures, formula (2×RBOB + 1×ULSD)×42 ÷ 3 − WTI:

Two conclusions:

1. The crack spread is indeed narrowing today, and ULSD (diesel) is down −3.37%, more than crude's −2.86% — products falling faster than crude is the definition of refining margin compression. The driving variable itself gives a direction consistent with the share prices.
2. More important is the absolute level: a 3-2-1 crack spread of $67.84/bbl is a historical extreme (the normal range is typically on the order of $15–30/bbl). That is, even though it has only retraced 2% today, it still has enormous room to normalize. Refiners' earnings forecasts are built on top of this number.

🔴 Added after finalization: three pieces of hard data that change this observation (EIA primary + SEC primary)

① The crack spread's historical percentile: even more extreme than my first draft said. Using EIA official daily spot prices to build a 20-year 3-2-1 series (n=4,969,2006-06-14 → 2026-09-15): the current $73.22/bbl (EIA basis, 9/15) sits in the 99.78th percentile of all history, with a 20-year median of only $16.80 and a 5-year median of $24.56. The 2026 year-to-date average is $47.75, whereas even the full-year average of the last super-cycle in 2022 was just $36.76. This strengthens the mean-reversion argument.

② 🔴 But the "Saudi recovery" is not complete — my first draft wrote an expectation as a completed fact.

• The East-West pipeline has been shut in since September 10, with repairs expected to take up to six weeks (i.e. possibly as late as late October).
• YASREF (the Aramco-Sinopec joint venture), one of the largest diesel export facilities on the Red Sea at 400,000 bpd, was struck.
• The 3% oil decline on 9/16 was based on a report that Saudi Arabia might restore half of the pipeline capacity within days — an expectation, not a fact; and on 9/18 there was another report that Saudi Arabia will cut off all crude supply to Europe next month.
• This report could not obtain primary confirmation from Saudi Aramco IR. Therefore the premise of a "supply recovery" currently rests only on media-grade evidence, and that evidence contradicts itself.

③ 🔴 The most important item: the clock is out of sync.

The quarter-to-date Q3 crack spread is 43% above Q2. That means the Q3 results released in late October will very likely be even stronger than the already-explosive Q2 — even if spot is retracing right now. "Refiner give-back" and "deteriorating fundamentals" are not synchronized in time, and the two cannot be used as evidence for each other. This is the biggest risk in this section, and the first draft was entirely unaware of it.

④ What the valuation actually reflects — the answer is counterintuitive. Regressing FY2022–2025 net income against that year's crack spread (MPC R²=0.986, VLO R²=0.954) and back-solving the crack spread implied by consensus:

The two independently compute $38.9 and $39.8, highly consistent (strong cross-verification). That is: the "cheap" 7.66x is not built on "crack spreads holding at peak" but already assumes they fall from ~$68 to ~$39 (−42%). So the real expectation gap is not "will it fall" but "will it fall below $39".

Magnitude of downward revisions (holding share prices constant): back to the 2022 peak-year $36.76 → EPS revised down only about 9–13% (essentially already in the price); back to the 5-year median of $24.56 → about a 60% cut, with P/E becoming 19x/24x; back to the 10-year median of $18.84 → about an 85% cut, with P/E becoming 47x/68x.

⑤ P/B evidence (supports the short, and is harder than P/E): MPC 6.25x, VLO 4.76x, both in the 99.9th percentile of five years, and 1.5–2.4x their 2022 super-cycle peaks (2.56x / 3.15x). ⚠️ Self-check: part of MPC's high P/B comes from buybacks depressing the denominator (about $32.7 billion of cumulative buybacks in 2022–24), so that evidence should be discounted; VLO's equity has been stable in the $23.5–26.0 billion range for five years, so its P/B rising from about 2x to 4.76x is clean evidence.

⑥ 🔴 MPC and VLO are not the same trade, and the first draft was wrong to list them side by side. MPC owns the midstream asset MPLX, with segment adjusted EBITDA of about $1.778 billion per quarter (≈$7.1 billion per year), a fee-based buffer essentially unrelated to the crack spread; VLO is the purer crack-spread beta. Consequently, in a "back to the 10-year median" scenario, VLO's P/E (67.5x) looks worse than MPC's (47.0x). If you want to express this direction, VLO is the purer instrument; MPC comes with a built-in cushion. On this basis, this report changed the §9 first choice from MPC to VLO.

⚠️ The following are three weak spots already listed in the first draft (found in QA, accepted):

• The cross-section has not yet differentiated: the complete ranking of today's pre-market declines is MPC −1.98% > COP −1.62% > VLO −1.40% = PSX −1.40% > DVN −1.15% > OXY −1.14% > XOM −1.05% > CVX −0.77%. COP is an E&P, not a refiner, yet it fell more than VLO/PSX. The first draft's ranking omitted COP, which was selective evidence. The honest conclusion is: "refiners are falling more than integrated oil & gas" holds, but "refiners are falling more than all energy stocks" does not, and a 0.3–0.6pp gap is still within the noise band. The real evidence is the crack-spread table above and the cumulative gain differential since 8/20 (+18.6% vs −1.6%), not today's 1pp pre-market ranking.
• Pre-market volumes are all thin: PSX about 14,000 shares, COP about 17,000 shares, MPC about 32,000 shares, VLO about 39,000 shares — all equal to or below CIEN (52,000 shares), which this report judged "too thin". By the same yardstick, these pre-market quotes are likewise insufficient as ranking evidence and can only serve as a directional hint. The first draft applied this threshold only to the long side, not the short side; corrected here.
• Whether the struck Saudi facilities included refineries (Ras Tanura, Yanbu SAMREF) is a link for which this report obtained no primary-source confirmation; the report cited in item 1 of Section 1 describes crude export and loading disruption, which is not the same mechanism as "refining capacity shutdown". The transmission chain "refinery shutdown → product shortage" is therefore marked unverified, and this section's conclusion is revised to rest directly on the crack-spread readings, not on that chain.

---

7. Within-Theme Ranking

Theme 1: Crypto-related stocks

Theme 2: Semiconductors

Theme 3: Optical communications

Theme 4: AI power and neocloud

---

8. Opening Verification Signals

⚠️ Primary reminder: today's list has only one variable

This is the risk that most needs emphasizing in this report. Every single name on today's long side — semiconductors, optical modules, crypto stocks, neocloud, AI power — traces its gain back to the same chain: oil falls → inflation pressure eases → the hiking path is repriced → yields decline → duration assets are revalued.

The evidence is that sector gains form a perfect "beta ranking": SMH +2.02% > ARKK +1.89% > XLK +1.43% > QQQ +1.14% > IWM +0.82% > SPY +0.72%, while the only decliner, XLE −0.93%, sits precisely at the other end of that chain.

This means the diversification is fake. Holding NVDA, IREN, COIN and CIEN together looks like four themes but is actually 4x exposure to the same macro bet. If oil reverses on a single geopolitical headline, they will draw down together. The verification points below are therefore deliberately designed to test different variables, rather than being six copies of the same one.

The crypto side is especially obvious: MSTR +6.80%, COIN +5.69% and HOOD +5.17% are not three independent pieces of evidence but three copies of the one event "BTC +4.77% overnight"; last Friday the same three were up +16.39% / +11.66% / +9.12% together. Listing them side by side on a watchlist manufactures a false sense of mutual corroboration.

⚠️ Second reminder: this sector fell across the board on earnings days

Verification against primary SEC filings produced a three-sample piece of independent evidence, and it is harder than any narrative:

All three beat, all three fell sharply. This is not "bad results" but a sector that had already been repriced before earnings — by the time the good news arrived, the buyers were gone.

The direct implication for today is: "good fundamentals" is not by itself a reason to buy today, because all three of these had good fundamentals on their respective earnings days and all fell. Today's real variable is oil and rates, not fundamentals. This is also the core basis on which I labeled INTC, AMD and CIEN — three of the top gainers — all as "watch only".

Verification points (by priority, each testing a different variable)

① Structural (highest priority) — tests "is this still only a handful of names rising" Last Friday's market breadth was alarm-level: Nasdaq +0.39%, Philadelphia Semiconductor +2.78%, but S&P equal weight −0.48% and Russell 2000 −0.50%. That is, the indices were carried by a tiny number of AI-weighted names while the "majority" of the market fell.

• Watch: whether RSP (S&P equal weight) and IWM can both turn green and hold today. If equal-weight/small-cap keeps underperforming the index in the first 30 minutes, today is still a narrow-ridge tape with no basis for broadening — this is a signal about the entire list, not about any one name in it.
• ⚠️ The first draft used "pre-market IWM +0.82%, slightly stronger than SPY +0.72%" as a positive preliminary signal; that conclusion has been withdrawn: a 10bp gap is within the pre-market ETF bid-ask spread and carries no information. Breadth cannot be judged pre-market; you must wait for the actual relative strength of RSP/IWM after 09:30.

② Commodity — tests the source of the entire logic chain

• Watch whether WTI can hold below $100 (Friday's close was 100.30, and it printed 97.43 at 08:12 ET this morning). This is the foundation of every long today. USO (−2.58%) and BNO (−2.08%) can serve as independent cross-verification.
• Falsification condition: if WTI recovers 100.30 (i.e. erases today's entire decline, especially if UNGA-related headlines appear), the entire logic of the long side fails on the spot, regardless of single-stock fundamentals.
• ⚠️ The first draft mistakenly wrote "hold below $97" here, while WTI's current price is already 97.43 — that condition was unsatisfied the moment it was written, an invalid test, now corrected to 100.

③ Rates — tests whether the market is betting against the Fed

• ⚠️ An evidence gap that must be confessed (found in QA, accepted): if today really were "a repricing of the hiking path", the front end should move the most. But the actual shape is 2Y −3.8bp < 30Y −3.9bp < 10Y −4.9bp — the front end moved the least. That looks more like an overall downshift in the term premium/duration than a repricing of the policy path. And the magnitude of a 4.9bp decline in the 10Y cannot by itself support SMH +2.02%.
• The report's statement of the causal chain should therefore be understood as: oil falls → inflation expectations and the term premium fall back → duration assets revalue + risk appetite returns. The link "the hiking path is repriced" has no direct witness in this report — to confirm it you would need the change in the fed funds futures-implied probability of a December hike from 9/18 → today, and this report did not obtain that reading.
• Note one key fact: the Fed's dot plot shows 16 of 18 members expect one more hike, while the market today is trading easier financial conditions. There is tension between the market and the dot plot.
• Watch Goolsbee's 10:30 ET remarks and the 12:30 ET Chicago Fed National Activity Index. There are 10 Fed speaker appearances this week, and each is headline risk. If hawkish language pushes the 2Y yield back up, the duration trade will reverse quickly.

④ Single-stock — tests "is it beta or is someone really buying"

• INTC: whether volume in the first 30 minutes sustains the abnormal pre-market level (4.59 million shares pre-market). If volume is heavy but the stock spikes and fades, that indicates profit-taking rather than new money entering — a negative signal for the entire semiconductor sector.
• Testing "is there any single-stock alpha today at all": after verification, this report found that there is almost no genuinely same-day single-stock catalyst today (IREN's was 9/14, INTC's was 9/16, crypto's were 9/18 and 9/16). A clean test is therefore: see whether there is any correlation between today's biggest gainers and their fundamental quality. If there is none at all (a pure beta sort), then every single-stock-level conclusion on this list should be de-weighted today, keeping only the macro-level judgments.

⑤ Gap-fill risk (gap-and-go vs gap-fill)

• The high-gap group (INTC +5.48%, MSTR +6.80%, CIEN +6.24%, CRCL +5.71%, COIN +5.69%) should be watched for whether they can hold above the opening price after the open. ⚠️ Note: the pre-market change is a cumulative quantity, and the 08:05 snapshot is not the 09:30 open, especially when pre-market volume is thin (CIEN only 52,000 shares pre-market, AMAT only 86,000) — these quotes are clearly less representative than INTC's 4.59 million shares.
• Historical note: the names above all saw single-day drawdowns on the order of −5% on 9/10 and 9/14; gap-fill has been the norm rather than the exception in this move.

⑥ Options/volatility

• VIX closed Friday at 14.81 (an absolute low), and printed 14.85 (+0.27%) at 08:16 ET this morning; October VIX futures are 17.75 (−0.29).
• Two details worth noting: (a) futures are in contango of nearly 3 points over spot, meaning the options market is pricing the coming month well above the current calm — consistent with "geopolitical headline risk has not gone away"; (b) index futures are up 0.7%–1.1% while spot VIX has not fallen in sync. In a healthy return of risk appetite, VIX usually falls alongside. This non-confirmation suggests today's buying looks more like short covering and duration revaluation than a substantive improvement in risk appetite.
• ⚠️ Correction here: the first draft wrote "no major earnings this week, so IV crush risk is very low"; that judgment has been withdrawn. This week does include COST (9/24 after the close), ACN and NKE, and those names and their sectors still face pre-earnings IV expansion and post-earnings IV crush. It is only that none of the names on this list report this week, so IV crush risk is limited for this list — these are two different statements, and the first draft conflated them.
• One pre-market anomaly worth noting: NKE traded 551,000 shares pre-market yet barely moved (+0.31%) — volume without price is usually position adjustment during earnings week, not a directional bet.

---

9. Final Conclusions

① The 5 names most worth watching today

② The 3 strongest themes today

③ Directions to avoid today

1. Refiners (VLO first, then PSX; MPC carries its own midstream buffer) — the 3-2-1 crack spread sits in the 99.78th percentile of 20 years (20-year median of only $16.80), P/B is in the 99.9th percentile of five years, and almost the entire gain since 8/20 stopped right at Friday's high. But you must know three things at the same time: ① the Saudi pipeline remains shut in and repairs may take until late October, so the supply recovery is not complete; ② the quarter-to-date Q3 average crack spread is 43% above Q2, so the Q3 results in late October will very likely be stronger; ③ consensus already implies the crack spread falling to ~$39 (−42%), so only a break below $39 creates a real expectation gap. This is a direction with very hard evidence but possibly the wrong clock.
2. Integrated oil & gas (XOM/CVX) — they are falling too, but they never rose since 8/20 (XOM −1.6%, CVX +1.8%), so there are no gains to give back; this is not the same situation as the refiners and the previous item's observation logic does not apply.
3. Names that are severely overbought or whose catalyst has expired (INTC/AMD/ARM/CIEN/MSTR/IREN) — especially INTC: the heaviest pre-market money in the field (about $530 million notional), paired with a piece of 9/16 stale news that the party involved has partly denied, with the valuation in the 98th percentile of 5 years.
4. Concept stocks with no revenue at scale (OKLO/SMR) — they rise fastest in a pure-beta tape and fall fastest in the drawdown.

④ Final one-sentence judgment

Today is a highly correlated return of risk appetite driven by the single variable "oil falls → inflation and the term premium fall back" — and almost no genuinely same-day single-stock catalyst can be found anywhere in the field: INTC's is from 9/16, IREN's from 9/14, and crypto's two regulatory positives from 9/18 and 9/16. On a pure-beta day like this, the only long argument still standing is not news but valuation: NVDA's TTM P/E sits in the 1.7th percentile of the past three years, paired with +106% revenue growth; and the only inverse argument still standing is a crack spread in the 99.78th percentile of 20 years — but its clock has very likely not arrived, because the quarter-to-date Q3 crack spread is still 43% above Q2. The biggest risk is that this entire list is essentially a bet on one variable.

---

Data-Pull and Verification Log for This Edition (internal)

Data channel status

• The Nasdaq official API (api.nasdaq.com/api/quote/.../info) was fully available this time, 102/103 successful (only HES failed), with isRealTime=True; it is the primary channel for this report's pre-market prices and pre-market volumes.
• The CNBC quote API was available, used for indices/futures/Treasuries/FX.
• The pre-market change baseline was back-computed name by name via last price − change amount and compared against Nasdaq secondaryData 9/18 closing prices: everything matched exactly except META (back-computed prior close 665.225 vs secondaryData 665.75, a 0.08% difference, suspected to be an after-hours print; the back-computed value was used).

Bad fields identified and avoided (still to be checked next time)

1. CNBC's .DXY and .VIX were frozen values at 08:00 ET: both had change=UNCH and last == previous_day_closing, and .VIX9D's last_time was stuck at 09-18. → The dollar was switched to real-time cross-verification via EUR=/JPY=. .VIX revived at 08:16 ET (14.85, prev 14.81, with open/high/low), which conversely proves 14.81 was indeed the 9/18 close; .DXY was still frozen at finalization, so the body does not cite its value. Lesson: a UNCH in the early pre-market is not necessarily dead data — it may simply be that the day's first print has not occurred yet, so re-check before finalization rather than discarding it outright.
2. CNBC's US2Y change_pct field is broken again: change=-0.038 (correct) but change_pct=+0.0742% (contradicting its own last/prev; the correct value is about −0.80%). Consistent with past records, this field is unusable; always back-compute from last/prev.
3. The direction of the yield decline was independently cross-confirmed with TLT +0.70%, without relying on the CNBC Treasury fields alone. 3b. Friday's breadth was independently re-checked with ETFs: RSP −0.48% (matching S&P equal weight −0.48% exactly), IWM −0.47% (matching RUT −0.50%), QQQ +0.63% (matching NDX +0.67%), SMH +2.21%. The narrow-ridge conclusion holds.
   • ⚠️ One apparent conflict that is actually an ex-dividend: SPY was −0.12% on 9/18 while the SPX index was +0.17%, a gap of 0.29pp. 9/18 was the third Friday of September, SPY's quarterly ex-dividend date, and a dividend of about 0.25% exactly bridges that gap (same for DIA, −0.48% vs DJI −0.18%). This is not a data error; do not chase it as a source conflict.
4. The crude direction was cross-confirmed with USO −2.58% / BNO −2.08% against WTI futures −2.72%; all three agree.

Secondary-source errors intercepted (3, all of which would have changed conclusions)

1. Micron's earnings date: multiple secondary sources (TipRanks, Investing.com pages and a TipRanks article) said "September 23 after the close", which is actually Micron's fiscal 2025 date. Micron IR's 8/26 press release confirms September 30, 2026. Without verification, this would have conjured up "the biggest event of the week" out of thin air.
2. Friday's index closes: TheStreet's headline was dated 9/18 but the body gave "Nasdaq +439.87 points / +1.7% to close at 26,418.3" — that number is the 9/17 close. 9/18 was actually 26,522.545 (+0.39%). The +0.39% later in the same article contradicts it. All figures were switched to primary CNBC quotes.
3. Micron "Q4 free cash flow above $30 billion": that quarter's revenue guidance is only $11.1–11.3 billion, so FCF cannot exceed revenue; the magnitude is clearly wrong and it was discarded and never written into the body.

risk-auditor QA results: 22 items raised, 17 accepted, 0 rebutted, 3 not completed for lack of time

The 6 items that changed conclusions (all incorporated into the body):

1. "No major earnings this week" was wrong — COST on 9/24 after the close, ACN and NKE are all this week. I verified only one ticker (Micron) and extrapolated to the whole week, a textbook case of "check one → assert the whole week does not exist". "IV crush risk is very low" was withdrawn and rewritten along with it.
2. The closing ranking in §6 silently deleted the only counterexample, COP (−1.62%, an E&P that fell more than two refiners) — that is selective evidence, and it happened in the very passage I called "the hardest evidence in this report". It has been put back and the strength of that passage's conclusion reduced.
3. NVDA's/IREN's verification points held in both directions ("rose less" was used both as a reason to buy and as a falsification condition). Both have been unified to look only at the 09:30–16:00 open→close, with an explicit note that pre-market relative strength does not count.
4. MPC's verification point tested the wrong variable — it was written as WTI, while my own argument says the driver is the crack spread; the most likely money-losing path, "WTI falls + crack widens", would trigger no test at all. It has been changed to watch the 3-2-1 crack spread, and the missing primary crack-spread readings were added (this was the most valuable QA item: the original short thesis section had zero data and was entirely inferred backwards from share prices).
5. The thin-volume threshold was applied only to the long side — on the short side PSX at 14,000 shares, COP 17,000, MPC 32,000 and VLO 39,000 are all below CIEN (52,000), which I judged "too thin", and their ranking was precisely the closing evidence of §6. It has been annotated by the same yardstick.
6. NVDA's "expectation gap 9 pts" counted price position a third time (trading characteristics 10 pts and a risk deduction of only −3 had each counted it once). After revising it down to 4 pts NVDA fell from #1 to #2 and IREN rose to #1 at the time — but subsequent primary verification then overturned that IREN item; see below. This is the same error as the "one variable disguised as many" that I criticize in §8, only occurring inside the model.

Other accepted items: the two crypto catalysts (SEC 9/18, the bill 9/16) were both stale news yet graded S (now cut to A+, the same yardstick as INTC); the WTI test changed from 97 to 100 (the original test was unsatisfied the moment it was written); COIN's unattributed rating item was deleted; CIEN's source URL predated the event by two days and was replaced with the SEC primary source; the fact that the yield shape contradicts the causal chain was confessed; the 10bp IWM/SPY conclusion was withdrawn; compliance wording (exit immediately / not suitable for a large position / reduce position size / cleanest path) was all changed to observational language; §4's scope of applicability and §3's 102→10 manual screening basis were disclosed; 9/12 was a Saturday (pricing on 9/14); INTC was 10 trading sessions, not 11 (9/7 Labor Day); BTC was 6 days, not 4.

🔴 Third verification round after finalization (NVDA/IREN/MPC/VLO) overturned 3 conclusions — this round was the most expensive and must be remembered

1. IREN's JPM two-notch upgrade was 9/14, and I wrote it as "today". This is the most serious error of the session: it led me to rank it overall #1 and label it "priority deep-dive", while in the same report I was assigning INTC a full risk deduction precisely for "treating 9/16 stale news as a same-day catalyst". The test was applied to others but not to myself. It has been cut to #10, total score 69→30, label changed to watch only. Lesson: for any "rating upgrade today", the URL date of the instant alert must be checked; never trust a Monday/"today" in a search summary.
2. IREN's third-party share-count field was 43% stale (terminal 224.5 million shares vs the 10-K cover's 394.1 million), so the real P/S is 26x, not 14.8x. That field was even lower than its own FY26 diluted weighted share count (316.1 million) — this consistency check could have caught the problem on the spot, and must be done for highly dilutive names from now on.
3. The clock on "Saudi recovery → crack spread normalization" is out of sync. The pipeline has been shut in since 9/10 with repairs taking up to six weeks; meanwhile the quarter-to-date Q3 average crack spread of $67.37 is 43% above Q2's $47.11, so the Q3 results in late October will very likely set a record. "The stock should fall" and "fundamentals are deteriorating" are not the same thing, and I treated them as evidence for each other. Also, consensus already implies a crack spread falling to about $39, so "will it fall" is not an expectation gap — "will it fall below 39" is.

Still unresolved, priority next time (3 items)

• NVDA fundamental verification incomplete now completed: TTM GAAP P/E of 28.1x sits in the 1.7th percentile of the past 3 years, the only hard-evidence fulcrum on this report's long side, which is why NVDA returned to #1. Two bases worth remembering long term were found along the way: (a) NVDA's GAAP EPS is higher than non-GAAP (investment gains are included in GAAP but excluded from non-GAAP), with 12.8% water in the TTM; (b) the commonly cited "forward P/E of 18.4x" most likely corresponds to FY2028 rather than FY2027 (self-computed FY27 is about 23.6x), and putting it beside 28.1x makes the valuation look one notch cheaper.
• "Repricing of the hiking path" has no direct witness. It requires the change in the fed funds futures-implied probability of a December hike from 9/18 → today. The existing yield shape (2Y moving least) actually points to a different explanation, confessed in §8-③.
• INTC's +5.48% today (about $530 million notional) still has no same-day catalyst located. Three rounds of search + no new EDGAR filings. QA note: if the restructuring-charge disclosure behind that Benzinga "Intel stock gaining Friday" item 11 actually occurred on 9/18, then the heaviest judgment in the whole report — "no catalyst found + risk deduction −15" — would need rewriting. The original article's date could not be traced before finalization.
• Secondary sources say CIEN has a Barclays target of 548 and a Northland target of 550 (+57% versus the current price); the values are anomalous and could not be confirmed against a primary source, so they were not written into the body.
• Primary confirmation from Saudi Aramco IR is still missing. All existing evidence is media-grade and self-contradictory (9/16 "may restore half of capacity within days" vs 9/18 "will cut off all crude to Europe next month"). This is the fulcrum of the refiner observation, yet it is the weakest link in this report's evidence.
• IREN's Q1 FY27 (ending 2026-09-30) segment revenue is the first verifiable reading on the "AI pivot" (whether Microsoft lease revenue enters the income statement for the first time), along with the delivery progress of Horizon 2-4.

Methodological summary for this edition (worth making a habit) Three rounds of verification overturned 6 of my judgments that changed conclusions (COIN fundamentals, HOOD risk direction, NVDA double counting, IREN catalyst date, IREN share-count basis, refiner clock), of which 4 came from "date/basis" rather than "fact right or wrong". The common pattern is: the number itself is not wrong; what is wrong is which day it belongs to, which basis it is on, and which baseline it is relative to. Next time, start by hardening these four questions into a checklist — "What day is this number from?" "What is the denominator of this ratio?" "Has this catalyst already been priced?" "Have I applied the same yardstick to myself as well?"

• META's −2.43% on Friday has a questionable attribution: secondary sources attribute it to Christopher Cox selling 20,000 shares (about $13.5 million) on 9/15. At META's market cap, a $13.5 million insider sale is unconvincing as an explanation for a 2.43% decline; 9/18 was quadruple witching (27.6 million shares traded, 1.34x the 9/8–9/17 average of 20.6 million shares), and mechanical rebalancing is the more likely cause. The body does not adopt that attribution and treats META's pre-market +2.38% merely as a beta rebound.

---

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