US Pre-Market Brief | 2026-09-15 (ET) Tuesday

Coverage window: 2026-09-14 16:00 ET regular-session close → 2026-09-15 08:15 ET. Covers Monday's after-hours, the overnight Asia/Europe sessions, and this morning's pre-market. Methodology notes:

• Single-stock pre-market prices and pre-market volumes come from the Nasdaq official quote API (read at 09-15 07:49–08:13 ET), benchmarked to the 09-14 16:00 ET close, with pre-market price − net change = previous close reverse-computed and verified name by name.
• Indices/futures/commodities/Treasury yields come from the CNBC quote API (read at 09-15 07:50–08:00 ET).
• 52-week highs and lows were cross-checked and found consistent across two independent sources, Nasdaq and CNBC; distance-from-high percentages are benchmarked to the 09-14 close.
• Sector statistics are my own, computed from name-by-name pulls of 759 large caps (market cap ≥ $15 billion), not third-party sector metrics.
• ⚠️ Pre-market quotes are cumulative, not point-in-time facts: the same name can read differently at 07:50 and at 08:13, so every pre-market percentage below is tagged with its read time.
• 🔴 The data in this brief is valid through 08:30 ET. The September Empire State Manufacturing Index is released at 08:30 ET (after my main readings), so from that point on none of the pre-market percentages here include its reaction, and pre-market volume keeps accumulating. If you are reading this after 08:30, treat every pre-market figure as a historical snapshot rather than the current state.

---

0. Today in One Sentence

1. Today's pre-market is not a directional trade; it is a "Monday reversal." S&P futures −0.17%, Nasdaq futures −0.12%, SPY −0.14% — at the index level almost nothing moved; but the cross-section underneath the index is swapping places wholesale. I pulled pre-market quotes name by name for 759 large caps: 574 had live pre-market trades, 570 passed the "pre-market price − net change = previous close" reverse-computation check, and of those 562 also had a Monday return; across those 562 names (full sample), the correlation between "Monday's return" and "this morning's pre-market return" is −0.55. ⚠️ One selection bias has to be stated at the same time: if you look only at the 66 names with pre-market notional ≥ $10 million, the correlation rises to −0.79 (−0.77 for the 113 names ≥ $3 million). But that threshold is itself positively correlated with "fell hard on Monday" (you only get pre-market volume if you fell hard), so this is conditioning on a variable correlated with both x and y, which systematically inflates |r|. Two further technical caveats: the pre-market return and the Monday return share the 09-14 close as their benchmark, and bid-ask noise in that close mechanically manufactures negative correlation on its own; and r = −0.55 implies R² = 0.30, so 70% of the variance is unexplained. So this brief uses −0.55 throughout, and a claim like "almost no exceptions" does not hold — the reversal is the single strongest structure of the day, but it is not the only structure, and it cannot on its own serve as causal evidence of mean reversion.

2. The strongest catalyst is macro, not single-stock: the 10-year Treasury yield has moved above 5%, the highest since 2007. My 08:00 ET reading was 4.998%, and by 08:21 ET it was 5.012% (+5.1bp on the day), with an intraday high of 5.041% and 30Y at 5.366%. Yields are moving up into the open, not backing off. The FOMC's two-day meeting starts today, with the decision tomorrow (9/16) at 14:00 ET, and the market is pricing roughly a 92% probability of a 25bp hike to 3.75%–4.00% — which would be the first hike since July 2023. The real counterparty to every position today is two o'clock tomorrow afternoon.

3. What set off Monday's riot was an essay, not an earnings report. Anthropic CEO Dario Amodei published "We Must Pace the Frontier" on Saturday, calling for a slowdown in frontier AI development; Altman and Musk publicly echoed it, and Altman also said OpenAI will not IPO in 2026. The result: the Philadelphia Semiconductor Index fell nearly 6% on Monday (its worst since early July), while pure-play cybersecurity names exploded higher (CRWD +13.85%, ZS +16.52%, PANW +13.09%). This morning both legs are walking back at once.

4. The two legs are retracing at asymmetric speeds, but it must be said in absolute magnitude, not in ratios. Grouped into five buckets by Monday's return (all five shown, none cherry-picked):

   Bucket	n	Monday mean	This morning's pre-market mean (absolute)	"Give-back ratio"
   Q1	22	−9.02%	+1.36%	15%
   Q2	22	−4.69%	+0.62%	13%
   Q3	22	−0.85%	+0.16%	18%
   Q4	22	+1.69%	−0.70%	42%
   Q5	25	+6.86%	−2.06%	30%

   Table scrolls horizontally

   ⚠️ The first draft argued from "gave back 30–42%," and that was wrong: the smaller the denominator, the more inflated the ratio, and the 42% for Q4 corresponds to an absolute move of only −0.70%, the smallest of the three (on the full 562-name sample, the bucket with a Monday mean of −0.03% produces a ratio of 386%, a pure denominator artifact). The correct statement, in absolute magnitude: the fading leg retraced −2.06% this morning, larger than the +1.36% repair in the reversal leg. The direction of the conclusion is unchanged but its strength must be downgraded — the market's confidence in "an AI slowdown is good for cybersecurity" really is weaker than in "an AI slowdown is bad for chips," but that is a 0.7 percentage point gap, not a rout.

5. The one major single-stock catalyst today that is unrelated to the "Monday reversal" and confirmed by a primary source: VERA. Vera Therapeutics published the two-year final efficacy analysis of ORIGIN 3 on its corporate IR page this morning; TRUTAKNA (atacicept-vymj) met all pre-specified endpoints. Pre-market $39.10, +14.83% (08:12 ET), notional turnover $26.5 million — enough volume to support the move, and it is the only name on the entire list that does not depend on the variable "how much did it fall on Monday."

---

1. News Overview

---

2. Strongest Themes, Descending

---

3. Overall Single-Stock Strength Table

Pre-market data was all read at 09-15 07:49–08:13 ET; "distance from the 52-week high" is benchmarked to the 09-14 close (cross-checked across Nasdaq + CNBC). Bullish section

The conclusion column in this table follows §7 "Within-Theme Ranking" as the final word; totals are the sum of the §4 components (out of 95), re-ranked after QC corrections.

| 1 | VERA | Vera Therapeutics | Biopharma | Bullish | S | 65 | ORIGIN 3 two-year final analysis met all endpoints | Very high (own primary disclosure) | Small cap ($2.8 billion), single asset, no depth of moat | High, unpriced before the data | +14.83%, $26.5 million (08:12) | Gap-fill; single asset | Priority deep-dive | | 2 | AMAT | 应用材料 (Applied Materials) | Semicap | Bullish | A | 53 | Bounce after −7.07% Monday | Medium (sector beta) | Revenue +24.8% and accelerating sequentially; 12↑/0↓ consensus revisions over the past four weeks; OCF/net income 0.91 | Medium; ⚠️ PE/PS/PB — all three multiples are above any fiscal year-end from FY21–25 | +1.95%, $21.2 million (08:13) | Single variable; FOMC; valuation already above its five-year range | Watch closely | | 3 | ASML | 阿斯麦 (ASML) | Semicap | Bullish | A | 52 | Led the bounce after −7.25% Monday, no single-stock news found | Medium (pure beta) | Sole global EUV supplier, deepest moat; net cash +€5.597 billion | Low; ⚠️ Forward PE 35x depends on an H2 acceleration covered by only 2–3 analysts | +3.41%, $139.8 million (08:12) | No single-stock catalyst, pure sentiment; PE above every year from FY21–25 | Watch closely | | 4 | LRCX | 泛林 (Lam Research) | Semicap | Bullish | A | 52 | Bounce after −8.29% Monday | Medium | Operating margin 37.4%, the highest among equipment names; ROE 58.3% | Medium; ⚠️ OCF/net income 0.81 — profit is rising but operating cash flow has not kept up | +2.23%, $16.2 million (08:12) | Same as above; working-capital build to be verified | Watch closely | | 5 | CRDO | Credo | AI interconnect | Bullish | A | 52 (↑) | Bounce after −7.89% Monday | Medium | Net cash $738 million, total debt only $26 million — the cleanest in the group; revenue +114.7%, essentially organic | EV/S in only the 17% percentile of the past three years, the lowest in the group | +1.57%, $22.1 million | Extremely volatile; no EPS guidance from the company; acquisition target unnamed | Watch closely (↑ upgraded) | | 6 | MU | 美光 (Micron) | Memory | Bullish | A | 51 (↓) | Bounce after −5.25% Monday, second-largest pre-market notional in the market | Medium | Revenue +345.7%, gross margin 84.6%; OCF/net income 1.02, the cleanest earnings quality in the group | ⚠️ Forward PE 5.8x is extremely low while PB 10.36 is about 4x its historical peak — a textbook cycle-top reading | +1.26%, $439.6 million (08:13) | FQ4 prints 9/30, resetting the whole valuation baseline within two weeks; big volume, small move | Watch only (↓, aligned with §7) | | 7 | LLY | 礼来 (Eli Lilly) | Pharma defense | Bullish | B+ | 50 | +2.02% Monday, another +1.56% this morning, not on the reversal leg | Low (no news) | GLP-1 duopoly, very deep moat | Low | +1.56%, $25.7 million (08:12) | No catalyst, riding risk-off flows | Watch closely | | 8 | LITE | Lumentum | Optical | Bullish | B+ | 47 (↓) | Bounce after −9.92% Monday | Medium | Revenue +109.3%, non-GAAP EPS $3.23; post-conversion the balance sheet flips to net cash $1.067 billion | EV/S at the 100% percentile over both five and three years, the most expensive ever | +1.80%, $67.2 million (08:12) | Valuation has broken above its historical ceiling; share count +23.6% in one quarter | Watch only (↓ downgraded) | | 9 | INTC | 英特尔 (Intel) | Semiconductors | Bullish | B+ | 62 | Bounce after −5.59% Monday, pre-market notional $193 million | Medium | Turnaround incomplete, weak earnings quality | Medium, −31.7% from high | +1.96%, $192.7 million (08:13) | The weakest fundamentals of the group | Watch only | | 10 | MRVL | Marvell | AI custom silicon | Bullish | C (↓) | 40 (↓↓) | Bounce after −7.32% Monday | Medium | ⚠️ TTM GAAP profit includes a one-off pre-tax gain of $1.848 billion in the quarter ended 2025-10-31; excluding it, recurring PE is about 182x (72x on the face); operating margin 16.7%, the lowest in the group | ⚠️ The only name in the group whose consensus is being cut: 2↑/7↓ over the past four weeks; and two consecutive misses (−3.3%, −4.6%) | +1.55%, $63.5 million (08:13) | Both PS and PB are above the entire FY22–26 range | Avoid (↓↓ downgraded) | | 11 | GLW | 康宁 (Corning) | Optical fiber | Bullish | B+ | 61 | −13.70% Monday, the weakest name in the entire S&P, +1.29% this morning | Medium | TTM operating cash flow/net income 2.06x, the best cash conversion in the group; ROE 19.1%, the only respectable GAAP return in the group | Medium, −47.2% from high | +1.29%, $21.8 million (08:12) | Net debt $6.875 billion, the highest in the group; core and GAAP definitions differ | Watch only | | 12 | ARM | Arm | IP licensing | Bullish | B | 57 | Bounce after −9.74% Monday | Medium | IP monopoly but an expensive valuation | Medium, −47.2% from high | +1.47%, $17.2 million | Still expensive | Watch only | | 13 | KLAC | 科天 (KLA) | Semiconductor metrology | Bullish | B | 57 | Bounce after −6.39% Monday | Medium | Near-monopoly in metrology | Medium, −45.0% from high | +1.66%, $5.4 million | Thin pre-market volume | Watch only | | 14 | COHR | Coherent | Optical | Bullish | B | 55 (↓↓) | Bounce after −12.73% Monday (third-worst in the market) | Medium | ⚠️ TTM operating cash flow of only $79 million against net income of $770 million (0.10x); free cash flow −$1.023 billion; inventory +79% year over year while revenue rose only +22.5% | EV/S at the 95% five-year percentile | +2.62%, $39.2 million (08:12) | Severe divergence between the income statement and the cash flow statement | Watch only (↓↓ materially downgraded) | | 15 | GEV | GE Vernova | AI power | Bullish | B | 55 | Bounce after −8.62% Monday | Medium | Grid equipment cycle | Medium, −25.8% from high | +1.45%, $19.9 million | Same source as AI capex | Watch only | | 16 | NVDA | 英伟达 (NVIDIA) | AI compute | Bullish | B | 54 | −3.36% Monday, +0.70% this morning, largest pre-market notional in the market at $660 million | Low (fell the least) | The undisputed leader | Low, only −10.8% from high | +0.70%, $659.6 million (08:12) | Least room to repair | Watch only | | 17 | NOK | 诺基亚 (Nokia) | Network equipment | Bullish | C | 48 | −13.30% Monday, +2.89% this morning | Low | Monday's drop includes company-specific bad news | Low | +2.89%, $39.7 million | Impure attribution, weak base for a bounce | Avoid |

Bearish/fading section (see §6)

---

4. Single-Stock Scoring Model (positive components out of 95, risk deduction 0 to −15)

⚠️ Correction notice: in the first draft of this table the "total" was set by tier first and the components back-filled, so all six columns failed to reconcile with the sum of their components (errors of +15 to +20 points), and the heading was also mis-written as "out of 100" (the seven positive components actually cap at 95). After QC caught it, the totals were re-summed from the components and §3 was re-ranked accordingly. Every column below can be verified by hand.

Scoring note: after the correction, AMAT at 53 and ASML at 52 are still adjacent, but the supporting reasons changed entirely. The original draft said "AMAT wins because it is −42.6% below its 52-week high and therefore has more room to repair" — that reason has been falsified; see the red box below. AMAT's lead now rests on "revenue accelerating sequentially + 12 upgrades and 0 downgrades in consensus over the past four weeks + OCF/net income 0.91"; ASML is dragged down by "a forward PE of 35x that depends entirely on an H2 slope covered by just 2–3 analysts and requiring Q3 to rise +43% sequentially."

🔴 A ranking error that must be publicly corrected: "the deeper below the 52-week high, the more room to repair" has no predictive power in today's cross-section, and the sign is if anything reversed. Recomputing on the 14 reversal-leg names in §3, the correlation between "distance from the 52-week high" and "today's pre-market gain" is just −0.023 (roughly zero). Name by name it is worse: the four with the deepest drawdowns (CRDO −51.4%, GLW −47.2%, ARM −47.2%, KLAC −45.0%) are up only 1.29%–1.66% pre-market, the weakest in the group; while the shallowest drawdown, ASML (−21.2%), is +3.41% pre-market, the best in the market. The semiconductor sub-agent reached the same conclusion independently from another direction: all 7 names peaked between 2026-06-04 and 06-30 (4 of them on exactly 6/30) and have retraced 20.8%–45.0% since, with drawdown depth essentially unrelated to — if anything slightly inverse to — fundamental strength — KLAC, with the lowest growth (+15.2%), has the deepest drawdown (−45.0%), while MU, with the highest growth (+345.7%), has one of the shallowest (−26.4%). So this brief keeps "distance from the 52-week high" as a description of position/risk structure (used to show that CRWD is at highs and semis are in a hole), but has removed it from the ranking variables.

---

5. Detailed Analysis of Top Names

① VERA — Vera Therapeutics | Priority deep-dive | Total 85

• Related news: before the 09-15 open, the company's IR page published "TRUTAKNA™ (atacicept-vymj) Stabilized eGFR and Halted Two-Year Progression of IgA Nephropathy in the ORIGIN 3 Final Efficacy Analysis." The company says it met all pre-specified endpoints, with safety comparable to placebo, supporting atacicept's pursuit of full approval in adult IgA nephropathy; it also calls the drug "the first and only dual BAFF and APRIL inhibitor approved by the FDA." Source: VERA IR⁴ (primary disclosure, not a media relay).
• Catalyst logic: this is a regulatory-path catalyst, not a current-period earnings catalyst. TRUTAKNA received FDA accelerated approval on 2026-07-08; the two-year eGFR endpoint in ORIGIN 3 is the key dataset for moving from accelerated approval to full approval. The valuation impact lies in repricing "an accelerated approval that could be withdrawn" into "very likely to convert to full approval" — what falls is the regulatory risk premium inside the discount rate, not an upgrade to near-term revenue.
• Theme and stage: launch day. The data went public for the first time this morning, and the pre-market is the first pricing window.
• Fundamental verification: market cap about $2.82 billion (Nasdaq, 09-15). −39.3% below its 52-week high ($56.05) (based on the 09-14 close of $34.05); 52-week low $22.72. This is a single-asset biotech without meaningful revenue scale, so a P/E framework does not apply; cash runway and follow-on financing are the main variables (this brief did not obtain the latest cash balance; left blank, no speculation).
• Moat/market position: being the first and only FDA-approved dual BAFF/APRIL mechanism in IgA nephropathy is a temporary barrier, but with a single indication and a single asset, resilience is weak.
• Pre-market and technicals: $39.10, +14.83% (08:12 ET), pre-market notional turnover $26.5 million — ample volume for a company with a $2.8 billion market cap, not a thin-volume fake breakout. The current price is still about 30% below the 52-week high.
• Final judgment: the only logic on the entire list that has both a primary source and no dependence on the variable "how much did it fall on Monday." But it is already +14.8% pre-market, and the iron rules say do not chase; the real thing to watch is whether it holds the pre-market price in the first 30 minutes rather than spiking and fading.

2应用材料AMATWatch closely | Total 73 · Applied Materials

• Related news: no single-stock news. −7.07% Monday, +1.95% ($21.2 million, 08:13 ET) this morning.
• Catalyst logic: purely a sentiment retracement of Monday's AI-slowdown panic. To be clear: nothing in Amodei's essay asks anyone to buy fewer tools or fewer chips — this was a narrative-driven valuation swing, not a change in orders or guidance.
• Theme and stage: the first snapback after the fade, a very short stage. But to be clear: this is not day two of a fall from the highs — all 7 semiconductor names peaked between 2026-06-04 and 06-30 (4 of them on exactly 6/30) and have already been retracing for two and a half months. Monday was just one big red candle on that downward path.
• Fundamental verification (checked against SEC/Yahoo time-series data): FQ3 FY2026 (ended 2026-07-31) revenue $9,115M, +24.8% year over year, +15.2% sequentially, with growth accelerating quarter by quarter; gross margin 50.3%, operating margin 33.7%; TTM operating cash flow $8,396M, free cash flow $5,623M (OCF/net income 0.91); net cash +$1.887 billion (thin); ROE 36.2%. Consensus revisions over the past four weeks: 12 up / 0 down, the most unanimous revision direction in the group.
• ⚠️ But one piece of hard evidence directly undercuts my original reason for recommending it: AMAT's PE (36.6), PS (10.92) and PB (13.14) are all above every fiscal year-end from FY21 to FY25, with PB roughly 30% above the top of its historical range. In other words, "−42.6% below the 52-week high" does not mean "cheap" — it is only cheap relative to the June peak; relative to its own five-year history it is still in expensive territory.
• Moat/market position: platform leader in deposition/etch/ion implantation, forming the big three of equipment together with LRCX and KLAC.
• Pre-market and technicals: +1.95%, recovering only about 28% of Monday's decline.
• Final judgment: keep it at watch closely, but swap out the reason. What supports it is "accelerating revenue + 12:0 consensus upgrades + solid cash flow," not "it fell a lot so it must bounce" — the latter has been falsified by the cross-sectional evidence in §5④ below. It eats the same variable as LRCX/ASML/MU/KLAC, so holding several at once is not diversification.

3阿斯麦ASML / ASMLWatch closely | Total 72

• Related news: no single-stock news today. −7.25% Monday, +3.41% ($139.8 million, 08:12 ET) this morning, the largest pre-market gain among all large caps.
• Catalyst logic: pure sector beta. Wording correction: the first draft said "there is no new company-level information behind today's +3.41%" — that is an assertion that something does not exist, and the evidence does not support it. The accurate statement is: as of 08:13 ET, the most recent ASML single-stock item I found was Morgan Stanley cutting its target from €1,930 to €1,700 on 9/8 (maintaining Overweight), with the next earnings report on 10/14; I did not find any ASML single-stock news today, but I cannot conclude from that that none exists.
• ⚠️ One item in the forward valuation must be flagged as unverified: ASML's forward PE of 35.0x rests on FY2026 full-year consensus of $44.97, while actual H1'26 adjusted EPS was only $17.18 — to hit it, H2 must deliver $27.79, i.e. Q3 +43% sequentially and Q4 +76% year over year, and the quarterly consensus sample is extremely thin (only 3 estimates for Q3 and 2 for Q4). That slope has not been confirmed by primary company guidance. If it does not hold, ASML's forward valuation becomes meaningfully more expensive.
• Theme and stage: same as AMAT, day one of the snapback.
• Fundamental verification: market cap about $626.9 billion; −21.2% below its 52-week high ($1,999.96) — the shallowest drawdown in the entire equipment chain.
• Moat/market position: the sole global supplier of EUV lithography, the deepest moat on this list.
• Pre-market and technicals: the largest gain (backed by $140 million of notional, not thin volume).
• Final judgment: the deepest moat, but "biggest gain + shallowest drawdown + no single-stock catalyst" stacked together means its pre-market gain is most likely a sentiment premium rather than repair. Lower priority than AMAT.

④ LRCX / ⑤ MU / ⑥ LITE / ⑦ COHR — the rest of the reversal leg

The four share the same logic, described together so as not to manufacture false independence:

Key observation one: MU's pre-market notional turnover was $440 million (second-largest in the market), yet it is up only +1.26%, the smallest in this group. A large amount of money bought a small move = real sellers are absorbing the bounce in the pre-market. By the same logic as "judging whether something is priced in is about the volume ratio, not the return," MU's volume-price combination says the memory bounce has the weakest bid behind it, not the strongest.

Key observation two (I downgraded two names after sub-agent verification):

• COHR is the only earnings-quality negative outlier in this group. The income statement looks great (operating profit +75.3%), but the cash flow statement is bleeding (FCF −$1.023 billion), and the gap can be verified line by line on the balance sheet — inventory up $1.143 billion, receivables up $379 million, capex $1.103 billion. Inventory/quarterly revenue rose from 0.94x to 1.26x. This cannot distinguish "stocking for locked-in orders" from "accumulated demand misjudgment," but for the position I originally gave it — 7th place with a "watch closely" tag — that uncertainty is too large, so it is downgraded to watch only.
• LITE's −$84.65 per-share loss is not deteriorating operations, it is a one-off non-cash loss from the convertible conversion, and the conversion made the balance sheet stronger (flipping to net cash). But its EV/S is now at the 100% percentile of all available data and has broken above the top of its historical range, while share count was diluted 23.6% in one quarter — "the huge loss is fake" and "the valuation is genuinely expensive" are both true at once, so it is downgraded too.
• In the other direction: CRDO is upgraded. Its declining GAAP gross margin looks like a price war, but on verification it comes entirely from $11 million of newly added acquisition intangible amortization this quarter (both the prior quarter and the year-ago quarter were $0), and non-GAAP gross margin barely moved; it also has the cleanest balance sheet in the group. But leave one item open: goodwill + intangibles jumped from $122 million to $1.365 billion this quarter, the acquisition target is unnamed, and future quarters' revenue base will include the acquired contribution.

8礼来LLYWatch closely | Total 65 · Eli Lilly

• Related news: none. But it is the only large cap that rose on Monday (+2.02%) and kept rising this morning (+1.56%, $25.7 million, 08:12 ET), and it is not on the reversal leg.
• Catalyst logic: with the 10Y touching 5% and a hike due tomorrow, large-cap pharma with certain cash flow, short duration and no link to AI capex is a natural shelter. XLV was +1.45% Monday, one of the few sectors to close green.
• Fundamental verification: market cap about $1.088 trillion; −11.9% below its 52-week high ($1,292.65).
• Final judgment: the gain is modest and unsexy, but it is the only name on this list that is a third variable, unrelated to both the "Monday reversal" and the "AI narrative." From a portfolio standpoint, its value is its lack of correlation.

⑨ CRWD — CrowdStrike | Short watch | Total 33

• Related news: +13.85% Monday, driven by Amodei's essay warning that swarms of autonomous AI agents could cause hundreds of billions of dollars of damage, feeding the narrative "worsening threat environment → expanding security spend." −1.89% ($42.7 million, 08:13 ET) this morning.
• Weakness in the catalyst logic: this was pure sentiment pricing — not a single cybersecurity company published orders, ARR or guidance on Monday. And one control group is telling: CIBR (the cybersecurity ETF) rose only about 4% Monday, while the pure plays were up 12–16% — meaning the buying was narrow, concentrated in the few names with the purest label.
• Position risk (the most important point here, and confirmed by fundamental verification): CRWD is only −1.7% below its 52-week high, PANW −6.3%. More important are the valuation percentiles — after the sub-agent recomputed them from 20 quarters of filings: CRWD's EV/S is at the 95% five-year / 100% three-year percentile (i.e. the most expensive in three years), and PANW's EV/S is at the 100% percentile over both five and three years, having broken above the top of its historical range (a historical range of $7.8–23.6, currently 26.50). This spike added another 13% on top of the most expensive valuation in its history.
• ⚠️ A cross-comparison error that must be corrected (the most important methodological reminder in this brief): the media placed PANW's "NGS ARR +63%" side by side with CRWD/ZS's "ARR +25%," which is using three different rulers:
  • PANW does not disclose total ARR at all; what it discloses is NGS ARR (next-generation security ARR, a subset that excludes hardware, legacy subscriptions and services);
  • most of that +63% comes from the CyberArk acquisition, corroborated by three independent pieces of evidence: goodwill + intangibles jumped from $8.180 billion to $29.185 billion in a single quarter, share count rose from 703 million to 813 million (+15.9%), and revenue growth jumped from +14.9% to +31.1% in exactly that quarter (after four quarters steady at +13.9% to 15.8%);
  • the strongest counter-evidence comes from PANW itself: FY27 NGS ARR growth guidance is only +22% to 23% — the drop from 63% to 22% is precisely the base effect disappearing once CyberArk has been consolidated for a full year. +63% cannot be extrapolated.
  • Control group: ZS voluntarily disclosed an organic figure (ARR +20% excluding Red Canary, versus the reported +25%), while PANW disclosed no ex-CyberArk figure. The same situation, one discloses and one does not — that itself is information.
• But the counter-evidence must be given, and this one changed my conclusion: ZS's EV/S is at the 15% five-year / 17% three-year percentile — the cheapest of the three relative to its own history, and it is −43.1% below its 52-week high. So "cybersecurity is all crowded" is wrong: what is crowded is CRWD and PANW, not ZS. The short watch applies only to the first two.
  • That said, ZS's low percentile does not equal cheap either: its FY27 ARR growth guidance falls to roughly +16.6% to 17.4% (FY26 reported 25%, organic 20%). A low valuation percentile and a downshift in growth are two sides of the same coin.
• Pre-market: 0 of 6 pure-play cybersecurity names are up pre-market, mean −1.85%. On Monday all 7 names traded at 1.9–2.2x their 10-day average volume, which says Monday's repricing was genuine turnover, not a thin-volume head fake.
• Final judgment: short watch, not a short recommendation. Sentiment themes can run two days, and tomorrow's FOMC is a two-way risk. If one must be chosen between the two, PANW's evidence chain is more complete than CRWD's — it satisfies all three of "the most expensive valuation ever" + "growth mostly from M&A" + "the company's own guidance for a sharp slowdown," whereas CRWD satisfies only the first (its net new ARR of +51% is accelerating, and acquisition intangible amortization is only $13.3 million per quarter, so it can essentially be treated as organic).

10Dave & Buster's | AvoidPLAYTotal 22 · pre-market volume too thin to judge the open

• Related news: Q2 results after the close on 09-14. Revenue $544.1 million (−2.4% year over year), below the $556.8 million expected; an adjusted loss of $0.27 per share against consensus of a $0.18 profit; comparable sales −2.9%; adjusted EBITDA $98.9 million, versus $129.8 million a year earlier; GAAP net loss $12.5 million (versus an $11.4 million profit a year earlier). Source: company release⁵.
• Catalyst logic: this is a triple stack of "a large current-period miss + comps turning negative + EBITDA shrinking 24% year over year", not a one-off factor.
• Pre-market: −13.46% ($1.2 million, 08:11 ET).
• Risk note: pre-market notional turnover was only $1.2 million, extremely thin. A −13.46% print on that volume cannot be treated as a reliable prediction of the opening price, and a stock already down 13% is not a good short to chase.

---

6. Bearish/Avoid List

---

7. Within-Theme Ranking

Theme one: AI hardware oversold bounce (short duration, at most 2 position slots)

Theme two: cybersecurity fade (short watch)

Same-ruler warning: in the "FY27 growth guidance" column above, PANW gives NGS ARR (a subset), CRWD gives total revenue and total ARR, and ZS gives total ARR. The three use different definitions and cannot be ranked directly. I put them side by side only to show "the direction each company guides for itself," not to compare absolute growth rates. The only comparable ruler is "total ARR year over year," and PANW does not provide that ruler.

Earnings-quality supplement (using plain FCF throughout, not company-defined adjusted FCF): TTM FCF margin PANW 35.8% > CRWD 29.8% > ZS 25.4%. But look at SBC as a share: ZS's stock-based compensation already equals 96% of free cash flow (CRWD 76%, PANW 43%) — net of SBC there is almost nothing left of ZS's free cash flow, which is the earnings-quality item most to be wary of among the three, and a necessary hedge against "ZS is the cheapest" above.

Theme three: energy/refining (no recommendation slots in this brief)

Why the branch with the most seductive narrative gets zero position slots. The sub-agent verified the six companies against primary SEC XBRL filings and produced three mutually independent pieces of evidence pointing to the same conclusion:

① Not one cent of the record crack spread has reached the financials yet. The latest reported fiscal quarter for all six is 2026 Q2 (April–June), filed in late July to early August. September's record diesel crack spread appears in none of the disclosed statements; the first statements containing it will not arrive until late October or early November. So "using the latest quarter to prove the crack spread has been realized" is misaligned in time.

② A low forward P/E is not cheapness, it is a denominator that is exploding. VLO's trailing P/E of 15.97 → FY26e 8.49 looks extremely cheap; but this is the standard shape of a cyclical in an earnings-acceleration phase and contains no "undervalued" information in itself. What actually carries information is the quarterly path of consensus: the market has explicitly built Q3'26 as the cycle peak (VLO quarterly adj EPS estimate of $17.70, +40.3% above actual Q2), then a Q4 cliff of −38% to −44%, and a further FY27 full-year decline of 18.6% to 30.3%.

③ All six have P/B in the 94%–99.8% five-year percentile, and the four pure refiners already trade 11.8%–18.9% above sell-side average targets. A record spread is information already written into models and prices, not information still waiting to be discovered.

Conclusion in one line: the market has priced in the "high" of the crack spread, but not its "long." The unpriced variable is duration, not height — and duration is exactly what cannot be confirmed in this morning's pre-market.

④ There is one more price-action cross-check, more intuitive than the three above. Yesterday (9/14) all six closed below their own opening prices (VLO open→close −3.12%, MPC −1.13%, PSX −2.30%, XOM −2.15%, CVX −2.23%, DINO −2.36%), and for VLO, MPC and CVX the 52-week high is exactly yesterday's intraday high — on a day when the diesel crack spread was at record levels, they pushed to 52-week highs and then faded to close lower, 6 out of 6.

⚠️ Thin-volume warning (applies both ways): pre-market volume in the pure refiners this morning is extremely thin (MPC just 579 shares, PSX 261, DINO 1,253). Pre-market quotes on that volume are not facts and cannot be used to judge direction — so when I wrote "barely moving pre-market" in §2/§3 above, that can only serve as weak evidence of "no incremental buying," not as evidence that "money is selling." I did not use it to support a bearish conclusion.

XOM/CVX are different in kind, do not lump them in with the pure refiners: both still trade below sell-side average targets, by 3.4% / 4.1%, their FY26→FY27 expected declines are only 8.1% / 16.5% (the pure refiners are 25%–30%), their refining exposure is diluted by upstream, and their beta to the crack spread is markedly lower. Putting all six in one "energy" basket and ranking them is using the wrong ruler.

Not obtained in this brief: the per-barrel refining margin for all six is left blank. That metric is non-GAAP, appears only in the EX-99.1 earnings release and on the call, and does not go into XBRL. The theoretical crack spread of $106 and a company's actual capture rate can differ widely (affected by crude differentials, product mix, hedges and unplanned outages), and until the actual capture rate is in hand you cannot derive refiner profits directly from a futures spread. Left blank here; no back-solved substitute number.

---

8. Opening Verification Signals

Pre-market (already happened, read 07:49–08:13 ET)

• Breadth of the reversal leg: of the 21 semiconductor/AI hardware names, 20/21 are up pre-market; optical/networking 7/7; AI power/cooling 4/4. In the other direction, cybersecurity 0/6 up, application software 0/12, crypto 0/5. Breadth says more than the volume of any single ETF (CIBR's pre-market notional was only about $0.98 million, too thin to be usable as sector evidence).
• The index barely moved: SPY −0.14%, QQQ −0.10%, S&P futures −0.17%. All the action is in the cross-section, not in the index.

Actual readings for rate-sensitive sectors (added later, read 08:23–08:31 ET) — this overturns the intuition that "a hike = good for banks"

The first draft made "rate shock" the S-grade top theme without giving it any price evidence of its own. After adding that, the results are as follows, listing only names whose pre-market notional is large enough to be credible:

🔴 Here is a piece of evidence that directly conflicts with theme 1 of this brief, and it has to be written down. Against a backdrop of "92% priced for a 25bp hike tomorrow," bank stocks fell on Monday, and not by a little: BAC −5.14%, GS −3.96%, MS −3.64%, C −1.90%, WFC −1.75%. If the market really treated the hike as a net-interest-margin positive for banks, that reaction makes no sense. The more plausible reading is the tail risk added in §6: with the 10Y above 5%, banks' unrealized securities losses (already $325.1 billion in 2026Q1) and deposit costs outweigh the benefit to net interest margin. In other words, at this level of rates a "hike" is a net negative for banks, not a net positive. But the strength must be labeled honestly: apart from JPM ($316 million), pre-market volume in bank stocks is generally insufficient this morning; the conclusion above rests mainly on Monday's regular-session trading, not on this morning's pre-market readings.

To confirm during the session

1. First 30 minutes: is the reversal leg gap-and-go or gap-fill. This is the only question that matters today. Note the base-rate problem: on a day when the index is flat and the event risk is tomorrow, the market-wide pass rate for "can a gap-up hold" is usually low, so do not misread a single name's gap-fill as bad news for that name.
2. Whether the SMH/IGV scissors gap converges. (⚠️ The first draft wrote "SMH and CIBR" here while quoting IGV's numbers — the subject and the evidence were not the same ruler; corrected to IGV, because software is the broad-based representative of the leg that rose on Monday; CIBR's pre-market notional was only about $0.98 million, usable in the regular session but not pre-market.) The two opened in opposite directions on Monday (IGV +5.04% vs SMH −4.75%, a single-day scissors gap of 9.79pp), and they are converging this morning (IGV −1.08% vs SMH +0.81%). If they re-widen after the open, Monday was not sentiment but a genuine reallocation of capital, and theme 2 of this brief is wrong.
3. Whether MU's volume-price divergence persists: $440 million of pre-market notional bought only +1.26%. If it trades heavy and stalls after the open, the memory bounce should be downgraded.
4. Whether VERA holds the pre-market price of $39.10, and whether the volume carries into the regular session.

Options and event sentiment

• VIX 17.04 (08:00 ET), below Monday's close of 17.10 and also below Monday's pre-market 17.61. With the 10Y touching 5% and a hike due tomorrow, the volatility market is clearly not pricing panic — which can be read either as "the market thinks the hike is fully priced" or as "too little protection is being bought."
• Today is day one of the FOMC, with no decision. The real event is tomorrow at 14:00 ET. Any trend today could be wiped out in one stroke tomorrow afternoon.

Risk checklist

• Single-variable risk (the biggest structural weakness in this brief): of the 17 names in the bullish section of §3, 14 are rising for the same reason — "they fell a lot on Monday." They look like 14 independent names but are really 14 copies of 1 variable. Holding several of them does not diversify risk.
• The 08:30 ET Empire State Manufacturing Index is released after my readings, with August at 20.6 (a more-than-four-year high). Its "prices paid" component may do more damage than the headline index under the current inflation narrative. None of the pre-market data in this brief includes a reaction to it.
• Reversal after a gap: VERA (+14.8%) and PLAY (−13.5%) have both gapped hard, and by the iron rules neither is to be chased.
• August retail sales come tomorrow (9/16) at 08:30 ET, the same day as the FOMC decision, not today (confirmed against the Census Bureau's primary calendar).
• A tail risk the market is ignoring: a 10Y above 5% will re-widen banks' unrealized securities losses. U.S. banking industry unrealized securities losses were $325.1 billion in 2026Q1, up about $19 billion from the prior quarter, ending four consecutive quarters of repair. This is not a trading theme today (KRE was only −0.12% pre-market), but it is the most fragile link on the "rates keep rising" path, and it would become a theme fast if any regional bank saw unusual deposit movement. Sources: OFR¹³, Forbes (2026-08-21)¹⁴.

---

9. Final Conclusions

① The 5 names most worth watching today

🔴 Correcting a piece of self-congratulation from the first draft. The draft said "I deliberately diversified across variables: only 2 slots for the reversal leg" — that was said but not done. The short watch on CRWD and the long watch on AMAT/LRCX eat the same "mean reversion in Monday's move" variable (the correlation itself covers both the up leg and the down leg); it is the mirror image of that variable, not a third variable; and LLY's "defense under high rates" shares a source with theme 1 in §9② (rates). The real diversification is 3:1:1, not 2:3 — three slots for mean reversion (AMAT / LRCX / CRWD), one for an independent clinical catalyst (VERA), one for rate defense (LLY). Strictly speaking, the only one fully independent of today's two big narratives is VERA. Readers should judge position correlation accordingly and not treat these as 5 independent ideas.

② The 3 strongest themes today

③ What to avoid today

1. The entire crypto chain (COIN / CRCL / MSTR / BMNR): BTC ~$76,874, about $300 million of IBIT net outflows, Ark selling into Monday's strength, and a 5% risk-free rate directly raises the carrying cost of hoarding coins — three negatives pointing the same way.
2. Long-duration application software (NOW / CRM / ADBE): with the 10Y at 5% and a hike likely tomorrow, valuation duration is the sharpest ranking variable of the day, and these were still up 5–7% on Monday.
3. Energy and refining: not because I am bearish on the commodity, but because the equity side has already finished pricing the "high" — the four pure refiners trade 11.8%–18.9% above sell-side average targets, P/B sits in the 99.6%–99.8% five-year percentile, and consensus already builds Q3'26 as the cycle peak with a preset Q4 cliff. Add that all six closed below their opening prices yesterday, with three fading from 52-week highs. The branch with the most seductive narrative is exactly the branch the criteria support least. (Exception: XOM/CVX still trade below target prices and should not be treated the same as the pure refiners.)
4. Names with unclear attribution: NOK (the drop includes company-specific negatives), AXON (the convertible claim is unconfirmed), WAY (the acquisition rumor is unconfirmed and pre-market was only $2 million).

④ Final one-sentence judgment

Today is a "cross-section day," not a "direction day": the index barely moves, but Monday's two legs are walking back at the same time, and a full-sample correlation of −0.55 says mean reversion is the single strongest structure of the day — but with R² = 0.30 it cannot explain 70% of the cross-sectional variance, so do not treat it as the only logic. In absolute magnitude, the fading leg retraced −2.06% this morning, more than the +1.36% repair in the reversal leg, and the market's confidence in "an AI slowdown is good for cybersecurity" really is weaker, but the gap is only 0.7 percentage points. Yet the real counterparty to all these positions is not today — it is in tomorrow's 14:00 ET statement, which may carry the first rate hike since July 2023. A short-duration bounce thesis paired with an overnight event risk means the most rational approach is to compress the position horizon inside the day.

Data Sourcing and Failure Log (internal)

The data pipeline used this run and the failures actually observed:

1. The Nasdaq official quote API is the only trustworthy channel for this brief's pre-market data, and two checks must be added:
   • primaryData.isRealTime == true, and lastTradeTimestamp containing AM ET/PM ET (i.e. with hours and minutes).
   • Reverse-compute lastSalePrice − netChange ≈ secondaryData.lastSalePrice (2% tolerance).
   • Without these two checks things go wrong: the top five of the "pre-market gainers" list from the first scan (759 names) — AJG +5.40%, FOXA +3.93%, TRU, GOOGM, GOOGN — were all fake data: isRealTime=false, timestamps with a date but no time, absurd bid-ask spreads (AJG bid $242.86 / ask $296.12, TRU bid $70.56 / ask $88.55, SMCIP bid $35.00 / ask $99.69), while the volume field returns the prior day's full-session volume, not pre-market volume. With the checks added, 574 of the 759 had live pre-market trades and 570 passed the benchmark reverse-computation. This is the same family as the "CBOE stale data" lesson on a different API, and should be hard-coded into the data-pulling script.
2. Sources that returned WebFetch 403: cnbc.com article pages, reuters.com, benzinga.com article pages. CNBC's quote API works (indices/futures/commodities/Treasuries); only the article pages are blocked.
3. fiscaldata.treasury.gov's upcoming_auctions endpoint returns stale data from March–May 2024 (HTTP 200, content two and a half years out of date), so it cannot be used to confirm whether there is a 20-year reopening today. Yesterday's recap mentioned "a $13 billion 20-year reopening at 13:00 ET today," which this brief could not independently confirm, so that event was not written into the body. TreasuryDirect's upcoming auctions page is JS-rendered and WebFetch cannot retrieve the table.
4. econoday.com's calendar was wrong: it listed "retail sales" on 9/15 and "initial jobless claims" on 9/16 (Wednesday). After checking against the Census Bureau's primary calendar page: August retail sales are released 9/16 at 08:30 ET, not today. Corrected per the primary source, and §8 of the body says so as well.
5. Three media attributions that were falsified or unconfirmed (all intercepted, none written into client-facing conclusions):
   • investing.com's market wrap said "Circle fell 13% on the Open USD launch" — that is actually old news from 2026-06-30/07-01 (140 firms including Stripe/Coinbase/BlackRock/Visa backing Open USD; CRCL closed down 17.55% at the time). Today's CRCL −5.10% cannot be attributed to it; the body instead uses three dated pieces of evidence: "weak BTC + IBIT outflows + Ark selling."
   • "Waystar rose on an acquisition rumor" — the dated facts I found are raised 2026 guidance + the appointment of a new CFO (at a different point in time), and the acquisition claim has no primary source. Combined with pre-market notional of only $2 million, §6 explicitly marks it do-not-touch.
   • "Axon announced a $1 billion convertible offering" — searching turns up only $1.75 billion of senior notes from March 2025; the September 2026 convertible claim is unconfirmed. The body marks it "attribution questionable."
6. yfinance was completely unusable this run — a pure IP-level ban, not a leftover local process. The energy sub-agent's test: .info / .fast_info all threw YFRateLimitError; with curl_cffi chrome impersonation, query1 still returned Edge: Too Many Requests; query2's quoteSummary gave 401 and fundamentals-timeseries gave 429. Leftover processes were ruled out with ps aux. The user's skill-parameter requirement to "verify with fundamentals-analyst (yfinance)" had that leg broken this run. Substitute pipeline that worked (recommend hard-coding):
   • Financials (revenue/EPS/cash flow/balance sheet) → SEC XBRL companyfacts (primary 10-Q/10-K), all obtained, and more authoritative than yfinance;
   • Prices/52-week/historical series → Yahoo's query2 chart endpoint (the only Yahoo endpoint not blocked);
   • Pre-market/OHLC/52-week → stockanalysis /api/quotes/s/;
   • Consensus/target prices → stockanalysis forecast/__data.json (original source tagged spg = S&P Global).
7. Definition warning (avoided in the body, but recorded here): in the pipeline above, trailing P/E uses SEC's GAAP TTM EPS while forward P/E uses S&P Global's adjusted consensus; the two use different accounting definitions (GAAP includes impairments and inventory gains/losses). Comparing across those columns is a systematic bias, not a difference between companies — the body used only "the direction of the gap between trailing and forward" and "the quarter-on-quarter path," and never subtracted the two columns to produce a valuation conclusion.
8. Two facts still to be verified, flagged as blank in the body rather than filled with speculation:
   • MPC's apparent leverage is distorted: net debt/equity 131% and liabilities/assets 72.7% are the worst in the group, but MPC consolidates MPLX (a midstream MLP) whose debt is borne by MPLX, and the $6.643 billion of minority interest also sits on the balance sheet. Until that is split out it cannot be used for cross-company leverage comparison, so the body does not cite MPC's leverage figures.
   • XOM registrant change: in SEC's directory XOM now maps to CIK 2115436 "ExxonMobil Holdings Corp," while the old CIK 34088's last filing was a 10-Q on 2026-05-04, with the Q2 10-Q filed under the new CIK. Only the fact that it happened was confirmed, not its nature or its effect on shareholders, so the body does not mention it.
9. The per-barrel refining margin for all six companies was not obtained, because it is non-GAAP, disclosed only in EX-99.1 and on the call, and does not go into XBRL. The body explicitly leaves it blank and states that "the theoretical crack spread ≠ a company's actual capture rate," without back-solving a homemade metric from revenue/cost.
10. Scan data has been written to disk and the statistics can be recomputed (the first draft did not save it, and QC judged it unauditable on that basis):
    • work_uspre0915/premarket_scan_20260915.csv — per-name pre-market return/price/volume/notional/Monday return/read time for 570 names
    • work_uspre0915/stats_20260915.json — correlations for the three samples (−0.5479 @562 / −0.7698 @113 / −0.7937 @66), breadth counts, and sample-construction notes
    • Sample definitions have been aligned: 759 screened → 574 with live pre-market trades → 570 passing the benchmark reverse-computation → 562 that also have a Monday return (i.e. the sample used for the correlation). The first draft mixed 562/570/574 in the body; now unified.
11. Parts that risk-auditor judged "handled properly, do not touch" (recorded to prevent mistaken edits next time): pre-market definition discipline (per-name read times + reverse-computation checks), the two interceptors for Nasdaq fake data, the primary-source correction of econoday's date error, the proactive interception of the three media attributions, leaving VERA's cash runway blank rather than speculating, §5⑨ using ZS to overturn my own generalization that "cybersecurity is all crowded," and not writing the unconfirmed 20-year reopening into the body.
12. Errors found and fixed in this QC round (11 in total, by severity): ① The totals in all six columns of the §4 scoring table failed to reconcile with the sum of their components (errors of +15 to +20), and the maximum was mis-written as 100 (actually 95) — re-summed from the components and §3 re-ranked; ② §3 and §7 gave opposite action tiers for MU/LITE/COHR — unified to follow §7; ③ VERA's verification point was a volume threshold that would "necessarily pass" — changed to a gap-fill test ($39.10); ④ Three places cited the strongest correlation of −0.79 (which carries the selection bias of a liquidity threshold) — changed to the full-sample −0.55 with the bias disclosed; ⑤ "Distance from the 52-week high = room to repair" has no predictive power (corr −0.023), and was falsified independently by two separate sources — removed from the ranking variables, and AMAT's entire rationale for inclusion was replaced; ⑥ §8② had CIBR as the subject while the data was IGV — unified to IGV; ⑦ The thin-volume threshold was not applied to the energy branch (PSX had only $0.3 million yet was used to support a strong conclusion) — downgraded to "unknown"; ⑧ §0④'s give-back ratio was a denominator artifact and only three of the five buckets were reported — all five are now shown and the point restated in absolute magnitude; ⑨ §9①'s "I deliberately diversified across variables" was said but not done (the reality is 3:1:1) — corrected honestly; ⑩ The S-grade top theme (rates) had zero current-period price evidence and banks were not mentioned once in the whole brief — rate-sensitive sector readings were added, along with the counter-intuitive evidence that "banks fell broadly on Monday"; ⑪ PLAY's "short watch" tag contradicted the body — changed to neutral avoid.
13. Gaps still unresolved, left for the next brief: whether there is a 20-year reopening today at 13:00 ET (the fiscaldata endpoint returns stale 2024 data; suggest switching to TreasuryDirect's auction announcement XML); the actual 08:30 Empire State print to be back-filled; current orders and book-to-bill for AMAT/LRCX/KLAC, ASML's latest-quarter bookings, MU's HBM capacity contracts and DRAM spot prices (the long case for these four still lacks current-period demand-side evidence); SOX's "nearly −6%" on Monday is a media summary and differs by 1.25pp from my own computed SMH −4.75%, not cross-checked.
14. Status of the other sub-agents: the three fundamentals-analyst runs and risk-auditor all returned and have been integrated. All three yfinance agents failed (see item 6); the substitute pipeline is hard-coded above.

---

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