US Pre-Market Brief | 2026-09-17 (ET) Thursday

Coverage window: 2026-09-16 16:00 ET regular-session close → 2026-09-17 08:24 ET. Covers Wednesday's after-hours, the overnight Asia/Europe sessions, and this morning's pre-market.

Methodology notes:

• Single-stock pre-market prices and percentage moves are taken from the Nasdaq official quote API, benchmarked to the 09-16 16:00 ET close. Every name passes two gates: isRealTime=true and a timestamp that carries AM ET; each one is then reverse-checked with pre-market price − change = previous close. 651 large caps were scanned, of which 469 passed validation and 64 were dropped because of isRealTime=false or a timestamp with no hour-minute component (indicative fake quotes).
• Indices / futures / commodities / Treasuries come from the CNBC quote API. Yesterday's Treasury closes were independently cross-checked against the official US Treasury curve (daily_treasury_yield_curve): official 09-16 2Y 4.74 / 10Y 5.01 / 30Y 5.35.
• ⚠️ CNBC's US2Y change_pct field is broken today: it reports +0.0898%, but that same API's own last 4.681 versus previous close 4.727 reverse-computes to −4.6bp. This brief uses the reverse-computed value for the 2-year throughout, not that field. (The 10Y/30Y fields are internally consistent and usable.)
• Pre-market quotes are cumulative, not point-in-time facts. This brief gives two reading moments: the main table at 07:35–08:08 ET, with a re-read at 08:22–08:24 ET attached for key names; both numbers are listed.
• Sector/theme baskets are pulled name by name and equal-weighted by me, not third-party sector indicators; every basket lists its constituents and count.
• Financial figures are taken exclusively from primary SEC EDGAR filings (8-K / 10-Q / XBRL companyfacts), not media paraphrase. Wherever a media figure differs from the filing, both numbers are written out in the text.
• ✅ The 08:30 ET Philly Fed and housing starts have been released, and this brief has been updated with the actuals and the market reaction (supplementary reading 08:41–08:45 ET, see the "Supplementary reading" subsection after §1). The main table remains the pre-release reading; wherever a re-read was done post-release, both numbers are written out.
• ⚠️ Initial jobless claims were not obtained this time (DOL page 403 + FRED not yet updated); the slot is left blank in the text, and last week's figure was not passed off as this week's.
• 🔴 Next gates: 10:00 ET pending home sales, 11:00 ET Lennar call, tomorrow (09-18) quadruple witching — all three fall after this brief.

---

0. One-Sentence Summary of the Day

1. Yesterday the Fed hiked because of oil, and today oil collapsed on its own — that is the one-sentence explanation for this entire morning. The Fed hiked 25bp to 3.75%–4.00% at 09-16 14:00 ET, 12-0 unanimous, the first hike since 2023. Warsh put the reasoning bluntly at the press conference: "We cannot influence any single price, but what we can do and will do is make sure that changes in relative prices do not spill over, do not produce second- and third-order effects." (Fed press conference transcript¹) — the "single price" he meant is oil. And this morning: WTI broke below $100 to 99.48 (−2.88%), Brent 102.14 (−3.49%), down for a third straight day. The very input that forced this hike is deflating faster than the hiking cycle can move. That is why equity futures are broadly stronger and yields are still falling this morning, one day after a hawkish turn.

2. But the sentence has to be finished: oil fell because the problem was "routed around," not "solved" — the Strait of Hormuz is still effectively blockaded by Iran. The timeline has to be laid out (Wikipedia: 2026 East–West Crude Oil Pipeline attack², CNBC 09-13³): after Hormuz was closed, Saudi Arabia rerouted roughly 5 million barrels/day through the East–West pipeline to the Red Sea port of Yanbu; on 09-10~11 that pipeline's pump stations were hit by drones (launched from Iraq's Maysan province), Saudi Arabia halted flows, and oil surged more than 8% that week and broke $100; on 09-16 Saudi Arabia began repairs and set up ship-to-ship transfers offshore of Oman's Sohar port as a workaround, and only then did oil start to retreat. 🔴 So the foundation under every long position today is a "temporary workaround." Pipeline repairs are expected to take another 3–5 weeks, while Hormuz itself has not improved at all and roughly 4% of global supply remains impaired. A single Iran-related headline could reverse this entire table this morning — and VIX is only 15.59.

3. The dot plot is milder than the media says: the median is "one more hike and then hold for a long time," not "a hiking cycle." I pulled the raw SEP PDF⁴ and checked it line by line: median fed funds rate: end-2026 4.1% / end-2027 4.1% / end-2028 3.9% / end-2029 3.6% / longer run 3.2%.

• Current range midpoint 3.875% → end-2026 4.1% = one more 25bp this year (10/28 or December).
• But the end-2027 median is also 4.1%, i.e. flat — there is no "another hike in 2027" in the median path. ⚠️ Multiple outlets wrote "the FOMC expects two more hikes, one this year and one in 2027," which does not match the SEP medians. Go with the primary document: one hike, then hold for over a year. More important are the other four SEP items: 2026 GDP revised up from 2.2% to 2.3%, unemployment revised down from 4.3% to 4.1%, while PCE was revised up to 3.7% and core PCE up to 3.4%. This is "hiking into a strong economy," not a stagflation panic — which is exactly why equities could absorb it, and why the Dow fell 1.21% while the Nasdaq Composite fell only 0.01%.

4. This morning is not a "oversold-bounce day," it is a "momentum-continuation day" — I ran this on the full sample and the conclusion is the opposite of yesterday's brief. corr(Wednesday return, this morning's pre-market return) = **+0.238**, n=462, **t=+5.25** (p<1e-6), R² = 5.6%. The sign is positive. Yesterday's brief computed −0.184 (mean reversion); today it has flipped: what rose Wednesday keeps rising this morning. GEV (Wednesday +4.79%), BE (+4.11%), POWL (+3.70%), CRDO (+7.38%), ALAB (+6.59%) are all near the top of this morning's gainers. ⚠️ The same caveat has to be stated precisely: R² is only 5.6%. The relationship is real and statistically highly significant, but it explains only 5.6% of the dispersion. The real dispersion is in themes, not in "did it rise yesterday." Breadth is extremely strong: of 469 valid samples, 381 up / 86 down / 2 flat (81.2% advancing), pre-market mean +1.23%, median +0.92%. By contrast, on Wednesday only 169 of 462 names advanced, with a mean of −0.53%.

5. The strongest single catalyst is Generac, and its actual terms are not what the headlines say. The 8-K filed at 09-16 16:24 ET⁵ is explicit:

• The "$8 billion" is the vesting cap on the warrants (a cumulative payment ceiling), not an order and not backlog.
• The actual near-term number written into the filing is: "Initial deliveries of backup generators are expected to total $2.4 billion in 2027 and 2028" — and the wording is "expected to," not a committed quantity.
• The company simultaneously issued Amazon warrants to purchase 1,693,745 shares at an exercise price of $200.9266, exercisable through 2033-09-16, of which 307,954 shares vested immediately. GNRC pre-market $228.00 / +30.20% (08:24 ET), pre-market volume 167,000 shares ≈ $38 million notional — the volume is real, not a thin-tape fake quote. See §5.1, which contains one piece of background that nearly everyone missed.

6. Energy is the only sector today that "broke down Wednesday and is still falling this morning" — the cleanest avoid signal in the whole table. On a broad rally day with 81.2% of stocks up, the energy E&P basket (8 names) is −0.94% pre-market, 0/8 advancing, after falling −5.50% on Wednesday. EOG (Wednesday −5.73%, this morning −0.72%), APA (−5.53% / −1.18%), DVN (−5.63% / −1.08%), FANG (−8.03% / −0.84%). Not one of them is bouncing.

Pre-Market Board at a Glance (readings 08:14–08:24 ET)

Wednesday (09-16) Close Recap

The 1.2 percentage point gap between the Dow and the Nasdaq is the key to understanding yesterday: the hike hit banks (XLF −1.62%, KRE −1.77%, GS −3.96%) and long-duration industrials, while the AI compute complex (SMH +0.64%, XLK +0.10%) did not fall at all.

---

1. News Overview

✅ Supplementary reading: the 08:30 ET data have been released (added after drafting, reading moment 08:41–08:45 ET)

The main table remains the pre-release (08:14–08:24 ET) reading; below is the post-release re-read, with both numbers listed.

Market reaction (pre-08:30 → 08:41 ET):

Three readings:

1. The housing data are soft (starts missed, completions −27.1% YoY), yet LEN's decline is narrowing and DHI is holding +1.05%. This further confirms the judgment in §5.6: the market today is weighting "falling rates" above "housing fundamentals." But note — single-family starts +7.6% MoM is the one bright spot in this release, and single-family is exactly the builders' main battlefield.
2. The Philadelphia Fed prices-paid index at 48.6 is still elevated, the one hard data point today that does not support "inflation pressure is fading." It is a reminder: the oil decline has not yet passed through to manufacturing input costs.
3. After the data, equity futures strengthened further, the oil decline widened, and VIX kept compressing — consistent with the judgment in §0, but GNRC, far from showing the gap-fill I flagged in §8, made a new pre-market high ($232). The verification signal in §8 needs to be re-clocked against this new reading (see the addendum below).

⚠️ Addendum to §8 item 1: at 08:24 ET I wrote "GNRC pre-market high $236.86, now $228.00, already down 3.7% from the high," and set $220 as the gap-fill criterion. The 08:41 re-read is $232.00, back near the pre-market high. This does not change the criterion itself ($220 is still the key level near the gap midpoint), but it weakens the real-time description "the fill is underway" — the pre-market selling did not follow through. Go by the actual action in the first 30 minutes after the open; do not treat the 08:24 description as a conclusion.

---

🔴 Remaining Gates Today

---

2. Strongest Themes, Descending

Theme Basket Measurements (all pulled and equal-weighted by me, readings 08:02–08:24 ET)

Basket constituents (listed name by name for verification):

• AI data center power / electromechanical: GNRC, ETN, VRT, GEV, POWL, BE, CMI, PWR, NVT, HUBB, EMR, PH, TT, FIX, MOD, BWXT
• AI compute / connectivity: NVDA, AMD, AVGO, MRVL, CRDO, ALAB, MTSI, ARM, TSM, SMCI, ANET, NBIS
• Gold and silver miners: NEM, GOLD, AEM, GFI, HMY, CDE, AU, KGC, PAAS, WPM, FNV, RGLD
• Semicap equipment: AMAT, LRCX, KLAC, TER, MKSI, ASML, ENTG, AEIS, ONTO, COHR
• Energy E&P: XOM, CVX, COP, EOG, FANG, DVN, APA, OXY
• Software SaaS: CRM, NOW, WDAY, HUBS, SNOW, DDOG, MDB, TEAM, INTU, ADBE, OKTA, ZS, CRWD, PANW, NET
• Homebuilding: LEN, DHI, PHM, TOL, KBH, MTH (⚠️ see the data correction on PHM below)

⚠️ A data correction that has to be written out — the homebuilding basket was nearly contaminated by a fake price. The Nasdaq API gave PHM (PulteGroup) pre-market $99.185 / −15.62%, with isRealTime=true, which passes my usual two gates. But that quote's timestamp is 04:00 AM, its volume field is just 0.01, and at that same moment the bid $117.00 / ask $119.79 bracket Wednesday's close of $117.55 precisely. I went further and pulled the tick detail: PHM's real pre-market prints were $117.87–$117.90 at 07:42 ET, about +0.3% versus the Wednesday close of $117.55. Had I taken −15.62% at face value, the homebuilding basket would compute to −2.32% and produce the conclusion "the Lennar blowup dragged down the whole sector" — whereas the true value is +0.34%, the exact opposite direction. This brief has excluded that fake price. Likewise, NVR (−1.13%) and TMHC were dropped for isRealTime=false and are not included.

---

3. Overall Single-Stock Strength Ranking (descending by total score; direction listed separately)

---

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

Scoring definition: source authority 0-15 / catalyst directness 0-20 / earnings elasticity 0-15 / moat and fundamentals 0-15 / expectation gap 0-10 / catalyst persistence 0-10 / tradability 0-10 / risk deduction 0~−15.

Scoring rationale (no black box), plus the four scores I changed after the sub-agent review:

• GNRC gets full marks on source and directness, but only 4 on expectation gap (already +30% pre-market) and a 13-point risk deduction (30% gap + the media treating $8B as an order + Q2 results including a $71M one-time tariff refund). Its catalyst quality is the best in the field, and its executability is among the worst. GNRC versus CAT at 65 vs 64 is within noise — the real differentiator is not the score, it is whether you can get filled.
• 🔻 ETN cut from 63 to 56: verified GAAP TTM P/E at the 5-year 97th percentile, gross margin −348bp, GAAP EPS −15.9%, operating leverage 0.81× (negative). Earnings elasticity 11→8, risk −4→−8.
• 🔻 PWR cut from 60 to 53: PEG 2.72 (worst in the field), P/S, P/E and EV-EBITDA percentiles all clustered at 88–89%, while FY27 growth is only +14.9%. Expectation gap 5→3, risk −4→−9.
• 🔻 POWL cut from 51 to 45: orders +158% but revenue only +8.9%, operating leverage 0.75% (negative), and 40% of revenue still from oil, gas and petrochemicals. Earnings elasticity 12→9, risk −5→−8.
• 🔻 BE cut from 41 to 38, label changed from "watch only" to "avoid": P/S 25.08× is 3.1× the second-highest in the field, and a single customer was 73% of 1H revenue.
• 🔺 CMI expectation gap 7→8: it is the only name in the power chain that was still falling Wednesday and only caught up this morning, and its forward P/E of 17.5× is the lowest in the group. But earnings elasticity 11→9, because overall growth is only 9.4% and operating leverage is negative.
• AMZN gets only 3 on earnings elasticity: $2.4B is immaterial at its scale, and it is the payer.
• NBIS gets 2 on source authority: second-largest gainer yet no verifiable same-day catalyst can be found (see §5.9); a big move is not back-solved into a high score.

---

5. Detailed Analysis of Top Names

5.1 GNRC (Generac) — the catalyst is real, but the number the market is reading is not

Related news: the 09-16 16:24 ET 8-K (filing⁵), Item 1.01.

Three definitions that must be corrected first (all from a word-by-word read of the 8-K):

1. The "$8 billion" is the vesting cap on the warrants, not an order. From the filing: the warrants vest in tranches, "contingent upon aggregate gross payments ... up to a total of $8 billion". This is an incentive clause meaning "the warrants fully vest once Amazon has paid a cumulative $8 billion," not a purchase commitment.
2. The real near-term number is $2.4B, and the wording is "expected to": "Initial deliveries of backup generators are expected to total $2.4 billion in 2027 and 2028". Not backlog, not committed.
3. GNRC has written in its own 10-Q risk factors that contract terms with certain data center customers include "cancellation rights". And Exhibits 4.1 and 10.1 to the 8-K were both redacted under Reg S-K 601(b)(10)(iv), so pricing, delivery cadence, cancellation rights and breach remedies are all invisible.

Catalyst logic: $2.4B spread evenly is $1.2B/year, which is +27% against TTM revenue of $4.44B, and +54% against the C&I segment (Q2'26 annualized $2.23B). Relative to the company's size, this is a very large contract.

Fundamental verification (primary SEC XBRL / 10-Q; see the note at the end regarding the yfinance channel):

🔴 Two facts that change the conclusion:

(a) The data center business sits in GNRC's lowest-margin segment. C&I adj EBITDA margin ex-refund is 12.6% versus residential at 25.7%, a 13 percentage point gap. And the company itself wrote in Q2'26 that C&I margin improvement was partly offset by an "unfavorable sales mix shift" — within C&I, the data center product mix is unfavorable.

(b) The market-cap increment does not reconcile with the revenue increment. $175.11 → $228.00 on a fixed 59.0M shares is a $3.12B increase in market cap. If you run $1.2B/year of incremental revenue at C&I's true profitability (after-tax net margin roughly 6–7%), annual incremental net income is about $72–84M — to support a $3.12B market-cap increase you would have to put a 37–43× P/E on that incremental profit, above GNRC's own 5-year median P/E of 33.5x. Put differently, $228 is not pricing "what $2.4B is worth," it is pricing the option on "Amazon as an anchor customer + an $8B runway + replicating this with more customers" — the part that is hardest to falsify and easiest to overestimate.

On dilution, I want to correct an easy mistake: 1,693,745 shares against 59.0M outstanding is 2.87% nominal dilution, but under the treasury stock method (the actual mechanics of GAAP diluted EPS), an exercise price of $200.9266 at $228 produces only about 201,000 net new shares = 0.34% real dilution; even at $400 it is only 1.43%. Dilution is not the main cost of this deal; the real cost is that the company sold Amazon a 7-year call option at $200.93.

A widely overlooked piece of background (this is the most important one): GNRC's 6/22 close was $295.54, after the 7/29 earnings report it fell to $195.19,9/16 it was $175.11 — before this news it had already drawn down 41% from the June high, and that June rally was itself driven by the data center narrative. At $228 pre-market it is still 23% below the June intraday high ($296.44). The market is not granting it a new peak valuation; it is repairing an option that had previously been cut by 40%.

Pre-market and technicals: $228.00 / +30.20% (08:24 ET). Pre-market high $236.86 (05:34), low $224.31 (07:28) — already down 3.7% from the high. Pre-market volume 167,000 shares ≈ $38 million notional, real volume. 52-week range $134.80–$296.44, currently at the 63.8th percentile.

Final judgment: watch only. Catalyst quality is the best in the field (primary SEC filing, explicit amount, multi-year), but the +30% gap has already eaten the odds, and the "$8B" the market is pricing off is a misread number. If you want exposure to this line, the better risk-reward sits in CAT / CMI, not GNRC itself — and specifically not ETN/PWR, whose valuation percentiles are in §5.3.

🔴 The most important unknown for this name (and the best falsifiable criterion): is Amazon the "second hyperscaler customer" announced on 6/24? The 8-K does not say. That determines whether $2.4B is entirely incremental or a formalization of volume already in backlog and consensus. The criterion is set to be falsifiable: GNRC disclosed data center backlog of about $1.6B as of 7/29. If the Q3'26 report (around late October) updates that to ~$4.0B (=1.6+2.4), it is purely incremental; if it only reaches $2.5–3.0B, a substantial part is a formalization of existing backlog and today's move is borrowed from the future.

---

5.2 CAT (Caterpillar) — the same logic, a quarter of the gap

Related news: no disclosure of its own. The GNRC event turned "hyperscale data center backup generation" from a narrative into a contract with a dollar figure, and CAT is the largest and most liquid name on that chain (large diesel/gas generator sets).

Fundamental verification (primary SEC XBRL, Revenues / OperatingIncomeLoss / ProfitLoss):

Q2'26 revenue annualizes to about $82B versus FY2025's $67.6B — clear acceleration, not a plateau. Operating margin went from 16.5% in FY25 to 20.9% in Q2. GAAP EPS $7.77 (+68.2%), adjusted $8.17 (+73.1%).

✅ The data center exposure is quantifiable, not conceptual (primary 10-Q MD&A):

• Power Generation external sales within the Power & Energy segment were $3.098 billion (+29%), 15.1% of total company revenue, and the MD&A explicitly attributes this "primarily to data center applications."
• The P&E segment overall was $8.238 billion, 40% of total revenue, with a segment margin of 24.6% (22.1% a year ago).
• Committed backlog $72.1 billion, +92.3% YoY; RPO $44.1 billion, +123.9% YoY.

⚠️ But there is a blind spot that must be written out: the $72.1 billion backlog is not broken down. Over the same period Construction Industries revenue was +35% ($8.346 billion), the largest incremental contributor in the company, and it has nothing to do with data centers. Reading the +92.3% as "data center orders" is wrong.

⚠️ A second reason to discount: about 17% of the EPS beat is non-operating. Other income was $398M versus $84M a year ago (+$314M, from FX, total return swaps and investment gains), which at a 23% tax rate across 460 million shares works out to about $0.53/share, 17% of the $3.15 GAAP EPS increment.

Valuation: P/S at the 5-year 88th percentile and P/E (GAAP TTM) 33.2× also at the 5-year 88th percentile, forward P/E 28.6×, PEG 1.39. This is a cyclical at peak multiples on peak margins — MP&E machinery gross margin of 34.7% is itself at a historical high.

Pre-market and technicals: $809.04 / +3.36% (08:24 ET), only −0.10% Wednesday. 52-week range $443–$1,073.46, currently at the ~55–58th percentile, −26.5% from the 6/30 high.

Final judgment: watch closely. It is the top pick under this brief's "right logic, don't chase" principle: it carries the same catalyst as GNRC with only one-ninth of the move, its data center exposure is the only one in the field that has an amount, a backlog, and is already in revenue, and its liquidity and options depth are the best in the field. ⚠️ But three things must be said at the same time: it has no catalyst of its own today (directness only 11/20), 17% of the EPS increment is non-operating, and the valuation is at the 5-year 88th percentile. This is not a cheap stock; it is the stock with the cleanest logic and the least extreme positioning.

---

5.3 CMI / ETN / PWR / VRT / NVT / FIX — the "middle" of the power chain, and three judgments overturned by valuation percentiles

Shared logic: Amazon's backup generation order validates not just generators but the entire "hyperscale data center electromechanical" spend. ETN (power distribution and electrical), CMI (generator sets, a direct comparable to GNRC), PWR (power engineering EPC), VRT (thermal management and power distribution), NVT (electrical connection and enclosures), and FIX (data center electromechanical EPC) each own a segment of it.

🔴 Three findings that change the conclusion:

① ETN must be downgraded — it is the most concealed of them. Its P/S percentile is only 67% (the lowest in the field), which looks the "safest," but its GAAP TTM P/E sits at the 5-year 97th percentile, i.e. the most expensive 3%. That 30pp gap between the two percentiles is exactly the shrinkage in earnings quality: revenue +21.4%, gross margin −348bp YoY, GAAP EPS −15.9%, net debt/EBITDA up to roughly 3.5×. What adj EPS +6.8% strips out is precisely M&A amortization and transaction costs — and M&A (Boyd Thermal / Fibrebond) is exactly where its growth comes from. Looking only at P/S produces the opposite conclusion.

② PWR must be downgraded — three different rulers point to the same conclusion. P/S 89%, P/E 88% and EV/EBITDA 89% all cluster at the top of their ranges, while FY27 consensus EPS growth is only +14.9% (second-worst in the field) and PEG 2.72 is 1.6× the second-place name. Even on the company's own looser adjusted basis, forward P/E only falls to 37.1× with PEG 2.49 — the conclusion is unchanged. And its exposure is second-order — it benefits from overall grid capex triggered by data centers, not from data centers themselves.

③ VRT's growth has a gap that cannot be falsified. Reported growth this quarter was +24.1%, but organic was only +18% (including 5% from M&A and 1% from FX); yet Q3 guidance is organic +34%~36% and full-year +30~32%. All 16 percentage points of acceleration are loaded into the second half, which the company attributes to "temporary supply chain congestion and timing mismatches in multi-phase project execution." And VRT disclosed no backlog in either its 2026Q2 8-K or 10-Q — without order data that statement cannot be falsified. It is the least verifiable name in the group.

Within this group I still prefer CMI, but the reasoning has to be rewritten. It is the only one that was still falling Wednesday (−1.96%) and only caught up this morning, its positioning is the lowest (37th percentile of its 52-week range), its absolute valuation is the cheapest in the field (forward 17.5×, FY27 15.1×, EV/EBITDA 15.1×, all three at the lowest percentile tier), and its data center exposure is quantifiable (16.2% of revenue) with that segment being the company's only margin-expanding one. ⚠️ But its flaws must be labeled honestly: CMI's overall growth is only 9.4%, operating leverage is 0.46× (negative), and EBITDA margin is down 90bp YoY. A 16.2% data center share cannot carry the whole. It is "cheap and low in its range," not "high growth."

Final judgment: CMI / FIX / NVT — watch closely; ETN / PWR / VRT — downgraded to watch only.

---

5.4 GEV / POWL / BE — the three already-crowded names on the same line, each with an accounting trap

These three were already rising on Wednesday (a day when 81% of stocks fell), which means money was already inside them. The GNRC news is a "confirmation" for them, not an "ignition" — and chasing during the confirmation phase gives the worst odds. More importantly, each of the three hides a number that is easy to misread:

GEV (Wednesday +4.79%, this morning +3.24% to $955.08)

• ✅ Order evidence is the hardest in the field: RPO $176.28 billion (+37.0%); quarterly orders $24.2 billion (+88% organic); gas turbine backlog plus slot reservations of 116 GW (100GW in Q1, year-end target ≥125GW); electrification segment data center orders YTD >$5 billion, more than twice all of 2025.
• 🔴 But its GAAP P/E of 25.9× (2nd percentile) is fake and cannot be used. TTM GAAP net income is $9.529 billion while TTM GAAP operating income is only $1.799 billion — the difference is almost entirely non-operating: the 10-Q footnotes show 1H26 includes a $3.992 billion pre-tax revaluation gain on the acquisition of the remaining 50% of Prolec GE plus a $330 million gain on the Proficy divestiture. The true forward P/E is 60.2×, the highest in the field (BE aside).
• Likewise, FCF of $12.4 billion far exceeds its $1.799 billion of operating income — that is gas turbine customer prepayments, not profit.
• ⚠️ The wind segment posted a quarterly EBITDA loss of $275 million with orders −39%.
• Watch only.

POWL (Wednesday +3.70%, this morning +3.57% to $184.05)

• ✅ The orders are real: quarterly new orders $934 million (+158%), book-to-bill 3.0×, backlog $2.4 billion (+69%), including a single data center mega-order >$400 million.
• 🔴 But the orders have not turned into revenue at all: revenue is only +8.9%, gross margin −10bp YoY, operating leverage 0.75× (negative). Broken down by end market, "commercial and other industrial," which contains data centers, generated only $76.32 million of revenue this quarter (24% of the total), while oil and gas plus petrochemicals is still 40%, with petrochemical revenue −49% in the quarter.
• This is the largest orders-versus-revenue gap in the group, and the current multiple is already priced for completed conversion. A $6.5B small cap with only 4 analysts covering it.
• Watch only.

BE (Bloom Energy, Wednesday +4.11%, this morning +4.49% to $282.13) — valuation has left the comparable framework

• Q2 revenue +165.5%, swung to profit, the fastest growth in the field.
• 🔴 P/S 25.08×, the 5-year 90th percentile, against a 5-year median of just 4.54×. The second-highest in the field, VRT, is only 8.03× — BE is 3.1× that. Forward P/E 131.1×, EV/EBITDA 192.1×.
• 🔴 Revenue is highly concentrated in very few customers: from the 10-Q — two customers were 44% and 21% of Q2 revenue (the second is a related party); a single customer was 73% of 1H revenue; US revenue share rose from 59% to 90%. The company also disclosed an AI data center power partnership with Oracle, and recognizes the $251.6 million fair value of warrants granted to Oracle as a reduction of revenue, amortized over time.
• What supports 131× is an expectation that revenue quadruples in two years, and that curve rests on 1–2 projects. Any slip in one project hits the numerator and the denominator at once.
• This brief downgrades it from "watch only" to "avoid" — not because the logic is wrong, but because at this level being right does not pay commensurately while being wrong costs a lot.

---

5.5 AMZN (Amazon) — the one party in this story everyone ignored

Related news: it is the counterparty to the 8-K. CNBC's headline is "Amazon obtains right to buy up to $340M of Generac" (1,693,745 × $200.9266 = $340.3M, the numbers tie out).

The direction question that must be made clear: this agreement is "spending" for AMZN, not "revenue." The $2.4B of backup generation purchases is capex. The market reads it as bullish on the logic that it "locks in 2027-28 data center capacity and power assurance," which is execution confirmation, not margin improvement.

Pre-market $251.02 / +2.06% (08:24), Wednesday −0.99%. Final judgment: watch closely. It is the only mega-cap on this line where you do not have to worry about chasing a gap, but be honest too: how much of the +2.06% is market beta (Nasdaq futures +1.39%) and how much is this news cannot be separated.

---

5.6 LEN (Lennar) — the print really is bad, but the bearish reason it cited is reversing this morning

Related news: 09-16 16:42 ET 8-K EX-99.1 (filing⁷).

Primary figures (all from the company press release, not media paraphrase):

But three details run counter to the "blowup" narrative and must be written out:

1. Gross margin improved sequentially. The CEO's own words: "Our gross margin improved sequentially to 15.8%". And Q4 gross margin guidance is 15.5%–16.0%, essentially flat, not further deterioration.
2. Q4 delivery guidance of 22,000–23,000 homes is materially above Q3's 20,840.
3. Q4 SG&A guidance improves to 8.7%–9.0% (Q3 was 9.2%).

🔴 The most important item — the CEO himself attributed the weakness to the very variable that is reversing this morning:

"Mortgage rates increased through the quarter, with the 30-year rate at approximately 6.8% at quarter end and even higher since. Rates are responding as inflation remains above the Fed's target, driven by geopolitical tension and higher oil prices."

The causal chain Stuart Miller named is: geopolitics → oil → inflation → rates → housing demand. And this morning the first link in that chain is running in reverse (Brent −3.49%, 10Y −5.3bp).

🔴 But the ugliest part of this report is not in the income statement, it is on the balance sheet — and almost no coverage mentioned it:

Deliveries are falling, inventory is rising, cash is down by two-thirds, and net debt is up fivefold. That is a far more serious signal than a 5.4% EPS miss. And Q4 guidance embeds destocking: delivery guidance of 22,000–23,000 homes against new order guidance of only 19,500–20,500 — deliveries exceed orders, so backlog will keep shrinking (currently 16,857 homes / $6.345 billion, with the dollar value −4.6% YoY).

Valuation: rolling Q3 in, TTM EPS is about $5.32 and TTM P/E 14.7x; P/B 0.88x — the only one of the 11 companies across the two groups trading below book value per share (BVPS $89.49; ex-goodwill tangible book is about $75.2/share, i.e. 1.04x tangible book). Annualized ROE for the quarter is only about 5.3%. Judgment (not fact): a 0.88x P/B usually reflects market expectations of subsequent inventory writedowns or of ROE staying below the cost of capital — consistent with the direction of that 5.3% ROE.

Peer comparison (most recent quarter home sales gross margin, all company-reported): PHM 25.0% > TOL 23.9% > DHI 20.7% > NVR 19.2% > LEN 15.8%. All five compressed YoY, but LEN is the only one below 16%. Quarterly net margin: PHM 11.9% / TOL 10.5% / NVR 10.1% / DHI 9.8% / LEN 3.5%.

Pre-market and technicals: $77.56 / −1.02% (08:24, narrowed from −1.74% at 07:50). 52-week range $76.63–$139.44, currently at the 2.8th percentile of that range, essentially at the 52-week low (while the 52-week high of $139.44 was set on 2026-09-10, i.e. 7 days ago). The company repurchased 3 million shares in Q3 at an average price of $85.49 — the current price is 9.3% below the company's own buyback cost. ⚠️ Pre-market volume is only 33,618 shares against a 10-day average of 3.53 million — thin, so the indication is limited.

Final judgment: watch only (do not short). The reason to be short has already been eaten by the price (52-week low + guidance cut already announced + flat Q4 gross margin guidance), while the reason to be long has not yet been verified (needs the 11:00 call to confirm a rate peak). ⚠️ At the same time, be honest: the homebuilding basket is +0.34% this morning (after excluding PHM's fake price), with DHI +1.05% and TOL +1.16% both rising — the market today is weighting "falling rates" above "Lennar's results." If you want to express a bullish view on the housing chain, DHI / TOL are cleaner than LEN.

---

5.7 FLNC (Fluence Energy) — the only "short watch" in the table

Related news: 09-16 after hours, a sharp FY2026 guidance cut.

Cause: persistent and deepening production delays at the Houston contract manufacturer. On 09-14~15 the company had just reshuffled management, appointing AES veteran Bernerd Da Santos as EVP and COO.

Why this is more serious than an ordinary guidance cut:

1. This is the second cut (there was already one), which means management has lost visibility into its own capacity.
2. Going from profit to a $200M EBITDA loss is a change in the nature of profitability, not its magnitude.
3. Revenue was cut about 20%, and the reason is "we cannot deliver," not "demand is gone" — supply-side execution failure is usually harder to fix quickly than demand weakness.

Pre-market: $7.08 / −21.77% (08:24). 52-week range $7.26–$33.51 — the current price has broken below the 52-week low, a new low. Market cap is only $1.31B.

Final judgment: avoid / short watch. ⚠️ But the execution risk of shorting has to be stated clearly: a $1.31B market cap, a $7 share price, and a 22% decline already mean borrow costs and squeeze risk are both high, and most of the decline has already happened pre-market. This is a "do not touch it long" name, not a "good short" name.

---

5.8 Energy E&P — Wednesday's big red candle is not an oil story; unpacked, it has three layers

First, a correction to one of my own definitional errors. I originally measured energy stocks' oversold extent against "WTI fell about 2.4% Wednesday," but that number is actually the intraday move on 9/17 (today), not the 9/16 daily change. Per the EIA official Cushing WTI spot: 9/15 = $107.02 → 9/16 = $102.43, a −4.29% move that day (USO −3.52% that day, Brent settlement $105.83 — all three channels agree in direction). Comparing 9/16 stock declines against the 9/17 oil decline systematically overstates how "oversold" they are. (This is exactly why cross-market comparisons must annotate each side's reading moment.)

After the correction, the "oversold" conclusion still holds, and holds strongly. Using a two-factor regression over the 255 trading days from 2025-09-10 to 2026-09-16 (stock ~ USO + SPY, excluding 9/16 itself and then extrapolating):

These companies' measured beta to oil is only 0.28–0.44 (not 1), so a 3.5% oil decline "should" correspond to only a 1.0–1.6% stock decline. FANG's −6.41% residual is the single most extreme day in 254 trading days, bar none.

Strip out the sector too (regress on XLE) and it splits into three layers — they cannot be lumped together:

🔴 So "energy stocks blew up as a group" is wrong: XOM's and SLB's residuals are right around the median (125/254, 120/254), and their declines are fully explained by sector beta. Putting them alongside FANG and calling it "oversold" throws different things into the same basket.

The real culprit behind FANG has been found, and it is confirmed by a primary SEC filing — it is not oil

Form 144, accession 0001140361-26-036761,2026-09-16 accepted:

• Seller SGF FANG Holdings, LP, identified as a "10% Stockholder"
• 9,079,675 shares, totaling $1,920,623,653 (about $1.92 billion), proposed sale date 09/16/2026, broker Morgan Stanley
• Source of the shares: consideration shares from the 2024-09-10 acquisition of Endeavor Parent LLC (originally 114.9 million shares)
• 3.24% of shares outstanding

The volume signature matches perfectly: FANG traded 15.4 million shares on 9/16 versus a 20-day average of 1.95 million = 7.89×; that single 9.08 million share print was 58.9% of the day's entire volume. Peer comparison: COP 1.42×, EOG 1.49×, DVN 1.40×, XOM only 0.96× — only FANG saw an order-of-magnitude volume surge. Media reported the block priced at $205.80 (a 2.7% discount to the 9/15 close), while FANG closed at $194.54, 5.5% below the block price — the buyers were underwater the same day, the classic overhang pattern.

Another decisive piece of evidence: during the 26.5% oil surge from 8/28 to 9/15, these stocks captured only 7%–8.4% (consistent with a beta around 0.3). But in one day on 9/16, FANG gave back that entire move and then some — a net change of −1.58% versus 8/28, while oil over the same span is still +20.4%. The stock completed a full round trip; oil did not.

⚠️ On the "Morgan Stanley downgrade": searches returned claims of a 09-16 downgrade, but (a) rating actions do not appear in SEC filings, so there is no primary source to verify; (b) the same batch of results also contained "08-19 Equal Weight" and "maintains Overweight, price target $229→$216," which are mutually contradictory with questionable dates; (c) Morgan Stanley was simultaneously the broker on this block, so the "Morgan Stanley" in media headlines is likely partly that role being transcribed as a "downgrade." This brief's verdict: the decline was driven by the $1.92 billion block sale, and any rating action is a secondary factor.

For COP / EOG's residuals (−2.58% / −2.48%) I did not find corresponding single-stock news, and per "no news found ≠ no news" they are marked unexplained. Two pieces of context: both set their 52-week highs exactly on 9/15 (the day before the drop) and sit at the 83rd/82nd percentile of their ranges; and FANG's block discount spills over into a pricing reference for the entire Permian. This is inference, not evidence.

Breakevens and elasticity (regressing EPS on EIA quarterly average prices across 17 quarters, 2021Q1–2026Q2): EPS elasticity per 10% rise in oil is DVN +20.1% > FANG +18.8% > COP +18.1% > XOM +17.9% ≫ SLB +5.5%. ⚠️ SLB's slope is statistically zero (R²=0.07); its beta to USO is only 0.037 while its market beta is 0.896 — SLB is fundamentally not an oil-sensitive name, it is a market-beta name. Putting it in an "oil-sensitive" basket for cross-sectional comparison is using the wrong ruler. ⚠️ EOG's regression R² is only 0.25 (gas mix and hedging degrade linearity), so its implied breakeven price cannot be cited. With WTI around $100, all four meaningful companies (FANG/COP/XOM/DVN) are far above breakeven.

Final judgment: avoid (explicitly do not short). 🔴 This is the most asymmetric risk in the brief: shorting energy is equivalent to being long "the Saudi workaround keeps working + Iran does not attack again." Yet Hormuz is still blockaded, pipeline repairs need another 3–5 weeks, and the 09-10 attack itself came without warning. Downside may be another 5–10%; upside (one headline) could be a 15–20% gap. The odds are inverted. And now there is a second reason: the largest single component of FANG's big red candle was a one-off block sale, not fundamentals — which means a substantial part of its "oversold" condition should never have been extrapolated into a trend.

⚠️ Another definition that must be clarified: multiple outlets treated Tuesday's API "build of 7.14 million barrels" as Wednesday's inventory conclusion. The official EIA figure is a draw of 600,000 barrels (expected −1.4 million), the opposite direction. This brief uses the official EIA weekly¹⁰.

---

5.9 NBIS (Nebius) — second-largest gainer today, but I could not verify its same-day catalyst

Pre-market $229.36 / +9.55% (08:24), Wednesday +0.96%. This is second among the 469 valid samples (behind only GNRC).

🔴 A verification result that must be written out: multiple financial sites published under URLs dated today (09-17) attributing NBIS's rise to "the Palantir partnership making Nebius its preferred sovereign AI infrastructure partner." I checked the publication date at Nebius's official newsroom: that partnership was announced on 2026-09-08, nine days ago, not today. (Nebius Newsroom¹²) These articles stuffed an old announcement into a recurring column carrying today's date. Copying it would turn news that has been priced for nine days into a "catalyst of the day."

So my honest conclusion on NBIS is: catalyst unconfirmed. A possible explanation is Amazon capex confirmation → sentiment spillover across neoclouds (peers IREN +5.40%, HUT +5.20%, DOCN +4.52% are all up sharply this morning, and that resonance is real), but I cannot attribute the +9.55% to any verifiable same-day NBIS-specific news.

Final judgment: watch only. "Cannot find news" is not "there is no news," but until it is found, a big move should not earn it a high score. Its "source authority" in the scoring table is only 2 points.

---

5.10 Semiconductors and AI compute (ARM / AMD / MU / LRCX / CRDO) — a bounce, not a reversal

The background has to be laid out first, otherwise a bounce gets misread as a new trend:

• 09-14: "Stocks Slump on AI Slowdown Fears and Rising Bond Yields and Crude Prices"
• 09-15: "Stocks Settle Lower as AI Slowdown Fears Weigh on Chipmakers"; the same day "10-Year T-Note Yield Climbs to a 19-Year High"
• ARM is −39% over the last 3 months; GLW broke below its 200-day on 09-14.

I verified the "19-year high" claim against official Treasury data: the 10Y closed at 5.01% on 09-16, the highest of 2026; the 2023 peak was 4.98%, 2024's 4.70%, 2025's 4.79%, and the last time it was above 5.01% you have to go back to 2007 (peak 5.26%) — "19 years" holds.

The semicap equipment basket (10 names) is +3.13% pre-market. ⚠️ But this line has to be marked down a notch: equipment (WFE) revenue speaks to future supply, not current demand. Strength in AMAT/LRCX/KLAC cannot be treated as confirmation that "AI demand is back" — this is an indicator whose direction is easy to get backwards.

Final judgment: AMD / MU / LRCX — watch closely; ARM / CRDO — watch only. Reasoning: this group's gains this morning have no single-stock catalyst whatsoever; they are beta from "falling rates + risk appetite repair." ARM's +4.70% is a bounce out of a −39% hole, and CRDO has now risen two days straight. One day's cross-section cannot prove leadership has returned.

---

6. Negative / Avoid List

🔴 Software SaaS deserves its own sentence: this morning's mainstream narrative is "yields fall → long-duration assets recover." But the software basket, which is equally long-duration, is down (−0.70%, 4/15), while semiconductors are +3.13% and AI compute +3.74%. If this were genuinely duration-driven, software — which should rise most — would not be at the bottom. That tells you this morning's rally is a repair of the AI capex narrative, not a rate-driven broad rally — and that distinction determines which basket you should buy.

---

7. Intra-Theme Ranking

Theme one: AI data center power / electromechanical (strongest today, basket +4.77%, 16/16 advancing)

Graded by "strength of exposure evidence" (only two criteria: ① is there an amount in the financials attributable to data centers; ② is that amount already in revenue, or only in backlog)

Important: not one of the 9 is a pure conceptual mapping — each can produce at least one verifiable order or revenue data point. But "the exposure is real" and "how much is already in the price" are two orthogonal questions.

Composite ranking (exposure evidence × valuation position × operating leverage):

Theme two: energy (weakest today, basket −0.94%, 0/8 advancing) — inverse ranking

---

8. Open-Bell Verification Signals

Pre-market (before 09:30)

1. Is GNRC gap-and-go or gap-fill? Pre-market high $236.86 (05:34), now $228.00, already down 3.7% from the high.
   • Bullish confirmation: holds above $228 in the first 30 minutes and reclaims $232.
   • Bearish confirmation (the one to watch): breaks below $220 → the 30% gap is being filled, and the sentiment anchor for the entire power chain loosens.
   • 🔴 $200.9266 is the warrant exercise price, and it is now a meaningful psychological/structural level.
2. Is the power chain "following" or is GNRC "alone"? If GNRC opens high and fades while CAT/ETN/CMI/PWR still hold above +2%, the market is buying the theme rather than a single event — a healthier structure. Conversely, if only GNRC rises and the rest stall out, it is a one-off event-driven pulse.
3. Does energy see a snapback? If FANG/EOG/DVN turn positive while the broad market is +1%, shorts are starting to cover and the avoid logic needs reassessment.

Intraday (first 30 minutes after the open)

4. Direction of the reaction to the three 08:30 data points (not yet released when this brief was published):
   • Philly Fed expected 31.3 / prior 47.4, with an extremely wide error band, the easiest source of a surprise today.
   • Housing starts expected 1.320 million / prior 1.239 million, colliding directly with Lennar's guidance cut — if starts are strong while Lennar is weak, the housing chain will split internally and DHI/TOL will separate from LEN.
5. 11:00 ET Lennar call: management's view on whether the 30-year mortgage rate has peaked is the real pricing point for the housing chain today, more important than the reported numbers themselves.
6. Sector ETF linkage: whether XLI / XLU move with the power chain names; whether the SMH-versus-IGV divergence persists (this morning semis +3.13% vs software −0.70%). If IGV turns positive, this really has become a rate-driven broad rally; if the gap keeps widening, it is a structural AI capex move.

Options and volatility

7. VIX at 15.59 (−11.97%) is the classic post-FOMC volatility compression. ⚠️ But today it is abnormally low: the Fed just started its first hiking move since 2023, the 10Y is at a 19-year high, and Hormuz is still blockaded, yet VIX is under 16. That is not evidence that "the market is calm," it is evidence that "the tail is not priced."
8. Tomorrow (09-18) is quadruple witching. Position adjustment and index rebalancing starting in today's closing hour will contaminate price signals — the strength or weakness of today's final hour cannot be used directly to infer tomorrow's direction.

Risk signals (downgrade on appearance)

9. 🔴 Any new headline on Iran / Hormuz / the Saudi pipeline — this is the one variable today that can reverse the entire table.
10. Reversal after the gap: if GNRC spikes at the open then breaks $220, the power chain's chasers will be trapped at the highs.
11. Rising alone: if only GNRC rises while CAT/ETN/CMI stall.
12. Futures and Treasuries diverging: if equities keep rising while the 10Y climbs back above 5.00%, then the "rates have peaked" premise does not hold and this morning's bullish logic loses its foundation.

---

9. Final Conclusions

① The 5 names most worth watching today

② The 3 strongest themes today

③ What to avoid today, and why

1. Energy E&P and refiners (FANG / COP / EOG / DVN / APA / DINO / MPC) — the basket is −0.94% this morning with 0/8 advancing, the only sector completely wiped out on a broad rally day. But explicitly do not short: Hormuz is still blockaded and the odds are inverted. 🔴 And the attribution has to be corrected: the main cause of FANG's −8.03% on Wednesday was a $1.92 billion block sale (confirmed by SEC Form 144, 58.9% of the day's volume), not oil. And after stripping sector beta, XOM's and SLB's residuals rank 125/254 and 120/254 — they have no single-stock problem at all. "Energy stocks blew up as a group" is a mistaken bundling.
2. FLNC — guidance cut 20% and swinging to an EBITDA loss, broke the 52-week low. The only name on the "short watch" list, but hard to execute.
3. Power chain names already at extreme valuations (BE / PWR / ETN / POWL / GEV) — these are the judgments I changed after the sub-agent finished checking valuation percentiles:
   • BE: P/S 25.08×, 3.1× the second-highest in the field (VRT 8.03×), and a single customer was 73% of 1H revenue → avoid.
   • PWR: P/S 89% + P/E 88% + EV/EBITDA 89% all clustered at the top, FY27 growth only +14.9%, PEG 2.72, worst in the field.
   • ETN: the most concealed of them — a P/S percentile of only 67% looks safe, but its GAAP P/E is at the 5-year 97th percentile, with gross margin −348bp and GAAP EPS −15.9%.
   • POWL: orders +158% but revenue only +8.9%, conversion has not even started while the multiple is priced for completed conversion.
   • GEV: the GAAP P/E of 25.9× is fake (96% of profit from the Prolec revaluation), true forward P/E 60.2×.
4. Software SaaS — basket −0.70%, only 4/15 advancing, falling against the tape on a broad rally day. Do not file it under the "duration recovery" narrative.
5. NBIS — second-largest gainer but its same-day catalyst cannot be verified (the supposed Palantir partnership is actually 09-08 old news).

④ Final one-sentence judgment

Today is the day when "the reason for the hike is disappearing faster than the hike itself": the Fed hiked because of oil, and oil broke below $100 within 18 hours of the hike, so equity futures, yields and volatility all voted yes at once (81.2% of the 469 valid samples advanced). But the real lesson of the day is "do not bundle things together": GNRC's $8B is a warrant vesting cap and not an order, FANG's big red candle is a $1.92 billion block sale and not oil, ETN's cheap P/S hides a P/E at the 97th percentile, and GEV's 25.9× GAAP P/E is a one-time revaluation gain — each is internally consistent, and each will lead you the wrong way. What is worth doing is using CAT / CMI to express the power-spending logic that SEC filings have nailed down; what to watch out for is that the Strait of Hormuz is still blockaded and VIX is only 15.59, and the foundation under this table is a temporarily rerouted oil pipeline.

---

⚙️ Data-Sourcing & Ops Log for This Issue (Internal)

yfinance channel: local yfinance was again rate-limited by Yahoo (YFRateLimitError), and a direct call to query2.finance.yahoo.com/v8/finance/chart returned Edge: Too Many Requests. ps aux confirmed no leftover retry processes → this is an IP/endpoint-level block, and not a single number in this brief comes from yfinance.

• Substitute channels (all verified working): SEC EDGAR (XBRL companyconcept / companyfacts + 8-K/10-Q filings), Nasdaq api/quote/*/info (pre-market), CNBC restQuote (indices/futures/commodities/Treasuries), US Treasury daily_treasury_yield_curve CSV (Treasury close verification).

Channels that failed or were restricted in testing:

• CNBC web articles all 403 (cnbc.com/2026/09/16/fed-rate-decision-*, stocks-making-the-biggest-moves-premarket); only the quote API works.
• CNN returns 451, thestreet 403, stocktitan DNS timeout, Generac IR page 60s timeout.
• WebSearch intermittently unavailable, recovering after about 6 failed calls, which prevented the ARM / semiconductor catalyst searches from completing — the "no single-stock catalyst" conclusion in §5.10 is partly a result of restricted search, not of exhaustive verification, and the text states "not found" rather than "does not exist."
• The Nasdaq extended-trading endpoint returned empty rows for most tickers (only GNRC/PHM succeeded), so pre-market volume was obtained for only a few names and no full-sample notional threshold could be applied.
• EIA ir.eia.gov/wpsr/table1.csv returned a 302 empty response and psw01.html a 403 (outside the release window), so the EIA inventory figures come from secondary paraphrase and the primary CSV was not obtained.
• Initial jobless claims were not obtained today: dol.gov/newsroom/releases/eta/eta20260917 returned 403; the PDF downloaded from dol.gov/ui/data.pdf extracted as empty via pdfminer (likely an image-based or dynamically generated file); FRED ICSA was still stuck at 206,000 for 2026-09-05 as of 08:45 (i.e. the prior week, not yet updated for this week). The "206,000" returned by search is also the prior week's figure, not today's print — it is marked "not obtained" in the text, and last week's number was not passed off as today's.
• The Philly Fed and housing starts were both obtained from primary sources (Philly Fed website / Census CB26-147 PDF extracted via pdfminer), so those two are reliable.
• The Fed's SEP and press conference PDFs failed to parse via WebFetch (binary stream); local pdftotext -layout / pdfminer extraction succeeded instead. ✅ I re-verified the SEP's median fed funds rate with a second independent extraction (locating the raw text of the Federal funds rate line): the sequence is 4.1 / 3.8 · 4.1 / 3.6 · 3.9 / 3.4 · 3.6 · 3.2 / 3.1, i.e. one pair per year (September median / June projection), confirming end-2026 4.1% (June was 3.8%, revised up 30bp), end-2027 4.1% (flat), end-2028 3.9%, end-2029 3.6%, longer run 3.2%; the 2026 central tendency is 4.1–4.4 with a full range of 3.9–4.4 (lower bound 3.9 = some participants see no further hike this year, upper bound 4.4 = some see two more). The "read the wrong column" risk I worried about in the draft did not materialize, and the conclusion is usable.

Sub-agents: 3 fundamentals-analysts run concurrently, all returned (taking 13/19/23 minutes). All three switched to primary SEC EDGAR because of the yfinance rate limit, and the quality was high. Together they overturned 6 judgments in my draft (listed below), which shows this step cannot be skipped. risk-auditor QA: see the separate log below.

🔴 The 6 draft judgments the sub-agents overturned (by severity):

1. FANG's attribution was entirely wrong. My draft wrote the −8.03% as an oil-driven "oversold" move. It was actually SGF FANG Holdings' $1.92 billion block sale (Form 144, accepted 09-16, 9.08 million shares, 58.9% of the day's volume, 7.89× volume surge). This is a single-stock event, not a sector event. And after stripping sector beta, XOM's and SLB's residuals are near the median (125/254, 120/254) — they have no single-stock problem at all — so my draft's bundling of them with FANG into "energy blew up as a group" was wrong.
2. I read my oil benchmark at the wrong moment. I used "WTI fell 2.4% Wednesday" to measure how oversold energy stocks were, but that was actually today's 9/17 intraday move; 9/16 was actually −4.29% (EIA Cushing spot $107.02→$102.43). Cross-market comparisons must annotate each side's reading moment — this error systematically overstates "oversold."
3. I overrated ETN. The draft gave it 63 points and third place. Verification showed GAAP P/E at the 5-year 97th percentile, gross margin −348bp, GAAP EPS −15.9%, operating leverage 0.81× (negative). Its P/S percentile of only 67% makes it look safest, which is exactly what makes it the most concealed. Cut to 56 points and downgraded to watch only.
4. I overrated PWR. PEG 2.72 is the worst in the field, three valuation percentiles cluster at 88–89%, FY27 growth is only +14.9%, and the exposure is second-order. 60→53 points, downgraded.
5. I never looked at LEN's balance sheet. Net homebuilding debt $643M→$3,147M (5×), cash −67%, inventory +21% while deliveries fall, P/B 0.88× (the only one of 11 below book). This is far more serious than a 5.4% EPS miss.
6. GEV's GAAP P/E of 25.9× is fake (96% of profit from the $3.992 billion one-time Prolec GE revaluation gain), with a true forward P/E of 60.2×; I also computed GNRC's dilution too crudely (2.87% nominal, only 0.34% under the treasury stock method).

Data correction log:

1. PHM phantom price — Nasdaq gave −15.62% ($99.185, 04:00 ET, volume 0.01); tick-level verification showed real prints at $117.87–117.90 at 07:42 (≈+0.3%). Without excluding it, the homebuilding basket would go from +0.34% to −2.32%, reversing the conclusion's direction.
2. CNBC US2Y change_pct field is broken — it gives +0.0898%, while last/previous close reverse-computes to −4.6bp. Switched to the reverse-computed value.
3. Media definitional errors ×3 — (a) "the FOMC expects two more hikes (one in 2026 + one in 2027)" does not match the SEP medians (the 2027 median of 4.1% is flat against 2026); (b) several outlets wrote the API build of 7.14 million barrels as the EIA figure, while the EIA actually showed a draw of 600,000 barrels, the opposite direction; (c) NBIS's Palantir partnership was republished in a recurring column dated today, while the actual release date was 09-08.
4. Three additional data-pull traps the sub-agents found (unrelated to this brief's conclusions but worth recording): XBRL's TTM silently skips quarters (PHM/NVR/FANG do not separately tag the fourth fiscal quarter, so simply taking the latest 4 quarters yields a pseudo-TTM missing one quarter — after correction FANG's TTM P/E went from 15.4× to 37.0×); XOM changed its CIK (now 2115436; the old 34088 no longer has a ticker, and taking either alone gives an incomplete series); POWL did a 1-for-3 stock split on 2026-04-02, and SEC historical EPS is unadjusted while the price series is adjusted, so not handling it throws the entire P/E history percentile off by 3×.

---

⚠️ Risk disclaimer: this list is pre-market information review 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 stale information or factual errors. Rely on company disclosures/SEC filings, and do not use this directly as a basis for trading.