U.S. Market Recap | 2026-09-17 (ET), Thursday

Coverage window: 2026-09-17 09:30 ET open → 16:00 ET regular-session close; the after-hours portion extends to 17:06 ET.

Methodology notes:

• Single-stock/ETF closing prices, open/high/low, volume, and 52-week ranges are taken from the CNBC quote API (16:00 ET regular-session close basis), and percentage changes are back-computed name by name as close ÷ previous close − 1; the API's own change_pct field is not taken at face value. All 578 names requested, all 578 returned, not a single one missing.
• Market breadth and sector returns are my own equal-weighted, name-by-name calculations (S&P 500 constituents, n=503), not third-party sector indicators. Index levels are separately taken from CNBC.
• Theme baskets use exactly the same constituents and counts as this morning's pre-market watchlist, to keep pre-market/post-close comparable; constituents missing pre-market have all been filled in (see §3).
• Initial jobless claims are now primary-verified: the official DOL PDF was successfully extracted locally (this morning's pre-market piece failed to obtain it; this piece fills the gap, see §1).
• Crude uses the 14:30 ET official settlement price, with the 16:50 ET after-hours electronic reading also noted — both numbers are listed.
• ⚠️ CNBC's US2Y change_pct field is still broken today (it gives +0.1172%, while its own last 4.666 and previous close 4.727 back-compute to −6.1bp). This piece uses the back-computed value throughout for the 2-year, consistent with this morning's pre-market piece.
• ⚠️ Gold has a methodology conflict, and this piece makes no single assertion on it; see the end of §5.

---

0. One-Line Recap

1. Today was risk-on, but not the kind of risk-on this morning's pre-market watchlist described — the leader was not its No. 1 theme, it was its No. 3 theme. S&P +1.14%, Nasdaq +1.69%, Dow +0.61%, Russell +0.55%, VIX down another 12.82% to 15.44, 10-year Treasury −7.0bp to 4.934%. But the genuine leader was AI compute/semiconductors (basket +5.00%, 12/12 up), while the AI data-center power/electromechanical basket listed as the No. 1 theme this morning managed only +1.98%, with 5 of 16 closing down and an open→close of −2.73% — it gave back the most of all 12 baskets today.

2. The foundation of this morning's piece — "oil collapsed" — reversed on itself intraday, and all 7 energy names it singled out to avoid went up. WTI was at 99.48 (−2.88%) pre-market this morning and did indeed trade as low as 99.10 intraday, but it settled back at $101.91, just −0.51%; Brent settled at $104.82 (−0.96%), versus this morning's pre-market reading of −3.49%. The decline narrowed by roughly four-fifths. So: the energy E&P basket went from −0.94% pre-market (0/8 up) to +0.50% at the close (6/8 up); oilfield services/refiners went from −1.10% to +1.26%. The 7 energy names written explicitly into item 1 of this morning's "avoid" list (FANG/COP/EOG/DVN/APA/DINO/MPC) all 7 went up on the day, averaging +1.21%, with 4 of them beating the S&P. 🔴 And this is precisely the scenario preset by verification signal No. 3 in §8 of this morning's piece: "If FANG/EOG/DVN turn positive in an environment where the broad market is +1%, it means shorts are starting to cover and the avoid logic must be reassessed." All three turned positive, and the broad market was indeed +1%. The verification point judged it right while the headline conclusion judged it wrong — that is the single most worthwhile takeaway today.

3. Hit rate has to be measured with two different rulers, and the two rulers give opposite answers — that is today's real lesson. The 5 names singled out in §9① this morning (CAT/CMI/GNRC/DHI/GLW) were 4/5 up and 4/5 ahead of the S&P on closing returns, which looks like a good list; but measured as "executed at this morning's pre-market price," only DHI of the 5 made money (+0.43%), the rest all lost: CAT −1.29%, GLW −1.20%, CMI −2.99%, GNRC −9.11%. Widening the scope to all 20 names:

On closing prices, the Pass group beat the recommended group by 2.1 percentage points and the list looks inverted; on the executable pre-market price, the recommended group beat the Pass group by 2.15 percentage points and the list is right. The difference comes almost entirely from the gaps. Both numbers are true; the only distinction is whether you could actually get filled. This morning's piece judged GNRC "watch only (wait for the give-back)" — it closed +18.34%, the single best performer of the day, and looks like the big fish that got away; but it opened at $229.50 and closed at $207.23, open→close −9.70%, and buying at the pre-market price and holding to the close was −9.11%. That judgment was right at the execution level.

4. Rank the 12 baskets and there is only one conclusion: today's ranking is not a stock-selection ranking, it is a gap ranking. Spearman(this morning's pre-market basket mean, closing basket mean) = **+0.650** (direction broadly carried through), Spearman(this morning's pre-market basket mean, same-day open→close) = **−0.615** (the harder it ran pre-market, the harder it gave back intraday). The power chain that led this morning was open→close −2.73%, homebuilders −1.42%, semicap −1.69%; the three baskets that fell most this morning (software SaaS −0.70%, energy E&P −0.94%, oilfield services/refiners −1.10%) were open→close +2.03%, +1.48%, and +1.94% respectively — all positive, and the top three of the day.

5. The index rose 1.14%, but the median rose only 0.18% — this was not a broad rally, it was a highly concentrated one, and the 81.2% up-breadth of this morning's pre-market did not materialize. My own name-by-name calculation for the S&P 500: 293 up / 207 down / 3 flat, 58.3% advancing, mean +0.48%, median +0.18%. Against this morning's pre-market 81.2% advancing and mean +1.23% across 469 valid samples — breadth narrowed sharply after the open. Information technology at equal weight, +1.81% (57/73), carried the index on its own, while communication services was −1.31% (only 5/24 up), the only sector to close down.

6. The one storyline this morning's piece did not cover at all was the weakest direction of the day: telecom. TMUS −5.57% at $166.45, just 0.5% above its 52-week low of $165.66; CMCSA −3.46%, VZ −2.87%, PSKY −4.63%, T −1.82%. On a day the Nasdaq was +1.69%, this was the collapse of an entire sector, and this morning's list did not say a word about it.

---

1. Market Overview

Indices and cross-asset (16:00 ET closing basis)

Curve shape: 2Y −6.1bp / 10Y −7.0bp / 30Y −6.0bp, with the 2s10s spread tightening marginally from 27.7bp to 26.8bp (−0.9bp). This was a near-parallel shift lower, not a steepening and not a bear flattening — the "rates have peaked" trade was voted on by all three tenors simultaneously today, rather than being priced by any single segment.

Market breadth (S&P 500, my own equal-weighted name-by-name calculation, n=503)

🔴 The index was +1.14% while the median was +0.18%, a gap of 0.96 percentage points — more than three-quarters of today's index gain came from the mega-cap weights, not from breadth. Against this morning's pre-market sample of 469 names at 81.2% advancing: well over half of the pre-market broad rally was cashed out after the open, which is the same thing as §0 item 4's Spearman(pre-market, open→close) = −0.615, stated two ways.

Sector equal-weighted performance (GICS, computed within S&P 500 constituents)

Two things that need to be said clearly:

1. Energy at +0.64% equal-weighted with 14/21 up, ranking fourth — this morning listed it as the one sector facing a total wipeout and to be avoided, and that did not hold at all on the day.
2. Industrials managed only +0.19%, 44/83 — and this morning's No. 1 theme (power/electromechanical) sits almost entirely inside industrials. The flatness of the industrials sector is itself sector-level evidence that the power chain stalled out.

✅ Filling the gap this morning's piece could not get: initial jobless claims (DOL official, primary source)

This morning's pre-market piece hit a 403 on the DOL page with FRED not yet updated, so it left this item blank and explicitly marked it "not obtained," rather than passing off last week's number. This piece successfully extracted it from the official DOL PDF:

Original text: "In the week ending September 12, the advance figure for seasonally adjusted initial claims was 196,000, a decrease of 10,000 from the previous week's unrevised level of 206,000." (DOL official weekly release¹)

🔴 The direction of this data point runs opposite to the "rates have peaked" trade: the labor market is tighter than expected. It did not stop yields from falling all day, but it corroborates, in the same direction, the SEP revising the 2026 unemployment rate down from 4.3% to 4.1% — the "hiking into a strong economy" framework picked up another data point today, and that means precisely that the probability of one more hike on 10/28 should not be marked down because of today's decline in yields.

---

2. Pre-Market Watchlist Reconciliation

2.1 The 5 names singled out in §9①

Hit rate (the five names in §9①):

• By same-day change: 4/5 up (80%), 4/5 ahead of the S&P's +1.14%.
• By "executed at this morning's pre-market price": 1/5 positive (20%), mean −2.83%.

2.2 Grouped reconciliation of all 20 names

Recommended group detail: CAT +2.02%, CMI −0.69%, GLW +2.52%, AMZN +2.13%, FIX −0.84%, NVT +2.49%, MU +5.50%, AMD +6.36%, LRCX +0.03%, DHI +1.49% Pass group detail: GNRC +18.34%, ETN +2.93%, VRT +0.87%, GEV −0.02%, PWR −0.44%, POWL −0.54%, CRDO +4.19%, ARM +8.57%, NBIS +4.12%, BE +3.98%

2.3 Avoid-list reconciliation

Energy avoid group (7 names): 7/7 up, mean +1.21%, 4 beating the S&P. This is the single clearest misjudgment of the day.

2.4 A one-paragraph self-criticism

There is only one pathology in this morning's pre-market list: it treated "the moves that had already happened this morning" as "the moves that would happen today."

The two correlation coefficients at the basket level say it all — Spearman(pre-market, close) = +0.650 shows the direction was broadly right, Spearman(pre-market, open→close) = −0.615 shows that from the moment of the open, the pre-market ranking was an inverse table. This morning's piece itself wrote that "a +30% gap has already eaten the odds," and that sentence was right about GNRC, but it never applied the same ruler to the whole table: the power-chain basket was +4.77% pre-market and open→close −2.73%, the same disease; the energy basket was −0.94% pre-market and open→close +1.48%, the mirror image of the same disease.

Three specific errors that can be fixed directly next time:

1. 🔴 The energy avoid call is a hard error, and its falsification condition was written in the very same piece. §8 item 3 said "If FANG/EOG/DVN turn positive in an environment where the broad market is +1%, it means shorts are starting to cover and the avoid logic must be reassessed." All three turned positive, the market was +1.14%, the condition was 100% triggered. The problem is not that it was unforeseen, but that the verification point was written in §8 and the conclusion was written in §9, with no binding relationship between them — the verification point fired, yet no mechanism let it modify the conclusion. Next time: for any judgment with a falsification condition attached, the "what the conclusion becomes once the condition triggers" must be written down alongside it.

2. 🔴 It treated "falling oil" as a constant for the day when it was an intraday variable. At 08:14 this morning it read WTI at −2.88%, and the whole piece's No. 1 theme, No. 2 theme, and avoid direction were all built on that reading. But it closed the day at only −0.51%, with the decline having narrowed by four-fifths. A variable with 7.5 hours of free movement left after the report was published was treated as bedrock. Next time: label commodity readings as "the pre-market value as of 08:14," and explicitly write "if the closing decline narrows to within −1%, the energy avoid call is void."

3. 🔴 "Can't find a catalyst" was once again treated as "there is no catalyst," and once again it was bet on the wrong side. ARM was judged "watch only" this morning on the grounds of "no single-stock catalyst, none found" — it closed +8.57%, the best in the Pass group. The actual catalyst was Arm's CEO publicly stating that day that AGI CPU demand could reach the $2 billion scale (invezz 09-17², the original was a 403 from this machine, so this relies on the search snippet and could not be verified against a primary source). ⚠️ But a word in fairness to this morning's piece: the remarks were made during the session, and at 08:24 publication time it was objectively impossible to see them. The real error was not missing the news, but writing "I couldn't find it just now" into a directional judgment for the whole day.

---

3. Today's Theme Verification

3.1 Basket measurements (constituents and counts identical to this morning's pre-market, all computed name by name at equal weight)

3.2 Pre-market vs. closing rank comparison

🔴 The bottom three pre-market (software, energy, oilfield services/refiners) all closed positive, with open→close readings of +2.03% / +1.48% / +1.94% respectively, sweeping the top three of the day. The No. 1 pre-market (the power chain) was open→close −2.73%, dead last.

3.3 What this morning got right and what it got wrong

✅ Got right:

• The AI compute/semiconductor direction was right. This morning listed it as the No. 3 theme; it was No. 1 on the day. The rank was underestimated, but the direction and the constituents were both right.
• GNRC's "the gap has already eaten the odds" was completely correct, and the $220 criterion hit precisely.
• FLNC's "don't touch it long, but it isn't a good short either" was right on both sides.
• The relative call that DHI beats LEN: DHI +1.49% vs. LEN +1.71% — ⚠️ strictly speaking this one was wrong, LEN was in fact slightly stronger, even though the reasoning given this morning (net cash vs. net debt, gross margin 20.7% vs. 15.8%) holds on fundamentals. The ranking variable of the day was not balance-sheet quality.
• The utilities and gold/silver miner baskets both had perfect breadth (12/12); this morning listed it only as the No. 5 theme, but the constituents were very well chosen.

❌ Got wrong:

• The No. 1 theme was the wrong pick. The power chain was the basket that gave back the most on the day, and it was ranked first and given an S-grade intensity.
• The energy avoid call was wrong (7/7 up).
• Software SaaS's "should not be folded into the duration-repair narrative" was wrong — its open→close of +2.03% was the best of the day. This morning's argument was "if this were really duration-driven, software, which should rise the most, would not be at the bottom," and that reasoning holds on the pre-market cross-section, but it used an 8 o'clock cross-section to negate an entire day.
• Large banks were called in the wrong direction: this morning listed it as the No. 6 theme (repair); it was actually +0.37%, second from last. "Peak rates → bank repair" did not work today, and the reason is exactly what this morning itself had written: the curve shifted down in parallel rather than steepening (2s10s in fact tightened by 0.9bp), so the NIM logic does not hold. This morning wrote that as a risk note but never turned it into a ranking input.

3.4 🔴 The day's strongest and weakest, both missed entirely this morning

① The strongest single stocks were not on the list: the "second-tier catch-up" in semis/AI hardware

Common feature: most of these names are "legacy hardware + storage + x86," not the "power/electromechanical" focus of this morning's list. INTC's catalysts were reported to be analyst price-target upgrades, another ~10% PC CPU price increase in October, and rumored cooperation with SK Hynix on memory manufacturing in the U.S. (⚠️ all from secondary financial sites, no primary source could be verified from this machine, so treated as "unconfirmed").

② The weakest was an entire sector, not a single stock: telecom

🔴 I could not find a primary-source catalyst dated 9/17. TMUS filed no 8-K that day (verified on EDGAR; the most recent is the 09-03 CFO transition), the Wolfe Research downgrade occurred in August, and the CFO transition announcement was 09-03 — neither of those is today's news. Volume was 6.58 million shares = 1.64× the 10-day average, so the volume expansion is real. Handled under "can't find news ≠ there is no news," and flagged as catalyst unconfirmed. The referenceable backdrop is the months-long wireless/broadband price war and share erosion of telecom by cable operators (Comcast/Charter) and satellite (Starlink), but using a months-long structural theme to explain a single-day −5.57% is over-attribution, and this piece will not write it that way.

---

4. Earnings and After-Hours Movers

4.1 During the regular session: no S&P 500 constituent reported pre-market or after the close today

Today was a trading day with no heavyweight earnings (LEN and FLNC, which reported after yesterday's close, were handled in this morning's list). All the action happened after the close, and it was concentrated in one sector: steel.

4.2 🔴 Two after-hours items: the two steel majors issued quarterly guidance on the same day, both below consensus

This is the most certain catalyst for tomorrow's open, and the direction is negative.

NUE primary source (PR Newswire, 09-17 16:30 ET³):

• Steel mill segment earnings are expected to improve, due to "higher average selling prices and stable volumes," partially offset by higher production costs.
• 🔴 One methodology adjustment you have to strip out yourself: the Q2 figure of $5.04 includes a $130 million cash refund that will not repeat; Q2 adjusted was $4.84. So the sequential comparison should be $5.55–5.65 against $4.84, which is +15%+17%, not the +10%+12% against $5.04. Using the unadjusted Q2 as the base understates the sequential improvement in this guidance.
• There is also "higher corporate/eliminations expense, reflecting the absence of the Helion valuation adjustment" — another one-time item disappearing.
• Raw materials segment earnings are expected to decline ("lower pricing and shipments").

How to read these two (three points):

1. The absolute level is strong; it is weak only relative to expectations. NUE's $5.55–5.65 is more than double Q3'25's $2.63; STLD's $5.34–5.38 is +52% versus its own Q2 guidance of $3.51–3.55. This is not deteriorating fundamentals, it is consensus running ahead of the companies. This is exactly the case of "guidance is the exam paper set by the party being graded, and consensus is a different paper" — both companies are in a cyclical upswing, but the sell side drew the slope steeper.
2. ⚠️ The credibility of the after-hours declines has to be discounted; neither the reading time nor the volume supports a strong conclusion. NUE's −2.40% was read at 16:41 ET, only 11 minutes after the release, on volume of 46,902 shares (≈$12 million notional); STLD's −3.40% was on 51,004 shares. Both are thin-volume early reads. A third-party source separately reported NUE trading as low as $254 after hours (about −4%), which does not match my measured $258.78 — I did not obtain a later reading to determine which is the final value, so both are listed and no choice is made between them.
3. Sector spillover did not happen today: CLF +0.51%, RS +0.06%, CMC flat, ATI −0.35%, none of them followed lower after hours. That suggests the market is reading it for now as an expectation gap at two companies rather than a peak in steel prices. Whether CLF/CMC/RS follow at tomorrow's open is the real test of this one.

⚠️ I did not obtain STLD's guidance from a primary source. As of publication, neither STLD nor NUE has a 09-17 8-K on EDGAR (verified; STLD's most recent 8-K is dated 08-14), STLD's IR page returns a 404, and stocktitan timed out at DNS. The $5.34–$5.38 and the $5.60 consensus come from secondary financial sites, are not primary-verified, and should be cited with that in mind. NUE's numbers, by contrast, have been verified against the PR Newswire original.

4.3 Other after-hours movers (all thin volume, for reference only)

---

5. Flows and Sentiment

5.1 Sector ETFs (16:00 ET close)

Leaders:

Laggards / decliners:

🔴 One important cross-check: USO −0.55% / BNO −1.30% match the two official settlement prices (−0.51% / −0.96%) in direction and are close in magnitude — which means today's "sharp narrowing of the crude decline" is not a problem with any single data source, but a fact confirmed jointly by four independent channels (WTI settlement, Brent settlement, USO, BNO). This morning's pre-market readings of −2.88% / −3.49% were true at the time but did not hold for the day.

5.2 Risk-on / risk-off characterization

Today was risk-on, but a "narrow" risk-on:

Supporting risk-on:

• VIX −12.82% to 15.44, even the high of the day at 16.29 never got back above the previous close of 17.71;
• ARKK +4.50%, TAN +3.90%, SMH +2.76% — high beta, long duration, and the speculative end led;
• HYG +0.38%, no stress in credit;
• Consumer staples was the only sector to close down (defensives were sold);
• Software SaaS's open→close of +2.03% was the best of the day, long-duration assets were bought back intraday.

Not supportive / requiring a discount:

• 🔴 The Russell 2000 closed at the low of the day, open→close −0.30% — small caps did not participate, which contradicts a "breadth-type risk-on";
• 🔴 Only 58.3% advancing and a median of just +0.18%, 0.96pp below the index's +1.14%;
• 🔴 Financials −0.09%, regional banks exactly flat — on a day when yields fell across the board, banks had no reaction at all;
• The Dow's open was the high of the day, large-cap blue chips were being distributed all day.

Conclusion: today's risk-on was concentrated in three lines — "AI hardware + precious metals + long-duration tech" — and was not a systemic lift in risk appetite. The index's gain was mainly contributed by the information technology sector (equal-weighted +1.81%, 57/73), while industrials, financials, and communications, 183 constituents combined, essentially did not participate (+0.19% / +0.18% / −1.31%).

5.3 Rates and volatility

• All three tenors shifted down near-parallel by 6–7bp, while the 2s10s actually tightened marginally by 0.9bp. This is the shape of "falling inflation expectations," not "falling growth expectations" — in a recession trade the front end would fall more and the curve would steepen.
• The 10Y's high was 5.008% and it closed at 4.934%, never back above 5.00% — the risk signal in item 12 of this morning's §8 ("if the 10Y breaks back above 5.00%") was not triggered, so DHI's verification point passed.
• ⚠️ VIX at 15.44 is still low. This morning's piece wrote that "this is not evidence that the market is calm, it is evidence that the tail is not priced," and that sentence still holds today — and tomorrow is quadruple witching.

5.4 ⚠️ Gold: a methodology conflict I cannot resolve with primary data, laid out as is

Three channels contradict each other, and this piece makes no single assertion:

Known facts: @GC.1 today opened 4,301.40 / high 4,423.30 / low 4,294.50, while its previous_day_closing field reads 4,387.50 — the open is 2.0% below the "previous close," and a third party likewise reported that "the December contract opened 2.0% below Wednesday's close." A −2% gap-down open that recovers to −0.14% does not necessarily contradict, in direction, the three "up" readings from GLD/spot/miners (futures and spot have different reference times, and contract rollover is possible). The only thing that can be stated with certainty is this: the precious-metals complex was strong today (GLD +1.69%, SLV +3.37%, miners 12/12 up); and @GC.1's single-day change is not cited in this piece because its baseline is in doubt.

---

6. Next-Day Outlook (2026-09-18, Friday)

① Theme continuation

② Tomorrow's earnings and macro calendar

The next real gate is not tomorrow: MU's earnings after the close on 09-30 (the falsification point for the storage basket's 4/4) and the 10-28 FOMC (the SEP median implies one more 25bp this year).

③ Names to watch (ticker + falsifiable verification point)

④ Directions to avoid

1. Do not make main-line judgments from a single day's prices on quadruple witching. This is tomorrow's biggest methodological risk, higher priority than any individual stock. This morning's piece already warned about it once, and tomorrow is the day it actually takes effect.
2. Do not treat today's after-hours steel declines as confirmed fact — both are thin-volume early reads (46,902 / 51,004 shares), and STLD's guidance figures were not primary-verified in this piece.
3. Do not chase a bounce in the power chain. It was the basket that gave back the most today (open→close −2.73%), and two of the three downgrade signals set this morning have already triggered (GNRC breaking $220, CMI failing to reclaim $537.9).
4. Do not buy banks just because yields fell today. Today already falsified that once, with XLF −0.09% / KRE 0.00%, and the curve shifted down in parallel rather than steepening.
5. Do not catch the falling knife in telecom; the catalyst is unconfirmed.

⑤ Input notes for tomorrow's pre-market list

1. 🔴 A "vs. pre-market price" column must be added. The core conclusion of today's reconciliation — the recommended group losing to the Pass group by 2.1pp on closing prices and winning by 2.15pp on pre-market prices — can only be revealed by that column. Reconciling on closing returns alone systematically overstates list quality.
2. 🔴 For every falsifiable criterion, write down "what the conclusion becomes once it triggers." Today at least three of §8's verification points judged correctly (GNRC $220, energy turning positive, CMI $537.9), but not one of them went back and modified the conclusions in §9. There must be a binding relationship between verification points and conclusions, otherwise the verification points are only decoration.
3. Label commodity/rate readings as "the pre-market value as of XX:XX," and write down how to handle it when it becomes invalid. Today the foundation of the entire piece was WTI at −2.88% at 08:14, and it closed at only −0.51%.
4. Do not use pre-market changes as the ranking variable for themes. Spearman(pre-market, open→close) = −0.615. The pre-market gain measures "how much has already happened," not "how much is still to come." Suggest switching to a "breadth + intraday acceleration" ranking: today AI compute (12/12 and open→close +0.89%) and the power chain (11/16 and open→close −2.73%) differed by only 1.03pp in their pre-market means, yet were worlds apart on these two dimensions.
5. Tomorrow's scan must set up "telecom sector" and "steel sector" as two separate mandatory items, the former being today's weakest with an unclear catalyst, the latter having a clear after-hours catalyst awaiting verification.
6. Add one scanning dimension that was missed today: legacy hardware/x86/storage (INTC, HPE, DELL, SMCI, SNDK, STX, WDC). 7 of today's top 13 gainers belong to this cluster, and not one of them was in any of this morning's baskets.

---

⚙️ Data Sourcing and Operations Log for This Issue (internal)

yfinance channel: rate-limited by Yahoo for the fourth consecutive trading day. yf.Ticker('AAPL').history() throws YFRateLimitError: Too Many Requests directly, with the stack hanging at the crumb-fetching stage in _get_cookie_and_crumb_basic. Not a single number in this piece comes from yfinance (the skill parameter line saying "pull data with WebSearch+yfinance" could not be executed literally; everything was rerouted through CNBC).

Channels actually used (all verified working):

• CNBC restQuote (quote.cnbc.com/quote-html-webservice/restQuote/symbolType/symbol) — the workhorse for this piece. Up to 40 names per batch, returning open/high/low/last/previous_day_closing/volume/tendayavgvol/pcttendayvol/yrhiprice/yrloprice/fpe/ExtendedMktQuote/EventData. All 578 names requested returned, zero missing (name-by-name validation plus single-name retry fallback were applied per the "batch quotes silently drop samples" lesson; no retry was triggered this time).
• SEC EDGAR submissions JSON (CIK zero-padded to 10 digits) — used to falsify whether STLD/NUE/TMUS had an 8-K today; the conclusion is that none of the three did.
• DOL official PDF dol.gov/ui/data.pdf — WebFetch parsed it as empty (binary), but it did save the file to disk, and local pdftotext -layout extraction then succeeded. This morning's pre-market piece gave up at this step and this piece filled it in — this path can be reused directly going forward.
• PR Newswire — primary source for NUE's guidance.
• stockanalysis.com — used for independent cross-verification of TMUS's closing price plus the STLD/NUE consensus. Accessible this time, not blocked by Cloudflare.

Channels that failed or were restricted in testing:

• yfinance / Yahoo — rate-limited (see above).
• stocktitan.net — DNS resolution timeout (getaddrinfo ETIMEOUT), consistent with the historical record; this channel can be struck off the backup list.
• invezz.com 403, ir.steeldynamics.com/news/ 404.
• FRED ICSA CSV read timeout (fredgraph.csv?id=ICSA, no response in 40s) — but the DOL primary source was obtained, so the conclusion is unaffected.
• WebSearch was nearly useless for generic queries about the "9/17 close", returning articles from the same date in prior years or from other recent days. The effective query pattern is "company name + specific date + event type"; a generic search for "stock market close September 17 2026" returned articles from July and 9/8.

🔴 Three traps caught this issue (worth remembering):

1. TMUS had two mutually contradictory third-party readings. tradingkey reported "−3.09%, intraday low $167.77," while CNBC gave −5.57% and a close of $166.45. Independent cross-checking with stockanalysis confirmed CNBC was right (close 166.45, previous close 176.26), and the tradingkey figure was wrong — and since its "intraday low of 167.77" is above the true closing price, it is internally inconsistent on its own, which is a signal that can be caught on the spot.
2. CNBC's previous_day_closing can give a value that badly conflicts with open for individual instruments. @GC.1 had a previous close of 4,387.50 against an open of 4,301.40 (−2.0%), producing a three-way fight between "gold −0.14%," GLD +1.69%, and miners 12/12 up. Handled as "no single assertion, all three readings listed," with no forced pick. (Also: CRH/CSX/KHC/KRE all four returned UNCH; name-by-name re-checking confirmed they were genuinely exactly flat with normal volume, not a broken field.)
3. CNBC's US2Y change_pct field is broken for the second consecutive day — today it gives +0.1172%, while last 4.666 / previous close 4.727 back-compute to −6.1bp. Even the sign is inverted. Consistent with this morning's piece, this one uses the back-computed value throughout. This field should be permanently blacklisted.

⚠️ Three items in this piece that could not be primary-verified and are flagged as such in the body:

1. STLD's Q3 guidance of $5.34–$5.38 and the $5.60 consensus — secondary reports only; no 8-K on EDGAR, a 404 on the company site, stocktitan DNS down. The body explicitly marks it "not primary-verified."
2. ARM's CEO remarks — invezz 403, search snippet only. Flagged in the body.
3. INTC's three catalysts (price-target upgrades / October CPU price increase / SK Hynix cooperation) — all from stockstotrade, timothysykes, and similar low-quality recurring-column sites, which have a track record of "mass-republishing the same headline under new dates." The body treats them as "unconfirmed" and does not write them as fact.

⚠️ One methodological self-check: the §0 item 3 conclusion that "the recommended group beats the Pass group on pre-market prices" rests on a sample of only n=8 vs. n=10, and GNRC alone contributes −9.11% of the Pass group's −2.61%. Excluding GNRC, the Pass group is −1.89%, so the direction of the conclusion is unchanged but the magnitude narrows from 2.15pp to 1.43pp. The conclusion is confirmed not to depend on a single sample, but its strength should be understood at that smaller number.

Sub-agents: not invoked this issue (the user's instructions explicitly required not using the Agent tool). This piece therefore did not go through risk-auditor quality control, nor did it get an independent valuation check from fundamentals-analyst. Historically that step typically overturns 3–6 judgments — all self-checking in this piece was done by the main loop, and its credibility should be discounted as "not reviewed by a second pair of eyes."

---

⚠️ Risk disclaimer: this recap is post-close information review and observation only, and does not constitute investment advice. Data may differ in timeliness or methodology; please refer to company disclosures/SEC filings as the authority. It must not be used directly as a basis for trading.