US Market Recap | 2026-09-07 (ET) Monday

⚠️ This is a holiday-edition recap: there was no close to recap today.

2026-09-07 was US Labor Day. The NYSE, Nasdaq and the bond market were closed all day, there was no 09:30 open and no 16:00 close, and zero volume traded all day. The last trade dates this machine read for .SPX / .DJI / .IXIC were all 2026-09-04, which is machine evidence of "no trading today", not a data-retrieval failure.

This issue therefore does three things instead of a normal recap:

1. Reconciliation cannot be performed, and is deferred until after the 9/8 close. Section 2 only locks down the baseline and explains why a hit rate cannot be computed.

2. It sweeps the roughly 25-hour coverage blind spot the pre-market list flagged for itself (9/7 08:15 ET → 9/7 market close): today's European close, the stretch of Labor Day futures trading through the 13:00 ET early close, and Monday's corporate and geopolitical news. This is the main output of this issue.

3. It closes out the to-do item the pre-market list left behind: SKHY's ADS ratio and the two-market price gap have now been resolved with primary documents, and the conclusion is different from the pre-market guess (see Section 4.1).

• Coverage window: 2026-09-07 08:15 ET → 2026-09-07 17:00 ET
• Quote conventions: CNBC quote API (futures are readings at their respective settlement times, all cross-checked by reverse-computing against the prior close); SK Hynix's Seoul price is cross-checked across two sources, stockanalysis.com and investing.com; the ADS ratio is taken from the primary SEC 424B4
• This issue provides no single-stock same-day percentage changes — that number does not exist today

---

0. One-Sentence Recap

1. There was no session today, so all of this issue's information content comes from "that holiday futures half-session". And it delivered one clear directional reading: the "Nasdaq strong, Dow weak" structure of 9/4 did not converge during the holiday half-session — it widened out again. S&P 500 futures −0.17%, Nasdaq 100 futures +0.13% (turning positive), Dow futures −0.62%. The Nasdaq-minus-Dow spread widened from 0.45pp at 08:15 this morning to 0.75pp at the close, landing exactly back on the 0.72pp of the 9/4 cash session.
2. And once again it happened against a backdrop of rising yields. 10-year Treasury futures −0.10%, 30-year −0.17% (price down = yield up). That makes three consecutive observations: semiconductors / AI hardware strengthening on days when rates rise. This reinforces the judgment in Section 1.1③ of the pre-market list — the driver is a sector-specific event, not a "rate-driven high-beta squeeze".
3. But its information content has to be discounted first: the holiday half-session was extremely thin. S&P futures 137,293 contracts, Nasdaq futures 88,329 contracts, Dow futures only 10,972 contracts. On that kind of volume, a ±0.2% direction is not enough to be upgraded into a trend call — it is a hint, not a vote.
4. Today's big Asian rally was "storage / semiconductor specific", not global risk-on — and there is clean cross-sectional evidence for that. Korea +4.61%, Nikkei +2.12%, while Hang Seng −0.93% and the Shanghai Composite +0.07%; Europe was essentially flat (FTSE 100 −0.08%, DAX −0.15%, CAC +0.33%, Euro Stoxx 50 +0.17%). The only markets that rallied were the two with the highest memory weighting. If this were a macro-level revival in risk appetite, Hong Kong and Europe would not have stood still.
5. The pre-market list was right to reject that VIX reading of 15.25, and today produced better substitute evidence. Spot VIX printed 15.30 (+5.30%), but the VIX futures that actually traded over the same stretch (September contract) were up only +0.20%, on 3,818 contracts. A spot reading of "volatility jumping 5.3%" paired with a futures contract that did not budge is internally incompatible; and the futures are the one backed by actual volume. This issue's volatility baseline therefore remains the 9/4 close of 14.53.
6. SKHY, which the pre-market list flagged for verification, was resolved today — and the pre-market guess was wrong. The ADS ratio was not a citation error — the SEC 424B4 states in black and white that "each ADS represents one-tenth of a share of common stock". That 34% gap is a genuine structural premium, caused by the Korea Securities Depository's 2.5% conversion quota cap, which was already fully used up at issuance, physically sealing off the arbitrage channel. See Section 4.1 — this changes the usability conclusion on SKHY.
7. Crude continued to strengthen but has not reached the verification threshold. Brent $97.27 (+1.03%), WTI $92.70 (+1.33%); on Monday another tanker in the southbound Strait of Hormuz lane was struck by 3 unidentified projectiles (UKMTO). The falsification line set for XOM in the pre-market list was "Brent holding above $100", which has not been triggered.

In one sentence: the holiday half-session ran through the 9/4 structure one more time — AI hardware strong, large-cap value weak, yields marginally higher, volatility unchanged; but volume was too thin, so it counts only as a hint. The 9/8 open has to digest roughly 97 hours of news in one go, and the real verdicts come from ORCL and ADBE on 9/10 and the CPI on 9/11; until then, the pre-market list's stance of "downgrade rather than chase when attribution is unclear" has not encountered a single piece of evidence today that would require overturning it.

---

1. Market Overview (Holiday Half-Session)

1.1 US Equity Futures (9/7, 13:00 ET early close)

Convention note: the prior settlement is the 9/4 futures settlement price, which carries a normal basis versus the cash index closes (SPX 7,718.60 / NDX 29,544.155 / DJI 53,414.25); the two cannot be mixed in a calculation. The S&P futures' −0.17% has been given as exactly the same reading by two independent sources, CNBC and investing.com (7,708.75 / −13.25 / −0.17%, 12:59:59), so it can be relied on.

1.2 The Only Genuine Structural Signal Today: the Nasdaq–Dow Spread Widening Again

How to read it: when the pre-market list was finalized this morning, this spread had already compressed to 0.45pp, and at the time it looked as though the 9/4 divergence was fading; the holiday half-session pulled it back out to 0.75pp, and it did so via both legs at once — Nasdaq turning positive plus the Dow weakening — not one-sided.

⚠️ But three limitations have to be stated alongside it, or it will be read as carrying too much weight:

1. Volume was too thin — Dow futures traded only 10,972 contracts all day, and at that level a meaningful part of the spread is liquidity noise.
2. The risk signal set in Section 8.4 of the pre-market list was "if Tuesday reverses into Dow strong, Nasdaq weak, that means the rate-hike trade has overpowered the AI trade". Today went the opposite way, so that signal was not triggered; but "not triggered" is not the same as "disproven" — its verdict point is the 9/8 cash session, not holiday futures.
3. This is not rate-driven. Treasury futures fell in step (yields marginally higher), and the Nasdaq turned positive while yields were rising — the same structure as 9/4.

1.3 Rates, FX, Volatility, Commodities

The honest cost on yields: the bond market was closed, so there is no official closing yield for 9/7. The "+1–2bp" in the table above is estimated from the change in 10-year Treasury futures prices, it is not a Treasury Department reading, and it must not be placed in the same table as the Treasury CSV numbers used in the pre-market list for comparison. The next comparable official reading only arrives after the 9/8 close.

On that VIX print of 15.30 — how this issue handles it, and why:

The pre-market list rejected this morning's 15.25 reading on the grounds that "Cboe was closed, SPX options were not trading, and spot VIX had no calculation inputs". After the close today we obtained a piece of evidence the pre-market list did not have, and it supports that rejection — and is more useful than the rejection alone:

The two stopped updating in the same minute, yet gave incompatible directions: spot says the volatility expectation jumped 5.3%, while the futures, where people actually voted with money, moved only 0.20%. Spot VIX's calculation inputs (SPX option quotes) simply did not exist today, whereas the futures' 3,818 contracts of volume are real.

• This issue's conclusion: the volatility anchor remains the 9/4 close of 14.53; the credible reading from the holiday half-session is "volatility expectations essentially unchanged" (VIX futures +0.20%).
• The 9/8 verification: if spot VIX comes back to around 15 after the 9/8 open with a normal intraday range, that means 15.30 broadly reflected real pricing; if it drops straight back to the 14.5 line at the open, then 15.30 was an interpolation with no inputs and should be discarded entirely. This one can be settled on 9/8.

1.4 Overnight and Today's Overseas Markets: Only the Two With the Highest Memory Weighting Rallied

This table is the cleanest piece of cross-sectional evidence in this issue. Korea +4.61%, Japan +2.12%, while Hong Kong −0.93%, China +0.07%, and all six European readings land inside a narrow −0.15% to +0.33% band.

Implication: today's Asian strength is a "storage / semiconductor theme", not "a revival in global risk appetite". If it were the latter, Hong Kong and Europe could not have stood perfectly still. Korea's +4.61% therefore cannot be used as a reason for a broad higher open in US equities on 9/8 — it maps only onto the memory chain, and Section 8.1 of the pre-market list already warned that this portion is an overnight handoff, not incremental buying.

---

2. Pre-Market List Reconciliation — Not Possible This Issue; Baseline Locked

There was zero volume today, so not a single stock in the pre-market list has a "same-day change%", the hit rate cannot be computed, and this issue gives no hit-rate number. Forcing in the 9/4 close or futures as substitutes would only manufacture a table that looks like reconciliation but reconciles nothing.

The baseline is locked as follows (on a 9/4 close basis), for a single reconciliation after the 9/8 close:

⚠️ Convention requirement for the 9/8 recap (to avoid systematically overstating the hit rate): 9/8 is the first trading day after the holiday, and gaps will be materially larger than on an ordinary session. Reconciliation must give two columns, "closing change%" and "relative to the 9/8 open" — using the closing change alone would record "gapped up then bled all day, and executing the list was actually a loss" as a hit.

2.1 The Part That Can Be Verified: the Pre-Market List's Methodological Claims (Not Dependent on US Prices)

Several conclusions in the pre-market list can be verified without a US open, and each gets a verdict today:

One line of self-criticism: the pre-market list attributed SKHY's price gap to a "citation error" and excluded it from the list on that basis — the act of excluding it was right, but the reason was wrong. The correct reason (the arbitrage channel being sealed off by the 2.5% conversion cap) is not only more solid, it points to a completely different trading implication: this is not a data problem, it is a premium that can persist for a long time and that itself swings widely. Treating it as "data of questionable quality" leaves the impression that once verified it could be used; in fact, once verified it is even less usable as a proxy for Seoul.

---

3. Today's Theme Verification — No Same-Day Prices, Only the "Overseas + Futures" Layer Can Be Verified

Did the pre-market list identify the strongest theme correctly? — Not determinable this issue. The memory chain was indeed the strongest overseas (Korea's +4.61% was the strongest of any major market globally), which is consistent with the pre-market list ranking storage as the main line; but the pre-market list's core stance was "attribution unclear, downgrade rather than chase", and that stance needs US price and volume to verify it, which today did not provide.

Was there a surprise theme the pre-market list missed? — There is one worth recording:

Seoul Economic Daily reported today that Samsung's and SK Hynix's memory inventories have fallen below 10 days; KB Securities said next year's memory market will be "the tightest supply in history", with DRAM and NAND demand exceeding supply by more than 10%.

⚠️ But it has to be given the right evidence grade, and cannot be used to settle a verification point: this is a sell-side view plus a local media relay, not a contract price, not a company disclosure, not a primary filing. The verification point set for SNDK in the pre-market list explicitly demands "NAND spot / contract prices" (TrendForce / DRAMeXchange basis), and this does not meet that requirement, so it cannot be used to declare the verification point passed. Its correct use is: it shows that Korea's 4.61% today has an identifiable same-day news source — and that is exactly what was missing when the pre-market list said "the 9/4 up-candle has unclear attribution". What 9/4 lacked as a driver, 9/7 found one of in Korea; but it explains Korea, not the 9/4 US session.

---

4. The Substantive Output of This Issue

4.1 SKHY Delivered: the ADS Ratio Was Not Wrong, the 34% Premium Is Real, and Arbitrage Is Sealed Off Institutionally

The pre-market list listed this as "the first item of business next time we start work", and today it was resolved with primary documents, with a conclusion opposite to the pre-market guess.

Step one: the ratio (SEC 424B4, the 2026-07-10 final prospectus, primary text)

"We are offering 177,900,000 ADSs. Each ADS represents one-tenth of a share of our common stock, par value W 5,000 per share... The initial public offering price of the ADSs is US$149.00 per ADS."

Ratio = 1 ADS : 1/10 common share, confirmed correct. 177.9 million ADSs offered, offering price $149.00, 728,865,500 common shares outstanding after the offering. The pre-market suspicion that "the ratio was misquoted" does not hold.

Step two: so how big is the gap? — reverse-compute it ourselves

Note the relationship between these two numbers, which is itself a piece of information: SKHY did not move at all (closed today), and purely because Seoul rose 7.29% on Monday, the premium automatically compressed from 44.5% to 34.7%. Seoul is chasing that gap from below.

Step three: the cause (a layer the pre-market list never considered)

Multiple reports point to the same mechanism: the Korea Securities Depository (KSD) caps the quota for converting common shares into ADRs at 2.5% of total shares outstanding, and that quota was entirely used up at issuance. This means:

• No new ADRs can be created — unless existing ADR holders first convert back into Korean common shares, freeing up quota;
• therefore the standard arbitrage path of "buy in Seoul, convert to ADR, sell in New York" is physically sealed off;
• the result is that the ADR can sustain a premium over Seoul indefinitely, having historically reached roughly 51% at the peak and about 33% in late July.

This premium is known as the "reverse kimchi premium"; it is not a quoting error, it is the equilibrium outcome of an institutional constraint.

Step four: trading implications (completely different from the pre-market "data questionable" treatment)

1. SKHY cannot be used as a proxy for SK Hynix. Buying SKHY to express "Korean memory upcycle" means paying a 35% premium on top of the underlying, and that premium itself swings widely between 33% and 51% — the amplitude of the premium's swings is enough to drown out whether the fundamental call was right or wrong.
2. The specific tension on 9/8 can be written as a falsifiable verification point: Seoul's Monday +7.29% has not yet been reflected in SKHY.
   • If SKHY opens around $190 (= $177 × 1.0729), the premium has been fully maintained, and money is willing to keep paying for the closed arbitrage channel;
   • if it opens at $177–$185, the premium is compressing, and Seoul's gain has only been partially transmitted;
   • note that $190 is already close to its 52-week high of $194.80, so the room to chase is limited to begin with.
3. The decision "not to put SKHY on the list" was proven right today — but the reason has to be swapped out: not "the data has not been checked", but "it has been checked, and it is a premium trade, not a memory trade".

⚠️ One source conflict that must be disclosed: the two sources disagree on Seoul's 9/7 close — stockanalysis gives ₩1,767,000 (+7.29%), investing.com gives ₩1,783,000 (+8.26%, intraday range 1,733,000–1,783,000), a difference of about 0.9pp. This issue uses the more conservative ₩1,767,000 (it is arithmetically self-consistent with that source's own reported market cap of ₩1,299.57 trillion / 728.87 million shares). Using investing.com's number, the premium is +33.5%, which changes none of the conclusions.

4.2 What Actually Happened Inside That 25-Hour Pre-Market Blind Spot

Section 8.1④ of the pre-market list flagged it itself: "data is as of 9/7 08:15 ET, with roughly 25 hours after that uncovered, to be swept before the open." Item-by-item sweep results:

Sweep conclusion: nothing occurred in those 25 hours that requires overturning the pre-market list. The pre-market core stance (attribution unclear → downgrade rather than chase; refusing to interpret the 9/4 up-candle as a leadership rotation) still holds after the sweep, and gained two new pieces of corroborating evidence (Europe flat, and the Nasdaq turning positive while yields rose).

---

5. After-Hours Earnings Moves

No US earnings today — markets were closed, so there is no after-hours session.

The next after-hours window is after the close on 9/8 (Tuesday):

Three conventions on GME that must be separated out (one of which is new information the pre-market list did not have):

1. Revenue is declining year over year: $780–800 million vs $972.2 million in the year-ago quarter, i.e. −17.7% to −19.8%.
2. The vast majority of net profit does not come from operations: roughly $238 million of net profit comes from eBay derivative and equity investment gains, or 79.3% of the midpoint of net profit ($300 million).
3. ⚠️ One item the pre-market list did not cover: there is also roughly $75 million of digital-asset and related receivable losses offsetting in the same period. That means the "$238 million of investment gains" is not a net figure — restoring both non-operating items together, the actual scale of operating profit is far smaller than headline net profit.
4. Limited room for surprise: the company pre-disclosed the key metrics on 8/31, sharply compressing the probability of a major beat or miss; the ±9% implied move comes mainly from positioning structure, not from informational uncertainty.

Conclusion holds at the pre-market judgment: avoid GME. The reason is unchanged — the "surge in net profit" is investment mark-to-market, not operational improvement, while revenue is contracting at close to 20%.

---

6. Flows and Sentiment

There are no sector ETF readings today (XLK/XLE/IGV/XLE/XLF/XLV all had zero volume), so this section provides no sector-rotation table. Only three sentiment signals are available:

Characterization: a holiday session with "no panic, but no breadth either".

• One internal contradiction worth noting: the geopolitical event is escalating (another tanker attacked on Monday, Brent +1.03%), yet gold fell 0.55%, the dollar fell 0.26%, and Treasuries fell — the haven assets collectively refused to cooperate. This says the market treats the Strait of Hormuz as "an energy supply problem", not "a systemic risk event". That is consistent with the judgment in Section 5.6 of the pre-market list: rising oil is an event-driven hedge position, not a switch for macro risk.
• One counterpoint that cannot be ignored: the VIX futures basis (September futures 16.30 vs the spot anchor of 14.53), a positive spread of roughly 1.77 points, is normal contango, with no pre-event curve inversion — meaning the market has not priced in a meaningful risk premium for the 9/11 CPI or the 9/15–16 FOMC. The pre-market list said "14.53 is at a low level, the market has not priced in much risk premium for this week", and today's futures reading supports that.

---

7. Next-Day Outlook (for Tuesday 2026-09-08)

① Theme Continuity

② Tomorrow's Earnings and Macro Calendar (Tuesday 9/8, ET)

⚠️ One source conflict: on the timing of the 3-year auction, the eOption weekly calendar gives 13:00 ET, investing.com's calendar gives 12:00 ET. US Treasury auction convention is 13:00 ET, so this issue uses 13:00 and flags the discrepancy.

Rest of this week (unchanged): 9/9 the 10-year auction; 9/10 PPI + the 30-year auction + ORCL and ADBE earnings after the close + TSMC August revenue; 9/11 08:30 CPI; 9/15–16 FOMC; 9/21 index changes take effect; 9/30 MU earnings.

③ Focus List (Ticker + Falsifiable Verification Point)

Three more tape criteria to read right at the 9/8 open (carrying over pre-market Section 8.2):

1. Whether SNDK and the other three new S&P 100 additions (DELL/PANW/ANET) move in the same direction. Same direction → still index money; SNDK alone rallying → the market is trading NAND fundamentals. This one cleanly separates the two drivers.
2. Volume and price change must be read together, and the baseline must be stated. On 9/4 LRCX was 1.27x on the 10-day average volume and 0.79x on the 50-day average — the two conventions cannot be mixed, and the basis must be written out when cited.
3. Pre-market liquidity is always measured in notional value, never share count. MU at $1,016 and SNDK at $1,740 mean a small share count is already a large amount of money.

④ What to Avoid

1. Taking Korea's +4.61% directly as a reason for a higher US open — Hong Kong −0.93% and Europe flat across the board already prove that was a storage-specific move, not a revival in global risk appetite.
2. Treating ±0.2% in holiday futures as a trend — Dow futures traded only 10,972 contracts all day, and direction on that volume is noise.
3. Using SKHY to express the Korean memory upcycle — you would pay an extra 34.7% premium, and that premium has a historical range of 33% to 51%, an amplitude big enough to drown out the fundamental call.
4. High-beta interconnect names bouncing without volume (MRVL / ALAB / CRDO) — normalized P/E around 190–200x, and CRDO already shows negative operating leverage.
5. Chasing semiconductor equipment names (KLAC / AMAT) — P/E at the 92% / 89% percentile of the past 4 years with earnings at record highs, and 5 of the last 5 earnings reports over the past 6 weeks were sell-the-news.
6. Treating a "7x forward P/E" as cheap and buying it (MU) — that is peak-earnings pricing; on a realized basis it is 22.94x, the 86.7 percentile over 5 years.
7. The optically cheap WDC — 70% of TTM net profit is the already-liquidated SanDisk mark-to-market gain; on a consistent basis the true P/E is about 45.7x.
8. Mistaking investment gains for operational improvement (GME) — $238 million of eBay mark-to-market gains, plus roughly $75 million of digital-asset losses, while revenue is −17.7% to −19.8% year over year.
9. The three names removed from the S&P 100 (NKE / SPG / CL) — mechanical selling pressure ahead of 9/21.

⑤ Input Notes for the 9/8 Pre-Market List

1. This issue's futures, overseas-market and VIX futures readings can be carried over directly, but must be tagged "thin holiday volume" and cannot be used as normal-session price/volume evidence.
2. One SKHY item is settled: the 1:10 ratio is confirmed by the 424B4, and the cause of the price gap is the 2.5% conversion quota cap. The next issue can cite this directly without re-checking; but classify it as a "premium trade" and do not put it in the beneficiary table for the memory theme.
3. The three highest-priority items still unsettled (all carried over from the pre-market list):
   • NAND spot / contract prices (TrendForce / DRAMeXchange, weekly) — this is SNDK's only pillar, and there is still only one observation to date;
   • KLA's storage vs foundry end-market split — it determines whether KLAC is memory beta or foundry beta;
   • AMAT's NAND equipment share next quarter (around mid-November) — "DRAM is expanding, NAND is not" currently rests on a single observation.
4. One new item to verify: the Goldman Sachs Communacopia conference opens on 9/8, with NVDA on the roster. We need to confirm its speaking slot and whether there is any new capex or product guidance — this is a same-day potential catalyst the pre-market list did not flag.
5. The 9/8 reconciliation must add a "relative to the open" column (rationale in the Section 2 convention requirement).

⑥ Final One-Sentence Judgment

There was no session today, so today produces no "verification" — but it produces one exclusion. The holiday half-session ran through the 9/4 structure one more time (Nasdaq strong, Dow weak, yields marginally up, volatility unchanged), and Europe going flat plus Hong Kong falling cleanly excluded the "global risk appetite is reviving" explanation: the only markets that rallied were Korea and Japan, the two with the highest memory weighting. That narrows the attribution range for that unexplained 9/4 up-candle by one notch, but it still has not been explained. At the same time, the biggest to-do the pre-market list left behind was delivered, and the delivery overturned its own guess — SKHY's 34% gap is not a data error, it is an institutional premium locked in by a 2.5% conversion quota. The posture for 9/8 is therefore exactly the same as pre-market: do not chase, demand volume, wait for ORCL and ADBE on 9/10, wait for the CPI on 9/11. Any decision to add risk made on thin holiday futures is using noise as signal.

---

Data Retrieval and Pipeline Log (Internal)

1. Scheduling problem with this slot (needs a decision)

• This run's task was a "US market recap", but 9/7 was Labor Day, US markets were closed all day, and there was no after-hours session. The scheduler still fired this slot on its normal cadence.
• Recommendation: add a US trading-calendar precheck (NYSE holiday calendar) to the us slots in bin/generate.sh, so that on closed days it either skips outright or automatically switches to a "holiday edition" prompt, avoiding one manually rewritten draft for each of the 9 holidays a year. This issue is the product of a manual rewrite.
• Same class of risk: 11/26 Thanksgiving (closed all day), 11/27 and 12/24 (13:00 early close, there is an after-hours session but only half a day), 7/3, 1/1, 1/19, 2/16, 4/3, 5/25, 6/19, 7/4. Early-close days are especially dangerous — there is a closing price that day, so the draft will look entirely normal, but the convention is a half session.

2. Retrieval channels

• This issue used no yfinance (per the existing conclusion, this machine's IP is still on Yahoo's rate-limit list; no retry was attempted, to save time).
• Primary channel: CNBC quote.cnbc.com/quote-html-webservice/restQuote/..., root keys FormattedQuoteResult → FormattedQuote, exthrs=1. Every reading was cross-checked by reverse-computing the percentage change from last and previous_day_closing, with no discrepancies anywhere.
• ⚠️ New finding: CNBC's E-mini futures symbols are @SP.1 / @ND.1 / @DJ.1, not @ES.1 / @NQ.1 / @YM.1. The latter return all-null fields rather than an error — a textbook silent failure, and using them would be misread as "no futures data available". Tested: @ES.1, @ES.26U, @ES.1:CME, ES.1, @ES, @EMD.1, @RT.1 all return null. Russell 2000 futures were not obtained this time (@RT.1 is invalid and no substitute symbol was hunted down), so Section 1 has no Russell reading.
• Other valid symbols (tested this time): @TY.1 (10Y), @US.1 (30Y), @VX.1 (VIX futures), @CL.1/@BZ.1/@GC.1/@SI.1/@HG.1/@NG.1, @DX.1, .DXY, KRW=, EUR=, .IXIC (note: .COMP returns null, the Nasdaq Composite must use .IXIC).
• The S&P futures were independently cross-checked against investing.com, and the two sources gave exactly the same 7,708.75 / −13.25 / −0.17% / 12:59:59, so it can be relied on.
• WebFetch on cnbc.com/quotes/@ES.1 returns HTTP 403 — the CNBC web pages are unusable via WebFetch, though its quote API works. Do not use WebFetch on CNBC pages in future.
• Holiday close times as tested (consistent across product lines, safe to hard-code): equity index futures and Treasury futures 13:00 ET; energy/metals (NYMEX/COMEX) 14:30 ET; VIX futures 11:30 ET; ICE dollar index 16:51 ET; .DXY spot 17:00 ET.

3. USD/JPY field corruption (hit this time)

• JPY= returns last 154.35 and previous_day_closing 154.35, but change −1.89 / change_pct −1.2245% — last and prev are equal yet it gives a non-zero change, which is self-contradictory. Same family as the existing "CNBC Treasury change_pct single-field corruption". This issue therefore used no yen reading, and Section 1.3 lists only EUR/USD and USD/KRW.

4. Spot VIX validity (not fully settled)

• .VIX gives 15.30, intraday 14.99–15.32, timestamp 9/7 11:30:01 ET, which looks like a normal intraday series, not obviously dead data like _DJI/_COMP.
• But it is directionally incompatible with the VIX futures over the same stretch (+0.20%, 3,818 contracts), and spot VIX's calculation inputs (SPX option quotes) do not exist on a closed day.
• How this issue handled it: the main text treats only the futures as a credible reading, tags spot as "questionable", and gives the method for settling it on 9/8. We could not confirm whether Cboe ran a shortened session today — @VX.1 having 3,818 contracts of real volume shows CFE did trade, which is in tension with the general statement that "Cboe was fully closed". Next time we start work, check Cboe's official holiday trading schedule to settle this; it determines whether future holiday slots should trust spot VIX.

5. SKHY verification path (recommend hard-coding)

• EDGAR submissions JSON: https://data.sec.gov/submissions/CIK0002120882.json (here the CIK must be zero-padded to 10 digits), from which we get the 424B4 accession 0001193125-26-299963.

• Fetching the text: https://www.sec.gov/Archives/edgar/data/2120882/000119312526299963/d32785d424b4.htm(Archives path: the CIK must have its leading zeros stripped, otherwise you get HTTP 200 but 0 bytes). The file is 5.85MB, WebFetch is unsuitable; use curl + a regex to strip tags, then grep Each ADS represents.
• In the F-1/A that field is left blank ("Each ADS represents ___th of a common share") — only the final 424B4 carries the definitive value. This is a general trap: the A-round prospectus leaves it blank, WebFetch faithfully reports "the ratio is not disclosed", and that is easily misread as "cannot be found".
• The pre-market list's guess ("the ADS ratio was misquoted") has been disproven; if MEMORY has a related entry it needs updating.

6. Source conflicts (handled on a conservative / self-consistent basis)

• SK Hynix Seoul close on 9/7: stockanalysis ₩1,767,000 (+7.29%) vs investing.com ₩1,783,000 (+8.26%). We used the former — it is arithmetically self-consistent with that source's own reported market cap of ₩1,299.57 trillion ÷ 728.87 million shares. That investing.com page also lists "previous close" as 1,783,000 (identical to last), so the page's fields are themselves faulty.
• 3-year Treasury auction time: eOption 13:00 ET vs investing.com 12:00 ET. We used 13:00 (Treasury auction convention) and flagged the discrepancy in the main text.
• .KS11 returns UNCH for change, with prev equal to last (6,995.39) — CNBC has already rolled the 9/7 close into prev. The +4.61% is taken from the already-verified reading in the pre-market list; CNBC's UNCH was not adopted a second time.

7. Not obtained / left blank in this issue

• Russell 2000 futures (symbol not found, see item 2).
• The official 9/7 Treasury closing yields (the bond market was closed, so they objectively do not exist; the main text used a futures-based estimate and explicitly flagged that the conventions cannot be mixed).
• A new reading on NAND spot / contract prices (TrendForce is weekly/monthly, no update today) — this is the second consecutive issue in which the SNDK verification point goes unsettled, and it is listed as top priority in Section 7⑤ of the main text.
• The full agenda for Goldman Sachs Communacopia and NVDA's speaking slot (search results show the agenda is still being updated; no definitive time obtained).

---

⚠️ Risk disclaimer: this recap is an after-the-close information review and observation only, and does not constitute investment advice. Data may differ in timeliness or convention; company disclosures / SEC filings prevail, and this must not be used directly as a basis for trading.