US Market Recap | Monday, 2026-08-24 (ET)

Coverage window: 2026-08-24 09:30–16:00 ET regular session + 16:00–17:10 ET after-hours.

Quote conventions (this report does not mix readings taken at different moments):

• Prior close, open, high, low, close, volume and after-hours price for single stocks/ETFs come from the CNBC quote API, with the close taken at 16:00 ET (some ETFs settle at 16:10/16:15); after-hours readings are taken 16:00–17:10 ET, and every after-hours move is accompanied by its after-hours volume.
• The three major indices, Russell 2000, VIX, Treasury yields, the dollar index, commodity futures and crypto also come from the CNBC quote API, read at 16:15–17:07 ET.
• Gold/silver/oil are continuous futures contracts (@GC.1 / @SI.1 / @CL.1 / @BZ.1) and carry a basis versus spot; do not mix them with spot figures.
• The "vs. pre-market price" column in this report is the primary reconciliation basis: i.e. "today's close ÷ the quote printed in the pre-market list − 1". Reconciling on the day's percentage change alone systematically overstates the hit rate — the pre-market list was written after the gap, so a reader executing it buys at the pre-market price, not at yesterday's close. Today the two bases give conclusions a full order of magnitude apart; see §2.
• ⚠️ Market breadth (NYSE/Nasdaq advancers/decliners) could not be obtained for this report; the item is left blank rather than filled with an estimate. This report instead uses the equal-weight vs. cap-weight spread (RSP − SPY) as a breadth proxy, and labels it explicitly as a proxy.

---

⚠️ Read this first: today's switch was not anything on the pre-market list — it was two same-day memory headlines

The pre-market list framed today as "AAOI's $600 million ATM (8/21 SEC 424B5) + YMTC's STAR Market filing acceptance → AI hardware chain rolls over". The closing data says: the rollover was real, but neither switch named pre-market was today's switch.

What was actually being priced today were two same-day headlines that appear nowhere in the pre-market list — and both are about memory:

#	Headline	Content	Market reaction today
A	Nvidia notified customers: AI servers will be priced 15%+ higher next year, because of memory costs (first reported by Bloomberg 8/22, went broad 8/24)	The increase applies to systems shipping early next year, including Vera Rubin and Grace Blackwell; memory (HBM4+LPDDR5X) goes from 5–10% of rack BOM on Blackwell to 25–30% on Vera Rubin, and from roughly $370,000 per rack to roughly $2 million per rack (+435%)	NVDA −2.91% (close 208.48), breaking the $214 test the pre-market list had set for it; HPQ −3.80% (PC OEM; memory is one of its largest BOM items)
B	Reports say the Trump administration may, after Xi's US visit in September, allow Apple to source memory from CXMT / YMTC	This is a US-China negotiating chip, with no final decision yet; senators from both parties had previously pressed Apple to commit to not sourcing there	SNDK −6.45%, STX −6.51%, MU −5.83%, WDC −5.24% — the four memory names were the deepest-falling group of the day

Table scrolls horizontally

⚠️ The pre-market list named the right companies but the wrong transmission channel and the wrong clock — and today that difference was worth 5 to 6.5 percentage points.

Pre-market attributed the memory downside to "YMTC's STAR Market IPO filing accepted, raising RMB 33 billion", and explicitly wrote "supply transmission takes 1–2 years; no current-period impact". What actually hit the tape today was the same company, YMTC, but through an entirely different leg: not "how much more NAND it will produce in 1–2 years," but "Apple may be able to buy its NAND and DRAM right away." The former is forward supply; the latter is near-term demand reallocation. Pre-market had the right name attached to the wrong clock.

⚠️ And the thing most worth remembering today: two memory headlines pointing in opposite directions landed on the same day, and the market priced only the bearish one.

Headline A (Nvidia raising prices 15%+ because of memory cost) is directionally strongly bullish for memory — it is the buyer admitting out loud that memory is expensive enough to pass through to customers, a first-hand confirmation of memory pricing power. Headline B is bearish. The result: memory stocks fell 5–6.5%. This is not to say the market was wrong (B is genuinely nearer-term), but a reminder: when a supply-side confirmation and a demand-side threat show up on the same chain, the day's price only tells you which one is nearer, not which one is bigger. This report does not turn bullish on memory on that basis, but records it as a verification point for NVDA's 8/26 call (see §6③).

⚠️ One more thing that has to go at the very top: the falsification the pre-market list spent the most space on got its strongest possible confirmation today.

Pre-market used three independent pieces of evidence to argue that "Canadian tariffs are bullish for US steel and aluminum" was wrong (steel, aluminum and copper are explicitly excluded from Section 338; the 232 50% rate has been in force for 14 months; Canadian steel imports into the US are already −32%), and wrote item 6 of §9③ as "avoid any US steel/aluminum position taken on the grounds that 'Canadian tariffs are bullish'".

Today Trump announced on Truth Social: from 2027-01-01, a 50% tariff on Canadian cars, trucks, auto parts "and steel." That is a literal "50% Canadian steel tariff" headline. And the steel stocks' response was: the open was the high of the day, followed by a slide straight into the close.

Ticker	Pre-mkt	Open	High	Close	Today %	Open→Close %	vs. pre-mkt price %
NUE	+4.22%	249.13	257.13	244.64	+0.41%	−1.80%	−3.65%
STLD	+4.33%	235.64	244.25	229.27	+0.26%	−2.70%	−3.90%
CLF	+3.11%	11.92	12.33	11.30	+0.27%	−5.20%	−2.75%
CMC	—	67.41	68.70	64.87	−2.17%	−3.77%	—
RS	—	395.63	402.07	382.49	−1.39%	−3.32%	—

Table scrolls horizontally

A brand-new 50% steel tariff headline bought the whole sector a one-way slide from the open and a close near the day's low. The pre-market conclusion — "no incremental protection can be capitalized" — needs no better verification.

At the same time, pre-market's correction on "who actually gets hit by this tariff" was right again: F −3.33%, STLA −3.51%, HMC −2.34%, GM −1.08%, while AA (Alcoa) −4.74%, the largest decline in the entire metals complex. And the Canada equity ETF EWC fell only 0.38%. A 50% tariff aimed at Canada knocked US automakers down 1–3.5% and knocked Alcoa — which has 36.3% of its smelting capacity in Canada — down 4.7%, yet barely moved the Canadian index: the pain sits with the importer and with cross-border capacity owners, not with the exporting country's index (EWC is weighted toward banks and energy).

Finally, one line pre-market did not write but today requires: NVDA fell for the 7th consecutive session today.

Checking closes day by day (from the 8/13 close of 225.30): 8/14 −0.06%, 8/17 −0.07%, 8/18 −2.34%, 8/19 −0.99%, 8/20 −0.33%, 8/21 −0.98%, 8/24 −2.91%, seven straight down days, −7.47% cumulative. Media call it the longest losing streak since autumn 2022 (CNBC says "since September 2022", Yahoo says "since October 2022"; the two disagree, so this report uses only the independently checkable "seven straight declines / −7.47%"). It got into this shape ahead of its own earnings on Wednesday after the close.

---

0. Today in One Line

① Character: not risk-off, but "a single-point semiconductor collapse + broad gains inside the index." Of the 11 sector ETFs, 8 rose and 3 fell; equal-weight RSP +0.12% beat cap-weight SPY −0.29% by 0.41pp — the index's decline was contributed entirely by the mega-caps, and the median stock was up today. The three that fell were XLK −1.78%, XLE −0.83%, XLI −0.69%; the top three gainers were XLP +1.70%, XLF +1.29%, XLU +1.05%. SMH −2.43%, SOXX −2.67%, while SPY was only −0.29%.

② Today's strongest theme was "the memory cost shock," and the pre-market list did not carry it at all. It hit both ends at once: the buy side (NVDA −2.91% on the price increase, HPQ −3.80% on BOM cost) and the sell side (MU −5.83%, SNDK −6.45%, STX −6.51%, WDC −5.24% on Apple possibly turning to Chinese memory). Pre-market framed this chain as "the 1–2 year forward supply from YMTC's STAR Market IPO" — the clock was off by two years.

③ The reconciliation of the pre-market list produces an extremely split number: 90% on direction, 25% on executability. Of 40 judgeable names, 36 were directionally correct (90.0%); but the executable long side was only 4 "watch closely" names, of which just 1 (IBIT) made money if bought at the pre-market price (25%). Of the 13 short/avoid names, 12 closed lower (92.3%), but executed at the pre-market price only 7 produced a positive return (53.8%) — the declines had already been completed inside the pre-market gap. Details in §2.

④ Today's most expensive lesson is called NSSC. Pre-market §5.5 described it as "the only genuinely earnings-driven big gainer today / the largest liquid advance in the whole field" and rated it "watch closely." It was $45.54 pre-market (+19.56%), opened 46.16, high 51.77, closed 37.59 (−1.31%). Executed at the pre-market price, the close is −17.46%; from the open it is −18.57%. A real, better-than-expected earnings report plus a +19.6% pre-market gap produced the single worst trade of the day.

⑤ Pre-market's correction that "AAOI is not the reason for the whole chain" was the single best thing on this list. Today AAOI −13.77%, while COHR −4.85%, LITE −4.22%, CRDO −3.45%, FN −3.56%. AAOI's decline was 2.8–4x the sector's; the offering is its own affair. And COHR/LITE's −4~5% is clearly smaller than the five "no-news joint selloffs" pre-market listed (7–12%) — by pre-market's own test ①, not even "sector re-rating" holds today; it looks more like a day inside the base rate.

⑥ The Iran sanctions call was right, and right even though the event came in stronger than expected. Bessent today announced "Operation Economic Outcast", including language as strong as "countries that continue dealing with Iran will be removed from the dollar system after a grace period" — aimed at China (which takes roughly 90% of Iran's crude exports). Result: WTI −2.45% to 84.93, Brent −2.60% to 91.94, XLE −0.83%, USO −1.80%, XOP −1.74%, OIH −1.59% — more downside than at the pre-market read. Pre-market's two calls — "no directional bet on an event day" and "the expected version is already priced" — were both right.

⑦ The one-line framing: today was "memory invoice day" — the cost structure of the AI chain was put on the table for the first time in the form of "the buyer proactively raising prices 15%", and the market priced it by selling the buyer and the seller simultaneously; meanwhile money moved into consumer staples, the big banks and utilities, and 8 sectors inside the index were up. Pre-market read the day as "AAOI's offering + optical rolling over": direction right, switches all wrong. And the only genuinely expensive error was not in judgment but in execution — NSSC and PDD, both "watch closely," were both chases at the top of a gap.

---

1. Market Overview

⚠️ The two most important rows in this table are not indices — they are DIA and RSP. 100% of the Dow's gain came from the regular session (gap −0.07%, open→close +0.34%) — that is money actively bought during US hours, not something left over from overnight. RSP (equal-weight) +0.12% versus SPY (cap-weight) −0.29%, a 0.41pp spread. The median stock was up today; the S&P's minus sign was pressed out entirely by the mega-caps (the NVDA/AVGO/AMD/MU group). Reading today as "the market fell" is wrong.

Market breadth: ⚠️ NYSE / Nasdaq advancers/decliners could not be obtained for this report; the item is left blank. Using RSP − SPY = +0.41pp and 8 of the 11 SPDR sectors up, 3 down as proxies, the direction indicates positive breadth. A proxy is not an advance/decline count; readers must not cite it as one.

Sentiment & Rates

⚠️ One methodological confirmation: pre-market specifically wrote that "today's rally in hard assets cannot be attributed to a weaker dollar; that transmission chain does not hold today." The closing data fully supports it: the dollar +0.19%, gold +0.55% and BTC +2.12% all held simultaneously, while the long-end 30Y fell 4.7bp. The attribution should indeed land on long-end real rates, not on FX. This one pre-market got right.

⚠️ But the same section contains a popular attribution that was falsified today, and it deserves to be pulled out separately: "rising rates compress long-duration growth" does not hold today. Today the 10Y fell 4.2bp and TLT +0.62%, while XLK −1.78%, SNOW −3.00%, DDOG −4.18%, ARKK −2.66%. Yields fell, and long-duration growth fell the hardest. What hit tech today was not the discount rate — it was the cost structure of AI hardware.

⚠️ The only macro data point today: 08:30 ET Chicago Fed National Activity Index (July) = −0.08, prior (June) +0.06. That is below the −0.02 prior cited in the pre-market list (which was on a pre-revision basis); it counts as a mild softening but far from a recession threshold (three-month average −0.70). No identifiable impact on today's prices.

---

2. Pre-Market List Reconciliation

This section is organized by "the executability pre-market assigned," not by direction of the move. The reason is simple: a correct "watch only" makes no money, a wrong "watch closely" loses money, and blending them into one hit rate washes out today's real result.

2A. Long "watch closely" — the only genuinely executable group, 4 names

Hit rate for this group: 2/4 (50%) on the day's move; 1/4 (25%) versus the pre-market price.

2B. Short / avoid direction (13 names)

Hit rate for this group: 12/13 (92.3%) on the day's move; 7/13 (53.8%) versus the pre-market price.

2C. "Watch only" Pass group (23 names)

Test used: a Pass is correct if the stock did not rise ≥ +2% on the day (i.e. "not buying" did not cost you anything).

Hit rate for this group: 22/23 (95.7%).

⚠️ But two of the "correct" calls in this group deserve to be pulled out, because the way they were correct lands squarely on a pre-market blind spot: XLU +1.05% and XLP +1.70% were the 3rd and 1st best sectors of the day. Pre-market rated them B−, 29 pts, the lowest on the whole board, with the reasoning "this is a bounce, not a trend; on Friday it was the worst in the field." By the definition of a Pass it was not wrong (the gains sat just under / just over the 2% threshold), but the fact that "the best-performing sector today got this report's lowest score" should not be papered over by a technical Pass convention. Moving with them: DLTR +4.01%, ULTA +3.32%, BBY +1.85%, DG +1.56% — discount retail + consumer staples was a real theme today and is completely absent from the pre-market list; see §3⑤.

2D. Overall hit rate and post-mortem

The post-mortem (three items today, because these are three different kinds of problem)

① The 90% directional hit rate contains almost none of today's money. 12 of the 13 shorts closed lower, but the declines in AAOI (12.4pp), XPEV (4.4pp), SNDK (5.4pp) and MU (3.2pp) were all completed before 09:30; for COHR, LITE, CRWV, NBIS and MRVL the entire decline sat in the gap, and the session actually bought it back. That is not a judgment problem; it is a list-structure problem: a list finished at 08:10 can only identify what the market has already priced. The "vs. pre-market price" column compressed 92.3% down to 53.8% today, and that compression ratio is the number most worth remembering in this report.

② NSSC is a failure that must be reviewed on its own — and the way it failed was written up by the pre-market list itself. Pre-market §5.5's reasoning for NSSC was "the only genuinely earnings-driven big gainer today" and "the largest liquid advance in the whole field" — both statements are facts, and neither is a reason to buy. Its pre-market volume was 56,522 shares, only 1.8x this report's own 20,000-share threshold, and yet it was treated as "liquid." The top of that same report says "⚠️ pre-market volume is cumulative, not a snapshot; a quote on thin pre-market volume is not a fact," and §8 says "the pre-market volume threshold must be applied in both directions" — and then a 56,000-share, +19.6%-gap name got a "watch closely." The rule was written three times and skipped on the very name it most wanted to recommend. This is the same failure mode as "the checklist gets skipped on the branch you most want to recommend," recorded in last week's recap, repeated one day later.

③ The only one that made money (IBIT) is the only name on the whole list whose pre-market volume was genuinely too large to fake (5.036 million shares → 83.85 million by the close). Pre-market volume versus return-at-pre-market-price for today's four "watch closely" names:

Ticker	Pre-mkt volume	vs. pre-mkt price %
IBIT	5,036,000 shares	+0.81%
PDD	920,252 shares	−5.42%
NSSC	56,522 shares	−17.46%
NUE	4,026 shares	−3.65%

Table scrolls horizontally

The two ends are clean: the largest-volume name was the only positive one in the group, and the second-smallest was the worst (by an order of magnitude). The middle two (PDD and NUE) are inverted, so this is not a strictly monotonic relationship and this report does not write it up as a rule. But "pre-market volume < 100,000 shares should not qualify for watch closely" is supported today by NSSC and NUE simultaneously. Tomorrow's list should promote "pre-market volume" from a warning label to an admission threshold (see §6④②).

---

3. Theme Verification

① Optical: pre-market test ① gave a clear answer, and the answer is "does not hold"

The original text of pre-market test ①: support = "COHR/LITE's full-day declines significantly exceed the median decline of their 'no-news days' over the past 8 weeks, and the ranking of declines is still modules > systems/foundry"; falsification = "the declines land inside that group's normal range → this is not an event, it is the base rate."

Both conditions point to falsification:

(a) Not big enough. The five no-news joint selloffs pre-market itself listed over the past two months: 6/23 (COHR −9%, LITE −8%), 7/02 (COHR/LITE −10%), 7/15, 7/28 (COHR −11%, LITE −9%), 8/10 (COHR −12%, LITE −7%). Today COHR −4.85% and LITE −4.22% are smaller than every one of them — smaller by half.

(b) The ranking broke, and broke backwards.

The pre-market ranking logic was "the closer to optical-module pricing itself, the heavier the damage; the closer to foundry/system integration, the lighter", and it stated "if FN clearly outperforms AAOI today, the market is making precise segment distinctions". In fact: CIEN (a systems vendor, ranked 4th pre-market and explicitly described as "should be damaged less") fell 6.02%, underperforming both module makers COHR and LITE; POET (ranked 6th pre-market) fell 6.06%, the 2nd worst in the group. What the market did today was not segment differentiation.

Conclusion: AAOI's $600 million ATM is a clean single-stock event (it fell 2.8–4x the sector), but the theme "AI optical chain re-rating" did not hold today — it did not even reach this group's usual no-news-day volatility. Pre-market ranked it the No. 1 theme in the report and gave it A+; in hindsight that was an overestimate.

② Memory: "NAND purity pricing" was explicitly falsified

The test pre-market set for SNDK: "whether the intraday decline gap between SNDK and MU stays above 2pp (stays = the market is pricing NAND purity; converges = it is just broad semiconductor de-risking)."

Actual: SNDK −6.45% vs MU −5.83%, a gap of 0.62pp → converged, falsified.

And pre-market's line that "the decline is strictly positively correlated with NAND purity" was completely reversed:

STX, with the smallest NAND exposure, fell the most, and SNDK, the purest NAND name, only ranked 2nd. This is not NAND supply pricing; this is the entire memory/storage group being hit indiscriminately by one policy headline — and that headline (Apple possibly sourcing from CXMT/YMTC) covers DRAM and NAND at the same time, so it does not discriminate on purity. Pre-market used a NAND-only headline (the YMTC IPO) to explain a pattern that killed DRAM and NAND alike; naturally the ranking did not line up.

⚠️ There is one more piece of counter-evidence, and pre-market §5.5① had already written the test in advance, so it can be closed today: pre-market said "if the market really were reading 'Samsung's record shareholder return = capex has peaked,' equipment stocks should be the first casualties." Today AMAT −1.65%, KLAC −1.32% and LRCX −1.22% all clearly outperformed SMH's −2.43% (TER −2.88% excepted). Equipment names were not the first casualties; they beat the semiconductor complex. "Capex has peaked" was not priced today.

③ Steel & aluminum: today delivered a confirmation at the level of a counterfactual experiment

See the box at the top. One additional control test set pre-market:

Overall: pre-market falsifying the "tariff" reason = correct; identifying AA as the damaged party = correct; judging NUE/STLD's pre-market gains as "not a fact" = correct. But the substitute reason it put in place, "the HRC spread," did not buy any upside either — the whole group topped at the open. The reason was swapped correctly; the direction was not.

④ Crypto: the three legs went three different ways today

The new item: crypto brokers decoupled from BTC today too. COIN and HOOD were the two biggest winners of last Friday's Clarity Act move (+8.20% / +13.70%); today they gave back 3.76% and 4.17% respectively. What they track is not the coin price — it is the decay curve of last Friday's headline.

⚠️ Something that has to be owned up to: pre-market deleted a correct conclusion in §9③④. The pre-market draft had written "the momentum decay in COIN/MSTR is clearly visible," then deleted that conclusion because "the +0.24%/+0.81% at 08:08 became +0.93%/+1.48% by 08:26," rewriting it as "today is a do-not-chase, not an already-weakening." Today COIN −3.76% is weakening; MSTR +2.83% is not. In other words: the deletion itself was methodologically right (that evidence really was a pre-market snapshot that should not have been used), but the deleted conclusion was half right and the surviving one was wrong on the other half. The right move was not to keep the draft conclusion, but to point out at the time that COIN and MSTR are two different assets — one is volume beta (tracking the news flow), the other is coin-holding leverage (tracking the coin price). Putting them side by side in the same "already-obvious hard-asset high-beta end" bucket is the real error, not the deletion.

⑤ Two themes entirely absent from the pre-market list

(a) Consumer staples / discount retail — the best-performing sector today, and pre-market gave it the lowest score on the board.

XLP +1.70%, the best of the 11 SPDR sectors. At the single-stock level: DLTR +4.01%, ULTA +3.32%, BBY +1.85%, DG +1.56%. Among them DG, DLTR, BBY and ULTA all report this Thursday (8/27), pre-market or after the close — this is a "front-running into earnings + defensive rotation" overlay, and the pre-market list's entire description of XLP was "a bounce after Friday's XLU −2.28% / this is a bounce, not a trend / 29 pts, lowest on the board."

(b) Large banks — the 2nd best sector today, and pre-market said not a word.

XLF +1.29%, with JPM +1.37%, SCHW +1.20%, WFC +1.05%, BAC +1.04%, while GS −0.29%, MS −0.06%, C +0.05%, KRE −0.13%. It is "large commercial banks" rising, not "financials" rising — the investment banks and the regionals did not follow.

⚠️ This report was unable to trace XLF's +1.29% to a first-hand same-day catalyst. The curve bull-flattened today (2s10s −4.4bp), which from a net-interest-margin standpoint is unhelpful for banks, so this is not a rate explanation. Absent confirmation, this report records it only as "rotation absorbing the outflow from semiconductors" and assigns no causation. This is an unfinished question mark, not a conclusion.

---

4. After-Hours Earnings Moves

⚠️ There are no heavyweight after-hours earnings tonight. The 8/24 after-hours slate consists only of PICS and TUYA, two micro-caps (PICS −1.93% after hours, TUYA +3.41% after hours, on 91,000 / 90,000 shares respectively), which do not constitute a next-day catalyst.

Today's earnings events happened pre-market, not after the close:

⚠️ On the NSSC primary source: this report was unable to retrieve the text of the company's 8-K EX-99.1; the figures above come from secondary accounts cross-checked against two sources. The revenue/EPS figures are usable, but do not quote them verbatim.

The real after-hours catalysts are in the next two days, not tonight — full calendar in §6②.

---

5. Flows & Sentiment

Sector rotation: 8 up, 3 down; money moved out of semis into defensives and the big banks

One structure that must be unpacked: "tech" split internally today, and the way it split refutes the rate attribution

Original text of pre-market test ⑤: support = "the QQQ-minus-SPY spread stays around −0.4pp and software (CRM/NOW/SNOW/DDOG) does not follow lower"; falsification = "XLK weakens across the board and SPY's decline catches up to QQQ's → this is not rotation, it is risk-off."

Verdict: neither branch fully hit; record this as "the test itself lacks resolving power."

• QQQ − SPY = −0.71pp, nearly twice as wide as the −0.4pp in the support branch;
• half of software followed lower (SNOW −3.00%, DDOG −4.18%) and half did not (CRM −0.05%, NOW −0.33%);
• but SPY's decline did not catch up to QQQ's at all (−0.29% vs −1.00%), and XLK was not "weak across the board" — the mega-cap platforms were all up;
• and with 8 sectors up + RSP beating SPY by 0.41pp, risk-off is clearly ruled out.

Conclusion: today was rotation, but not the kind pre-market envisioned ("de-risking inside the AI chain, everything else unchanged"); it was cross-sector rotation from "AI hardware → defensives + big banks + platforms." The test pinned its observation point on whether software followed lower, and today software split in half internally, so the test cannot give an answer. Tomorrow should switch to indicators that do not depend on a single small sample, such as "equal-weight vs. cap-weight" and "sector advancers/decliners."

⚠️ And the most valuable use of this split is to falsify a popular attribution: today the 10Y fell 4.2bp, the 30Y fell 4.7bp and TLT +0.62%, while long-duration growth (SNOW/DDOG/ARKK) fell harder than the market. The "rates up → duration de-levering" chain does not even hold directionally today. What hit tech today was the cost structure of AI hardware (memory), not the discount rate.

risk-on / risk-off characterization

Characterization: mild risk-on, but concentrated in defensives and value; semiconductors are the only risk-off pocket.

Evidence supporting this characterization:

• 8 of 11 sectors up, RSP +0.12% > SPY −0.29%, and the Dow's +0.34% was actively bought intraday;
• VIX 15.85 — up 4.76%, but still below 16 in absolute terms, and the close is below the pre-market read (15.92);
• No stress in credit or rates: 10Y/30Y lower, TLT +0.62%;
• BTC +2.12%, IBIT +2.20% at recent highs of strength, which is not risk-averse behavior.

Counter-evidence (which must be listed alongside):

• Russell 2000 −0.76%, small caps clearly trailing the Dow by 1.02pp, and open→close −0.53%, i.e. sold during the session;
• XLE −0.83%, XLI −0.69% — the two cyclical sectors were down;
• ARKK −2.66%, high-beta growth is being trimmed.

Overall: this was a day of "money moving around inside the index, total exposure not reduced, but risk appetite dialed down a notch."

---

6. Next-Day Outlook (Tuesday, 2026-08-25)

① Theme continuation

② Tomorrow's earnings and macro calendar (all times ET)

Macro:

Earnings:

Later this week (the two things suppressing every position today have not changed):

③ Key names to watch (ticker + verification point)

④ What to avoid

1. Anything whose move was already completed inside the pre-market gap — this is today's most expensive lesson, not a theme. The direction on AAOI, SNDK, MU and XPEV was right, but 12.4pp / 5.4pp / 3.2pp / 4.4pp of the decline happened before 09:30, and the session added only 1.35% / 1.13% / 2.76% / 4.29%. Tomorrow's list must not give "watch closely" to any name with a pre-market gap >4%.
2. No name with pre-market volume < 100,000 shares gets into "watch closely." Today NSSC (57,000 shares) and NUE (4,000 shares) were the two worst trades on the entire list, and both were on this report's own thin-volume warning list. Raise the threshold from 20,000 shares to 100,000 shares, and apply it equally to names that are up and names that are down.
3. Directional trades in the optical chain. Today it did not even reach its own base rate, which says this group's volatility now comes mainly from liquidity rather than information. Until the actual pace of AAOI's ATM selling is visible, neither side has an edge.
4. Any directional position in NVDA into earnings (Wednesday after the close) + MRVL into earnings (Thursday after the close). Buying after seven straight declines is catching a knife; shorting into earnings means wearing the gap. Both should be treated as untradeable until Wednesday/Thursday.
5. The crypto "high-beta end" (COIN / HOOD / miners). BTC +2.12% today and all three groups fell; their correlation with the coin price has failed 3 days running. If you want crypto direction, trade IBIT or MSTR, not the proxies.
6. Energy direction (XLE / USO / XOP / OIH). The sanction details have landed, the intensity exceeded expectations, and oil accelerated lower — this is an event that has already cleared; going long has no catalyst, and going short means carrying Hormuz tail risk.
7. ⚠️ New: any equipment-stock short taken on the grounds that "Samsung's return program = semiconductor capex has peaked." Today AMAT / KLAC / LRCX all outperformed SMH; the market did not accept that narrative today.

⑤ Input notes for tomorrow's pre-market list

1. Promote "pre-market volume" from a warning label to a ranking variable. Today's four "watch closely" names ranked by pre-market volume (IBIT 5.04 million > PDD 920,000 > NSSC 57,000 > NUE 4,000) is almost exactly the ranking of today's returns. The §3 master board of tomorrow's list should be stratified by pre-market volume first, then ranked by event strength within each stratum.
2. Every theme must carry a "pricing progress" dimension. 5 of today's 8 themes were on "day 2–4," and the one that actually delivered magnitude was on day 1. The pre-market list currently labels only "persistence," not "how many days it has already been priced," which lets a fading old theme and a brand-new one receive the same score.
3. The scan window must cover English financial media published over the weekend and priced on Monday. Today's top driver (Bloomberg's 8/22 report on Nvidia's price increase) was published on Saturday, and the pre-market list's stated scan range was "8/21 16:00 close → 8/24 08:10" — the range nominally covered it but in practice did not catch it. Tomorrow, "first-hand media reports published over the weekend" should be a separate search item.
4. Tomorrow's list needs to actively answer two questions today could not: (a) what was the same-day catalyst for XLF's +1.29% (not obtained here; flagged as unfinished in §3⑤(b)); (b) is the rally in discount retail defensive rotation or front-running into earnings — DG/DLTR/BBY/ULTA all report Thursday, so this question will have an answer within four days.
5. Keep this week's holding period at ≤2 days. Neither suppressant — Wednesday's core PCE + NVDA earnings, and Friday's Warsh debut — has been lifted today.

⚠️ Failures and implementation details from data collection and verification for this report, for internal reliability assessment:

Data path

• Local yfinance was not used this time; single stocks/ETFs/indices/macro all went through the CNBC quote API (quote.cnbc.com/quote-html-webservice/restQuote, with exthrs=1), pulling OHLC + volume + after-hours price + after-hours volume in one shot. This is the established path, not a fallback.
• Symbol freshness was checked one by one: last_time for .SPX / .DJI / .IXIC / .RUT / .VIX / US10Y / US2Y / US30Y / .DXY / @GC.1 / @SI.1 / @CL.1 / @BZ.1 / BTC.CM= was 2026-08-24T16:15–17:07-0400 for every one, with none hitting the known _DJI/_COMP stale-data trap.
• CNBC futures symbols worked today (unlike the 8/24 pre-market run when @ES.1/@NQ.1/@YM.1/@RTY.1 were all empty), but this is a closing recap and index futures are not needed.

Channels that actually failed

• WSJ market diary (wsj.com/market-data/stocks/marketsdiary) could not be retrieved — that is the primary source for advancers/decliners, so §1's market breadth is left blank, replaced by the RSP−SPY proxy with a note in the body. Repeated searches also failed to turn up the specific NYSE/Nasdaq A/D numbers for 8/24.
• SEC EDGAR archive pages return 403: NSSC's 8-K EX-99.1 (sec.gov/Archives/edgar/data/0000069633/...) could not be retrieved, so §4's NSSC figures rest only on secondary cross-checking, flagged in the body as "do not quote verbatim."
• CNBC article pages return 403 (cnbc.com/2026/08/24/trump-canada-auto-tariffs-trade-war.html), so the original text of the Trump tariff item could not be retrieved; Fox Business's account plus multi-source cross-checking was used instead. This report therefore makes no claim about the exact publication time of that Truth Social post, and says only "intraday."
• stockanalysis's /api/charts/s/<t>/1M now 404s; NVDA's seven-day streak was instead obtained from the SvelteKit flat data at stocks/<t>/history/__data.json for daily OHLC, so the seven straight declines and −7.47% were computed here from closing prices, not quoted from the media.

Two errors that nearly made it in

• Yahoo's "Stock Market News for Aug 24, 2026" gives 8/21's closing data (Dow 53,277.01 = today's prior close; "advancers outnumbered decliners 1.68-to-1"). The old trap of a recurring column headline hiding the publication date — spotted and not used. Those two advance/decline ratios belong to last Friday, and copying them would have put an outright error into §1.
• OLLI's after-hours +5.19% ($80.32,26,531 shares) nearly got written up as "tonight's after-hours earnings move." Verification: Ollie's FQ2 report is scheduled for 9/2 pre-market, and there was no company news today; that after-hours print is a thin-volume mark. Removed entirely; §4 does not mention it. This is the same class of failure as "a search summary conjuring an earnings report out of nothing."

Open items

• The same-day catalyst for XLF +1.29% was not identified. Every bank-related article found had an unconfirmable publication date (several were from earlier in August or are cross-month recurring columns), so the body explicitly says "no causation obtained; recorded only as rotation absorption", and it is listed as a to-do in §6⑤. Better to leave it blank.
• On the Apple / CXMT / YMTC item, only secondary accounts and reposts were obtained; the full text of the original report was not. The body is written as "reports say" and "no final decision yet," and it is not treated as established fact.

---

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