US Market Recap | 2026-08-12 (ET) Wednesday

Coverage window: 2026-08-12 09:30 ET open → 16:00 ET regular-session close; Section 4's after-hours earnings moves are updated through 18:08 ET. Quote conventions:

• Closing prices, open/high/low, volume and after-hours prices for single stocks and ETFs come from the stockanalysis.com quote API, regular session on a 16:00 ET close basis, after-hours price snapshots at 17:04–18:08 ET (tagged per ticker in Section 4).
• S&P 500, Nasdaq 100, Russell 2000, VIX / VIX9D / VIX3M come from the CBOE delayed-quote API, close basis (last trade time 16:09–16:15 ET).
• The Nasdaq Composite comes from the official Nasdaq quote API (26,588.49, +0.54%).
• Treasury yields come from the official US Treasury daily yield curve (treasury.gov), on a 15:30 ET basis, two decimal places.
• The Dow uses the media / index-vendor convention of 53,770.27 (−21.58 points, −0.04%), cross-checked against the DIA ETF (−0.02%) for directional consistency.
• Earnings figures are taken from the company press release wherever possible (flagged in the tables); where the original was unavailable, authoritative media reprints are used and noted. Data not obtained for this issue: ① NYSE / Nasdaq advance-decline counts were not obtained for this issue — one breadth dataset that surfaced claimed "US stocks closed lower on 8/12, tech led the decline", which directly contradicts the close direction confirmed by three independent sources used here, so it was discarded in full; this issue instead uses the spread between the equal-weight ETF (RSP) and the cap-weighted one (SPY) as a breadth proxy, explicitly flagged as a proxy rather than raw breadth data. ② The Dow's intraday high/low was not obtained for this issue, so the intraday-shape analysis in Section 1 covers only the S&P 500, Nasdaq 100 and Russell 2000. ③ The convention gap between Cerebras's "core revenue" and GAAP total revenue could not be explained in this issue (see Section 4.3); that item is flagged as pending verification. ④ Crude oil, gold, the dollar index and bitcoin are represented in this issue by the closing changes of ETF proxies (USO/GLD/UUP/IBIT), not futures or spot raw quotes; the readings are directional reference only.

---

0. One-Sentence Recap

Today was a "very strong single stocks, very weak indices" day, and that gap is itself the conclusion. CPI landed in line, event risk was cleared — VIX closed at 14.55 (−4.78%), while the 9-day VIX9D outright collapsed 11.42% to 11.09, draining front-end fear completely; AI infrastructure earnings all delivered too, with Nebius (NBIS) closing +34.14%, CoreWeave +19.28%, Super Micro +19.02%. Yet the S&P 500 closed only +0.26%, the Dow was actually down 0.04%, and the S&P's open was the high of the entire day (7,765.46), with the close landing at the 38% percentile of the intraday range; the Nasdaq 100 closed at just the 16% percentile. In other words: money did not leave, but money is only willing to sit in a handful of names, and index-level buying had been fading from the very first minute of the session.

The pre-market list's directional hit rate was 18/19 (94.7%), but that number is deceptive — executed at the pre-market prices, only 12/18 (66.7%) made money (reconciliation conventions in Section 2). The strongest theme was not "AI cloud compute", which we ranked #1, but "AI server hardware", which we ranked #4 and tagged "highly contested" — beyond SMCI +19.02%, DELL (+9.87%), HPE (+8.11%) and ANET (+6.39%), none of which appeared anywhere in the pre-market list, were the cleanest chain of the day.

Next-day tone: leaning cautious. The four earnings reports after the close delivered a highly consistent signal — Cisco beat and guided FY27 above the market, and fell 4.37% after hours; Coherent's non-GAAP EPS of $1.74 punched straight through the top of its own guidance range ($1.52–1.72), and it fell 3.91% after hours; Cerebras raised full-year core revenue guidance, and fell 15.66% after hours. All three did the homework, all three were sold. "Buy the expectation, sell the fact" has already begun at the edge of the AI chain, and that is the first question tomorrow's open has to face. The 10Y Treasury at 4.68% (−2bp) and VIX at 14.55 mean neither the bond market nor volatility is sounding an alarm; tomorrow's real variable is the 08:30 ET PPI (prior −0.3%, consensus +0.2%), plus Applied Materials after the close.

---

1. Market Overview

1.1 Indices and Intraday Shape

The most important line in this table is not the percentage change, it is the last column.

• The S&P 500's opening price of 7,765.46 was already in its high zone for the whole day (high 7,766.01, only 0.55 points above). In other words, all the buying brought by the CPI print and the AI earnings was used up in the first minute after the open, and the remaining six and a half hours were net selling.
• The Nasdaq 100 closed at the 16% percentile of its intraday range — it pushed as high as 29,881.50 and closed at 29,742.60, giving back 64% of the day's gain. Tech rose the most today, and was also sold the hardest into the close.
• The one exception is the Russell 2000, closing at the 74% percentile. Small caps traded steadily all day with none of the large-cap pattern of spiking then fading. This is one of the few healthy signals today, consistent with rate expectations easing after CPI landed.
• The Dow closed down 0.04% — on a day when the S&P rose 0.26% and the Nasdaq rose 0.54%, the 30 blue-chip components were net lower.

1.2 Market Breadth (Proxy Indicator)

⚠️ Raw NYSE / Nasdaq advance-decline counts were not obtained for this issue; the following is a proxy calculation:

How to read it: today's "narrowness" is not extreme narrowness. Equal-weight lagged cap-weighted by only 0.07 percentage points, while midcaps and small caps actually outperformed large caps — this is not the classic narrow advance where "the Magnificent Seven drag the index up and everything else falls". The real divergence is not along the market-cap dimension but the sector dimension (see 1.3 and Section 5): tech and semis rose alone, while consumer discretionary, materials and communication services fell.

1.3 Cross-Asset and Sentiment

The volatility term structure is the single chart most worth reading today: VIX9D 11.09 < VIX 14.55 < VIX3M 18.53, a steep upward-sloping curve. The front end (9-day, 1-day) was crushed to extreme lows once CPI landed — VIX1D touched 7.60 intraday, which is pricing close to "nothing happens". But the 3-month at 18.53 fell only 2.01%, essentially unchanged.

What this combination means: the market thinks the next two weeks will be calm, but has not loosened its pricing of the autumn (September FOMC) one bit. Yesterday's pre-market report flagged something counterintuitive — current market pricing for September is roughly half-and-half between a hike and a hold, not a cut. After CPI landed, that balance shifted only slightly: hike odds fell from roughly half to about 40%, hold about 55%–60%. In other words, today's CPI removed a short-term risk but did not change the medium-term question. That is exactly why the front end dared to fall to 11 and the back end did not.

Gold +0.99% was the strongest cross-asset of the day, which on a day of "rising risk appetite and falling VIX" is a somewhat discordant signal and worth noting.

Sentiment call: mild risk-on, but the quality is deteriorating. The indices rose while the broad-index heavyweights fell; buying was concentrated early and selling into the close; front-end volatility collapsed while the back end did not move. This is the classic shape of "the last wave of trend-following buying", not the shape of "a new leg starting".

---

2. Pre-Market List Reconciliation

Reconciliation uses two conventions, and both must be looked at:

• "Change% today" = close ÷ 8/11 close − 1 — measures whether the directional call was right.
• "vs pre-market price%" = close ÷ the pre-market price quoted in the pre-market report − 1 — measures whether someone who read that report and executed at the pre-market price made or lost money today.

The second column is the real report card. Names on the pre-market list had generally already gapped 8%–26%; looking only at the first column systematically overstates the hit rate.

2.1 Long Side (19 names)

2.2 Short / Avoid Side (7 names)

2.3 Hit Rate and Post-Mortem

Those 28 percentage points between 94.7% and 66.7% are the price of one word: "gap." SMH rose 2.08% today and the entire AI chain rallied broadly, so getting the direction right was nearly free; the real exam was whether these names were still worth buying after already gapping 8%–26% pre-market. By that standard, one-third of the list lost money.

Three errors that must be acknowledged:

① SMCI: the heaviest misjudgement in the whole issue. The pre-market report used six reasons to push it down to 56 points and a "watch only" call; it closed +19.02% on 165 million shares, and was still +9.65% versus the pre-market price. Where the error lies — not in the facts, but in treating the facts as a same-day ranking device. The one-off gross margin item, the working-capital bleed, the guidance delivery record, the unaudited status, the regulatory investigation listed pre-market — not one of them was falsified today, and no new information today overturned any of them. The real failure was an assumption: pre-market we believed the 7/21 pre-disclosure had already reset market expectations, so "the expectation gap has been consumed". In fact sell-side consensus never marked down with it — EPS consensus was still sitting at $0.94 against an actual $1.70; FY27 revenue guidance of $65–72 billion is roughly 30% above the $52.5 billion consensus. Far from being consumed, the expectation gap was one of the largest of the day. Lesson: "the news was pre-disclosed" does not equal "expectations have been reset". To judge whether an expectation gap has been consumed you must look at whether the sell-side consensus numbers actually moved, not at whether the news has been made public before. This is the same class of error as "financial quality is not a short-term ranker" in two different forms — the former infers expectations from old news, the latter infers same-day moves from statement quality, and both skip the step of "how money is actually pricing it".

② KTB: the avoid rationale was too thin. The basis was only "Q2 revenue $584 million vs $587 million expected", a tiny 0.5% miss, plus "two consecutive down days". It closed +8.86% today, and this after being down 0.96% pre-market before opening straight up at $78.65. A 0.5% revenue miss should never have been treated as bearish evidence in the first place — the pre-market report itself graded its impact level as C. If it is C-grade evidence, it should not have shown up inside a binary conclusion like "avoid". The right handling was to leave it off the list, not to include it and attach a directional call.

③ Theme ranking dislocation (the item with the widest impact). See Section 3.

What was also done right today, for the record: ranking NBIS #1 (+34.14%), EAT's "not pulled forward by a gap" (+7.85%), CAVA's "unattractive after the gap" (−3.20%), HYLN's "watch only" (−21.10%), CRWV's "position too high" (+0.66%), COHR's and CSCO's "don't buy ahead of the event" (both fell after hours), and 5/5 on the software group. What these calls have in common is that none of them were judging direction — they were judging "how much room is left at this price". Today proved that dimension is worth more than the direction dimension.

2.4 MRVL's Volume Anomaly: the Answer Is Now Clear

The pre-market report specifically flagged one item: MRVL printed 4.15 million pre-market shares (the largest in the sample) with no company-specific news at all, and gave two asymmetric explanations — the base case was "money is using it as the liquidity vehicle for the AI semiconductor theme", the alternative was "there is undisclosed news", with the latter tagged as having zero positive evidence.

Today's result: MRVL closed +2.25%, essentially no excess over SMH (+2.08%), the open at $225.90 was in the high zone of the whole day, open→close was −3.90%, and no news landed all day.

The base case holds, the alternative is falsified. Pre-market we explicitly marked the evidence weight of the two explanations as unequal and gave only "watch closely" rather than "priority deep-dive"; that handling was right. For a stock with no catalyst of its own, however large the pre-market volume, it is only the shadow of the theme — it will rise along, but it will be the first to be sold.

2.5 HYLN: the Most Complete "Gap Trap" of the Day

From the intraday high to the close, a drawdown of −28.6%; from the pre-market price to the close, −21.10%. A genuine piece of good news — "full-year revenue guidance raised + a Navy contract" — on a small-cap name with a tiny revenue base was fully priced within 20 minutes of the open, then fell in a straight line all day, and finally broke below the price it traded at before the news came out.

The pre-market rationale was "small cap, high volatility, already gapped 26%", with a "watch only" call. The value of that call today was a hard 21 percentage points. It also explains why "extremely direct catalyst" (the company's own guidance) cannot on its own be a reason to buy — when a piece of good news can be fully priced within 20 minutes of the open, the directness of the catalyst is precisely what accelerates its exhaustion.

---

3. Theme Validation

3.1 The SMH / IGV Divergence: the Test Set Pre-Market Passed

In the "opening validation signals" section, the pre-market report wrote an explicit, falsifiable test:

The test of whether the theme is real is whether the SMH / IGV divergence persists. If SMH rises and IGV turns higher too, it is broad beta, not a theme; only if the divergence persists does the "sell the shovels vs use the shovels" call hold.

Result: SMH +2.08%, IGV −0.81%; the divergence not only persisted but widened further from pre-market (+1.69% / −0.99%) to 2.89 percentage points.

The test passed, and the "sell the shovels vs use the shovels" framework held today. This is the most valuable framework-level call in the whole issue — because it was written pre-market in falsifiable form, not induced after the fact.

Another test also passed: the pre-market report required that "the optical chain needs to see at least three names rising on volume before it can be confirmed as a theme rather than a single-stock event around LITE" — in fact five (LITE, CIEN, FN, COHR, GLW) rose more than 5%, so the theme holds.

3.2 The Biggest Theme Misjudgement: AI Server Hardware Was Ranked #4

Pre-market put "AI server hardware" at #4 with strength A, direction written as "bullish but highly contested", and logic hardness written as "soft". It turned out to be the strongest chain of the day:

The problem is not that SMCI was called wrong on one stock, it is that only one stock in the whole chain was written up. Pre-market did write "SMCI, DELL, MU" in the "representative names" column of the theme table, but DELL never appeared once in the single-stock master list in Section 3, and HPE and ANET were not mentioned anywhere in the whole issue.

The cause of this hole is worth recording: all of the pre-market analytical effort on this chain went into arguing that "the good news on SMCI the stock is fake" — five facts that had to be spelled out, six risks, the longest section in the whole issue. But the fact that "SMCI's margins will be halved" is precisely what does not affect DELL and HPE. SMCI's FY27 revenue guidance of $65–72 billion reads to fellow box builders as "end demand for AI servers is far larger than people thought", and DELL and HPE do not have SMCI's gross margin problem, working capital problem or regulatory problem.

In other words: we did the subtraction on the leader correctly, and forgot to do the addition on its peers. An analysis that is bearish on one stock cannot automatically be equated with being bearish on the chain it sits in — those are two things that must be argued separately, and today only the first was argued.

3.3 AI Cloud Compute: the Ranking Was Right, but the "Leaders Are Equivalent" Assumption Was Wrong

Pre-market put NBIS ahead of CRWV, and that ranking was right — thoroughly right: the gap was not 8 points, it was 15.5 percentage points (on the vs-pre-market-price basis).

The pre-market rationale was that NBIS satisfied all three of "profit inflection + extremely low net debt + not yet at the 52-week high", while CRWV carries $51.6 billion of lease obligations, an equity buffer of only 6.5%, and a marginal funding spread that widened 100bp over three months. Money confirmed that ranking today in the most direct way possible: with equally strong beats, the one with the clean balance sheet gained 15 points more.

But one assumption was wrong: pre-market also listed CRWV as "priority deep-dive" (merely annotated "position too high"). The actual result is that CRWV's net contribution after the open was −0.89%; buying it meant paying the entire 18.49% gap premium to last night's sellers. Those four words "position too high" should have been elevated into the conclusion rather than left as an annotation — within the same theme, when the #1 and #3 names have balance sheets that differ by a factor of five, they should not share the same conclusion label.

One more item on NBIS that must be recorded (data explicitly left blank pre-market, now obtained):

Pre-market wrote: "NBIS's end-of-Q2 ARR was not obtained for this issue (end-Q1 was $1.92 billion), and ARR is the core pivot of the company's full-year guidance." The actual figure: annualized run-rate revenue (ARR) at end-June was about $3 billion, +598% YoY and +56% QoQ from end-Q1. The full-year target is year-end ARR of $7–9 billion — meaning another 133%–200% of growth is required in the second half just to reach the low end of guidance.

At the same time, the most critical attribution call made pre-market held up. Pre-market insisted that this press release contained no forward guidance at all, and that the six pieces of guidance circulating in the market (year-end ARR $7–9 billion, 2026 capex $20–25 billion, full-year revenue $3.0–3.4 billion, roughly 40% EBITDA margin, contracted power >4GW) were all on the old May basis, and therefore "the rally should be attributed to the profit inflection being proven, not to a guidance raise". Re-check result: the company today did indeed merely reiterate full-year revenue of $3.0–3.4 billion, year-end ARR of $7–9 billion and roughly 40% EBITDA margin, and stated explicitly that formal 2027 guidance will be given later this year. Refusing to treat three-month-old guidance as a fresh catalyst today was the right restraint. ⚠️ One further piece of information this issue actively excluded: during searching, a statement appeared claiming "NBIS gave revenue guidance of $14.5–15.5 billion, far above the $11.68 billion expected". That figure is nearly five times the full-year $3.0–3.4 billion the company actually reiterated, and no primary source supports it; this issue judges it a garbled second-hand summary and does not adopt it.

The complete attribution of today's +34% should be three items, not one: ① the profit inflection in the actual Q2 numbers (adjusted EBITDA $236.2M, margin about 41%, versus a market that still expected a net loss); ② the newly disclosed operating figure of ARR +56% QoQ to $3 billion; ③ about 24% of the free float held short, with the beat triggering a short squeeze. The third is a dimension pre-market did not consider at all, and it explains why the move reached the 34% magnitude rather than 15%.

3.4 AI Power: A+ Was Too High

Pre-market gave this derivative theme A+ strength and the #3 rank, on the grounds that "CRWV active power +500MW to 1.5GW; VRT/TLN/ETN/CEG already led yesterday".

Actual result: ETN +0.15%, CEG +0.11%, VST +1.21%, TLN +2.01%, GEV +2.77%, VRT +2.32% (−1.82% vs pre-market price). Apart from BE +12.29%, the whole chain basically did not move.

Pre-market had in fact characterized it correctly — "derivative theme, strongly follower-like, weak independent catalyst", "ETN / CEG / TLN watch only (already delivered yesterday)" — yet still assigned A+ strength and the #3 rank, so the strength label and the risk description contradicted each other. A theme that is "weak on independent catalysts and already delivered yesterday" should not be placed just behind the two S-grade themes. The characterization was right, the strength assignment did not follow.

3.5 Unexpected Themes Missed Pre-Market

---

4. After-Hours Earnings Moves (Next-Day Catalysts)

What tonight's four earnings reports have in common matters more than the details of any one of them: three beat, three were sold.

4.1 CSCO — Beat + Strong Guidance, −4.37% After Hours

FY27 guidance (all above prior market expectations): revenue $72.2–73.4 billion (FY26 was $63.3 billion, +14%+16%), non-GAAP EPS $5.05–5.11 (FY26 was $4.33, +17%+18%). FQ1'27 revenue $18.0–18.2 billion, non-GAAP EPS $1.32–1.34.

This is a report you cannot find fault with, and the stock fell 4.4%. The pre-market call on CSCO was "watch only (pre-event) — information is incomplete ahead of the event, it cannot be priced until FY27 guidance lands". The conclusion was right but the reasoning was only half right: what today proved is not "it cannot be priced before guidance lands", but "it may not be priceable after guidance lands either" — because what gets priced was never the guidance itself, it is the difference between the guidance and the portion of expectations already bought into the price. CSCO was +2.86% in today's regular session with a substantial one-year gain, and $9.3 billion of AI orders is only "meeting" the target, not beating it.

One convention note: on Q4 AI orders, the primary press release says $4 billion, while some media reported $3.7 billion. This issue uses the press release's $4 billion.

4.2 COHR — Punched Through the Top of Its Own Guidance, Still −3.91% After Hours (Full Validation of the Pre-Market Call)

FQ1'27 guidance: revenue $2.2–2.4 billion, non-GAAP EPS $1.85–2.05, non-GAAP gross margin 39.5%–41.5%.

Pre-market wrote two things about COHR, and both hold today:

① On the dispute over the "pass mark" — we were right. Pre-market explicitly refused to use the widely quoted "consensus $1.43", on the grounds that it was below the company's own guidance floor of $1.52, and wrote that "the more reasonable bar is above the guidance midpoint of $1.62". The actual $1.74 not only cleared $1.62 but also cleared the top of the guidance range at $1.72. Had we used the $1.43 figure at the time, we would have concluded "a huge 21% beat" today and been completely unable to explain the after-hours decline. Choosing the company's own guidance as the benchmark rather than a consensus number of unknown provenance preserved the internal consistency of the whole analysis today.

② On the three-layer argument that "the spillover is actually the risk" — all of it played out. The pre-market wording was: "the day before the event, already up 5.6% pre-market, expectations raised by peers, implied move ±10–12%, and even the 'pass mark' number itself is in doubt — five conditions stacked together make this a textbook asymmetric downside structure". Result: a report that punched through the top of its own guidance earned −3.91% after hours.

There is an inference here that can be used directly tomorrow: the actual after-hours move of −3.91% was far smaller than the ±10.5%–12.4% the options market priced beforehand. This is a standard IV crush — the report itself did not supply enough new information to match its implied volatility. Combined with LITE's +8.3% yesterday (inside its implied ±12.2%), the optical chain has now produced two consecutive "beat but did not beat the pricing" outcomes. This main line's information edge is running out, and the next driver can only come from policy (FCC), no longer from earnings.

4.3 CBRS — −15.66% After Hours, but One Convention Question Remains Unresolved

Guidance: Q3 core revenue $214–216 million; full-year core revenue raised from $855–865 million to $880–890 million.

⚠️ One convention question this issue could not resolve, and therefore draws no conclusion on: the company-defined "core revenue" is $210 million while GAAP total revenue is only $180.1 million — core is $30 million higher than total. Normally "core revenue" is a subset of total revenue after excluding items, and should be smaller than total revenue. This issue did not obtain the company's original definition of that measure and cannot judge where the $30 million difference comes from (it might be some contra-revenue accounting treatment, but that is only speculation, without evidence). Until that difference is explained, any conclusion about whether "Cerebras beat or missed on revenue" is unreliable — because the answer depends entirely on which number you use. This item is flagged as pending verification.

The pre-market call on CBRS was "watch only (pre-event)", on the grounds that "this is only the second public earnings report, insufficient volatility history for reference". That call was worth 15.66 percentage points today — and its correctness came precisely from "admitting we did not know", not from predicting the direction right.

4.4 The Common Signal in the After-Hours Moves

Three of the four reports "did the homework and got sold anyway", and the size of the selling bore no relation to the size of the beat:

These three names' regular-session performances were +8.24%, +2.86% and +11.63% — every one of them went into earnings already up. Compare with the daytime session today: CRWV opened with an +18.49% gap and net-contributed 0.66% for the day; CAVA opened +18.02% and closed 3.2% below the pre-market price; HYLN opened +25.77% and closed −21.10%.

Daytime and evening were the same phenomenon staged twice: at this level, the marginal pricing power of good news is decaying fast. That is the single judgement most worth carrying into tomorrow.

---

5. Flows and Sentiment

5.1 Sector Rotation (SPDR Sector ETFs, 16:00 ET Close)

Today's rotation is not "growth vs value", it is "AI hardware vs everything else".

• The leading end is extremely concentrated: SMH +2.08%, XLK +1.49% — beyond those, the third-best gainer XLRE managed only +0.93%, and it rose because rates fell, not because of the theme.
• The lagging end is three sectors with nothing to do with AI: materials −1.24%, consumer discretionary −1.13%, communication services −0.90%.
• XLK +1.49% versus IGV −0.81%, two ETFs both under the "technology" banner, 2.3 percentage points apart — this is the most straightforward picture of where money went today: money is flowing from "software companies that use AI" to "hardware companies that build AI", not from bonds into equities.

XLY −1.13% deserves a separate note. Pre-market attention to consumer fell only on the four single names ONON, KTB, CAVA and EAT, while the entire consumer discretionary sector was second-worst of the day. The strength of CAVA (+14.24%) and EAT (+11.07%) was purely stock-specific event-driven, the opposite direction to the sector — a point that was in fact already written into the pre-market call of "restaurant consumption (stock-specific), not a sector", and was confirmed today.

5.2 Risk Appetite Assessment

Assessment: risk-on at the surface, rotation underneath.

Supporting risk-on: inflation landed, rates fell, front-end volatility collapsed, small caps strengthened. But at the same time: the indices ground lower all day, the Dow closed down, gold led, back-end volatility did not fall, and three of four after-hours earnings reports were sold.

This is not a "good news exhausted" day, it is a "good news being exhausted" day. The key difference: today's money is still willing to pay for new information on AI hardware (NBIS +34%, SMCI +19%), but is no longer willing to pay for confirmatory information (CRWV +0.66% after the open, CSCO and COHR down after hours). The dividing line is not the theme, it is "is this the first time we are hearing it".

---

6. Next-Day Outlook (2026-08-13, Thursday)

6.1 Theme Continuity

6.2 Tomorrow's Calendar

Macro (08:30 ET, pre-market):

⚠️ Why PPI is more worrying than today's CPI: after CPI came in line today, the market cut September hike odds from roughly half to about 40%, and front-end volatility (VIX9D 11.09, VIX1D briefly 7.60) has been crushed to almost no event premium at all. In other words, a "benign PPI" is already fully priced and will not produce further upside; while an "upside PPI" would meet a front-end volatility structure with no cushion whatsoever. That is an asymmetric risk distribution, skewed to the downside.

Earnings (after the close):

• AMAT (Applied Materials) FQ3'26 — the company previously guided to revenue of about $8.95 billion and non-GAAP EPS of about $3.36. AMAT is already up 4.29% today to $548.15, going into earnings already up, structurally identical to tonight's CSCO / COHR.
• TPR (Tapestry) — against today's backdrop of XLY −1.13%, an independent read on the consumer.

6.3 Key Watch List (Ticker + Explicit Validation Point)

6.4 What to Avoid

1. Anything "going into earnings with a double-digit gain". Today's daytime session (CRWV net-contributing 0.66% after an +18.49% gap, CAVA −3.20%, HYLN −21.10%) and evening (COHR, CSCO, CBRS) have demonstrated the same thing with six samples. AMAT tomorrow evening is the next sample.
2. The AI power derivative chain (ETN / CEG / TLN / VST). Zero independent catalysts, and it barely moved on the day the theme was strongest — names like these fall faster when the theme turns.
3. Software / SaaS. No inflection yet in the structural derating, 5/5 down today.
4. Energy. Two consecutive days of "oil up, stocks not"; the geopolitical premium can no longer transmit into share prices.
5. Consumer discretionary at the sector level (XLY). Second-worst of the day today; the strength in CAVA / EAT is a single-stock event and should not be extrapolated to the sector.
6. Small caps after a one-day surge. HYLN went from +25.77% pre-market to −0.77% at the close today, the most complete sample of the week.

6.5 Input Notes for Tomorrow's Pre-Market List

1. Add coverage of DELL / HPE / ANET — today the entire AI server chain was written up with only one name, SMCI, and that call was wrong. And change one thing methodologically: a bearish argument on a given leader may not be automatically extrapolated into a bearish view on the chain it belongs to; the two must be argued separately.
2. Establish "semiconductors / memory" as a standalone theme — today SMH +2.08%, MU +4.92%, CRDO +8.26%; it has always been used as background beta to rule out single-stock attribution, and never treated as a theme in its own right.
3. The test for "has the expectation gap been consumed" must change. Today, using "the news was pre-disclosed on 7/21" to infer that expectations had been reset on SMCI was wrong — you must directly check whether sell-side consensus numbers actually marked down, not whether the news had been made public.
4. Strength labels must be consistent with risk descriptions. Today we gave AI power an A+ while also writing that it has "weak independent catalysts, already delivered yesterday" — when the two conflict, defer to the risk description and cut the strength.
5. Keep the "vs pre-market price" reconciliation column. Today it corrected an apparent 94.7% hit rate down to 66.7%, the single most important debiasing tool in the whole issue.
6. Do not attach binary conclusions to C-grade evidence. The KTB avoid was built on a 0.5% revenue miss with evidence graded C, yet it received a definitive "avoid" conclusion. When the evidence is not strong enough, the right move is to leave it off the list.
7. Track the actual progress of the FCC optical module ban — the earnings driver for the optical chain has now twice been shown to "beat but not beat the pricing", and policy is the only variable left on this main line.

Data-Retrieval Failure Log (Internal)

Channel status for this retrieval run:

1. yfinance unavailable throughout — YFRateLimitError: Too Many Requests, rejected on the very first call, no data obtained. This issue used no yfinance at all; of the "WebSearch + yfinance" pipeline written into the skill, only the WebSearch half worked today.
2. The CNBC quote API is dead — quote.cnbc.com/quote-html-webservice/... returns an Akamai Access Denied (Reference #18.b8823417). This means the remembered conclusion that "the CNBC quote API is the most reliable substitute" no longer holds, and that memory needs updating. Cross-assets (oil/gold/dollar/bitcoin) therefore switched to ETF proxies (USO/GLD/UUP/IBIT), with reduced precision, faithfully flagged in the conventions section of the main text.
3. stooq is still behind JS PoW anti-scraping (consistent with existing memory); both the /q/l/ and /q/d/l/ paths were unusable.
4. CNBC article pages 403 (market close wrap, Cerebras earnings page), TheStreet 403, stocktitan DNS timeout, Yahoo Finance live page 429, businesswire 60s timeout, investor.cerebras.ai DNS does not exist, slickcharts and the stockanalysis index pages returned no grep output.
5. The four channels that actually worked this time, recommended for hard-coding into the skill:
   • stockanalysis.com /api/quotes/s/<sym> — OHLC + volume + after-hours price (ep/ecp/eu) for single stocks and ETFs, fully usable after the close, the workhorse channel of this issue. Note that its /i/<index> index paths all return 400.
   • CBOE cdn.cboe.com/api/global/delayed_quotes/quotes/_<SYM>.json — _SPX/_NDX/_RUT/_VIX/_VIX9D/_VIX1D/_VIX3M are all usable and carry OHLC.
   • treasury.gov daily yield curve CSV — official primary source, all tenors, the most stable.
   • api.nasdaq.com /api/quote/<SYM>/info?assetclass=index — COMP and NDX work (DJI/SPX/RUT/IXIC all return "Symbol not exists").

⚠️ Three traps hit and dodged this time, worth recording separately:

• CBOE's _DJI and _COMP return stale data without raising an error — _DJI's last_trade_time is 2026-07-31, _COMP's is 2021-03-23, yet both have current_price / price_change fields filled in like normal data. This is exactly the failure shape described in the "silently stale data source" memory, except the lag is not one day but two weeks and five years. It was found only by checking last_trade_time field by field — that field must be checked every single time, you cannot just take the price.
• The breadth data source contradicted itself — one source claimed "US stocks closed lower on 8/12, tech led the decline, advance/decline 40.4% vs 56.3%", while three independent sources all confirmed S&P +0.26% and Nasdaq +0.54%. The whole line was discarded, replaced with the RSP/SPY proxy and flagged in the main text.
• A second-hand summary conjured up an NBIS guidance out of thin air — search results contained "NBIS gave revenue guidance of $14.5–15.5 billion, beating the $11.68 billion expected", nearly five times the full-year $3.0–3.4 billion the company actually reiterated. What exposed it was the order-of-magnitude comparison, not the news check — the same class of failure as "a search summary can conjure up an entire earnings report". Two further second-hand numeric conflicts were handled in the main text on a primary-source basis: SMCI's closing price (one source said $38.09/+20.52%, which is actually an intraday price near the day's high of $38.15; the primary OHLC basis is a close of $37.61/+19.02%) and CSCO's Q4 AI orders (one source said $3.7 billion, the press release says $4 billion).

Items therefore left blank in this issue: raw NYSE advance-decline counts, the Dow's intraday high/low, and the definition of Cerebras's core revenue measure. The first two have been filled with proxy indicators and flagged in the main text; the third is marked pending verification.

---

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