U.S. Market Recap | 2026-08-26 (ET), Wednesday

Coverage window: 2026-08-26 09:30–16:00 ET regular session + after-hours through 17:20 ET. ⚠️ This piece went to press while the NVDA earnings call was still in progress (the call started at 17:00 ET). After-hours prices were still moving violently; §4 tags the read moment on every single line, and after-hours percentage moves are not final values. Price sources: single stocks and ETFs from the stockanalysis.com quote API (16:00 ET close basis); indices from CBOE delayed quotes and the official Nasdaq API; Treasury yields from the U.S. Treasury's official daily yield curve; all earnings figures taken from the original SEC EDGAR 8-K / 10-Q filings of the day, not from media paraphrase.

---

0. Recap in One Line

1. At the index level this was the flattest day of the year, but internally a theme handover was completed. S&P 500 closed 7,675.70 (−0.02%), Nasdaq Composite 26,130.20 (−0.08%), Nasdaq 100 29,224.52 (+0.05%), Russell 2000 3,005.90 (−0.14%) — all four indices landed inside ±0.15%. But among S&P 500 constituents (501/503 actually measured), 281 advanced, 217 declined, equal-weight +0.21%, and the top of the gainers list was occupied wholesale by optical communications and networking hardware: LITE +6.04%, ANET +5.92%, FFIV +4.40%, CIEN +4.22%, GLW +3.82%, plus storage names WDC +4.02%, STX +3.01%, NTAP +3.43%. This is exactly the theme that the pre-market list ranked 6th and downgraded to "already priced in" — the single largest misjudgment in this report.
2. 🔑 Of NVDA's three verdict numbers, two went the bulls' way and one went the bears' way — and the one that went to the bears is precisely the gross margin the pre-market piece flagged as "the only one where a single direction can bring new information." Quarterly revenue $96.221 billion (+106% YoY), beating the midpoint of company guidance by +5.74% (the last four quarters ranged +3.87%~+5.57%, a new high this quarter); next-quarter guidance $108 billion, above consensus by +4.05%; but Q3 non-GAAP gross margin guidance of 74.0% (±50bp), −100bp QoQ, landing squarely inside the bear range set pre-market (≤74.5%). This explains why the stock went flat first and only then rallied after hours instead of gapping straight up: the two forces genuinely collided inside the same press release.
3. That pre-market crack labelled "SaaS guidance dulling" was closed today by its own falsification condition. Pre-market measured software equal-weight −1.77%, semiconductors −0.67%, a gap of 1.10pp, and wrote explicitly: "if it converges to within 0.5pp → this morning's software selloff is mainly INTU/ZM single-stock spillover and does not constitute a sector conclusion." Measured at the close: software (excluding the 6 names with their own earnings that evening) 21 names equal-weight +0.29%, semiconductors 20 names equal-weight +0.30% — a gap of 0.01pp, essentially fully closed. After hours it reversed further: software 27 names equal-weight +3.47% (24 up, 2 down).
4. All three 08:30 data releases landed, and not one of them constituted a surprise: July core PCE +0.2% m/m / +3.3% y/y, exactly in line with expectations and with June; headline PCE 3.7% y/y, slightly above the expected 3.6%; Q2 GDP second estimate +1.5%, unchanged from the advance print; July durable goods orders +1.1%, materially better than the expected +0.5%. Inflation neither deteriorated nor improved, which is exactly why the indices sat pinned to the flat line all day.
5. Tone for the next session: leaning risk-on, but the risk exposure has moved from "earnings" as a whole to "Warsh." VIX closed 15.21 (−1.55%), 10Y Treasury 4.66% (+2bp) — neither shows a stress signal. The four software earnings already delivered after hours (CRM/CRWD/OKTA/VEEV) will produce a positive gap at tomorrow's open, but 36 hours later it all gets handed over to Warsh's debut at 08:00 ET on 8/28.

---

1. Market Overview

1.1 Indices and Rates (read moments tagged item by item — they are not on the same clock)

Note on the Dow: across all three index data channels used, none returned a valid close for the Dow today (the CBOE Dow quote timestamp stops at 11:15:52 ET, covering only the morning, and was discarded). This piece gives no Dow index level, and uses only its ETF (DIA) close of −0.19% as a directional reference. This is a data gap disclosed as such.

1.2 Market Breadth (full-sample measurement, not a sample)

Taking the day's close-to-close move for every S&P 500 constituent, measured coverage was 501/503 (missing BRK.B and BF.B, two dual-class names):

🔑 There is a divergence here that has to be spelled out: equal-weight is +0.21%, while the cap-weighted S&P 500 is −0.02%. The 0.23pp difference comes entirely from the drag at the mega-cap end (NVDA −1.59%, LLY −3.59%, GOOGL −1.43%). But this does not mean "small caps outperformed" — Russell 2000 was −0.14%, IWM −0.10%; genuine small caps did not outperform. The accurate statement is: mid-weight stocks inside the S&P 500 were the strongest today, while both tails (mega-cap and small-cap) lagged.

1.3 Full Sector ETF Ranking

Additional reference: IGV (software) +0.53%, SOXX (semiconductors) +0.26%, SMH (semiconductors) −0.01%, JETS (airlines) −1.10%, USO (crude) +0.95%, BNO (Brent) +0.56%.

Sentiment read: mildly risk-on, but very weak in intensity. Of 11 sectors, 5 up and 6 down; VIX down, yields slightly up, dollar marginally higher — none of the three gives a directional signal. Today's real information is not at the index level but in the ordering inside the sectors.

---

2. Pre-Market List Reconciliation

2.1 Name-by-Name Reconciliation (including a "vs pre-market price" column)

Why the "vs pre-market price" column is mandatory: reconciling on close-to-close moves alone systematically overstates the quality of avoid-type calls — a name already down 9% pre-market that closes down only 3% looks, on close-to-close, like "it fell, call was right," but executing it as written at the open was actually a loss. The table below gives both columns side by side.

Pre-market prices are the 08:27–08:36 ET re-check values from the 2026-08-26 pre-market list.

Long-leaning

8 long-leaning names: today equal-weight +4.68%, vs pre-market +2.80%. 🔴 But that number has to be broken apart: excluding ANF, the remaining 7 names are only +0.26% equal-weight today and +0.04% vs pre-market — the entire return of the long-leaning list came from ANF alone. This is not a "broad-based" list; it is a "hit one" list.

Short-leaning / Avoid

🔴 The 11-name avoid list: today equal-weight −0.72% (looks right), vs pre-market equal-weight +2.60% (execution was actually a loss); 8 of the 11 rose from their pre-market price.

This is the most important methodological conclusion in this piece: the "accuracy" of the avoid list came almost entirely from that pre-market gap, and that gap had already happened by the time the list was published. Judged on close-to-close moves, the avoid group was 6 down / 5 up and looks acceptable; judged on the executable pre-market price, it is a systematically inverted list.

Watch Only

10 watch-only names: today equal-weight +1.18%, vs pre-market −0.66%. The avoid judgment on the two nuclear names (OKLO/SMR) was the highest-quality call — the pre-market piece stated explicitly that "the catalyst was verified as having no contract, no order, no revenue," and on the day the two fell −6.03% / −5.50% respectively, the most accurate fundamental rebuttal on the entire list.

2.2 Hit Rate and Self-Critique

Self-critique in one line: today's list performed very well on "fact-checking" and very poorly on "theme ranking," and these two scores should not be averaged into a single number.

Verification side: AMD's rating catalyst was verified as containing no new information (+0.37% on the day, right); the nuclear names were verified as having no contracts and no revenue (−6.03% / −5.50% on the day, right); INTU's −39% guidance was falsified against the original SEC filing with a "don't chase the short" call (+7.08% vs pre-market on the day, right). All three were only right because of going back to the primary documents — the highest value-add.

Ranking side: the strongest theme of the day (AI networking / optical communications) was ranked 6th and actively downgraded to "already priced in," while one of the weakest themes of the day (oil price pass-through) was ranked 4th and given two watch-closely slots. See §3.

🔴 SMTC deserves a separate critique — this was a failure of the type "the re-check mechanism was right, but the conclusion was led astray by the re-check." Pre-market at 08:08 ET it printed +4.10%; the 08:27 ET re-check turned it red to −1.20% with volume rising from 419,000 to 536,000 shares, on which basis it was downgraded from "watch closely" to "watch only," with corresponding corrections made in five places. It closed +10.41%, and +11.76% versus the pre-market price.

• The act of re-checking is not itself wrong; on 8/25 (WSM) it genuinely saved a conclusion;
• What was wrong was treating 19 minutes of pre-market price movement as a reason to overturn a three-for-three earnings report of "record revenue +33% + Q3 guidance $50 million above consensus + operating margin +660bps QoQ." The weight given to price evidence was placed above fundamental evidence;
• More notably: SMTC belongs to the optical communications / AI networking sector, and that sector was +3.92% equal-weight on the day. The downgrade of this one name and the downgrade of the entire theme are two expressions of the same judgment error, not two independent errors.

---

3. Theme Verification for the Day

3.1 Did the Pre-Market Piece Get the Strongest Theme Right? — No

The "three strongest themes today" given pre-market were: ①NVDA earnings event ②SaaS guidance dulling ③oil giving back the war premium.

The actual strongest themes of the day ranked by equal-weight gain were: ①AI networking / optical comms (+3.92%) ②storage hardware (+2.76%) ③industrials (XLI +1.09%).

• NVDA event: intraday it was a negative contributor (NVDA −1.59%), with pricing pushed wholesale to after hours. As an "event," ranking it first was not wrong; as "the strongest theme of the day," it was ranked wrong.
• SaaS guidance dulling: net zero on the day.
• Oil giving back the premium: inverted on the day.

The exact wording used pre-market to downgrade AI networking / optical comms was: "the strongest earnings report in the theme (SMTC) had already turned red pre-market at the single-stock level — showing the theme was fully priced before yesterday's close." 🔑 That chain of reasoning has one clear defect: it uses "the pre-market price of a single stock" to infer "the pricing progress of the entire theme." And on the day, none of LITE, ANET, CIEN, FFIV, GLW, WDC, STX, NTAP had earnings of their own — all were follow-on repricing. The theme's fund flow was simply not present in SMTC's pre-market quote.

3.2 The "Software vs Semiconductors" Gap: The Pre-Market Falsification Condition Hit Itself

The criterion written out in §8 pre-market (reproduced verbatim here):

If 30 minutes after the open the gap widens beyond 2pp and the count of declining software names is still 20/23 or higher → the divergence is real fund flow. If it converges to within 0.5pp → this morning's software selloff is mainly INTU/ZM single-stock spillover and does not constitute a sector conclusion.

Measured result:

✅ Verdict: the falsification condition holds. This morning's software selloff was indeed mainly INTU/ZM single-stock spillover and does not constitute sector-level fund flow.

This should be recorded as a plus for the pre-market list, not a minus — it did not write an unestablished lean up as a conclusion; it wrote down in advance the threshold that would overturn itself, and kept the label at "pending verification, not settled." What genuinely saved this list today was that criterion, not that conclusion (consistent with past experience: write the verification point first, then reason back to the conclusion).

After hours this crack opened further in the opposite direction: software 27 names after-hours equal-weight +3.47% (24 up 2 down), semiconductors 20 names after-hours equal-weight +1.58% (20 up 0 down) — software instead became the strongest side after hours.

---

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

⚠️ The read moment matters enormously here: the after-hours moves in the table below are the latest per-name reads between 17:12–17:20 ET (last trade times differ by name); the NVDA call began at 17:00 ET and was still in progress. NVDA's measured path in the 20 minutes after hours was oscillating, not one-directional: 17:01 +0.03% → about 17:10 +4.78% → 17:17 +3.08% → 17:20 +4.11%; over the same span CRWD pulled back from +11.27% to +9.53%, and SNPS narrowed from −2.93% to −2.20%. These numbers are not final values; after-hours prices accumulate continuously and repeatedly, and no single snapshot constitutes a fact.

4.1 NVDA: The Three Verdict Numbers Set Pre-Market, Adjudicated One by One

All data taken from the original 8-K / EX-99.1¹ and the original 10-Q².

Actuals for the quarter (FY27 Q2, ended 2026-07-26)

🔑 The supporting referee rules first: the pre-market piece wrote that "if quarterly revenue comes in below $94.5 billion (a guidance beat of less than +3.87%), the foundation of the bulls' low-bar argument collapses on the spot." Actual was $96.221 billion, a +5.74% guidance beat — not only did it not collapse, it was the largest guidance beat of the last four quarters (the prior four were +3.87% / +5.57% / +4.81% / +4.63%).

The three verdict numbers

Criterion 1 in detail — the absolute level cleared the bar, the QoQ did not, and neither is an illusion: Guidance of $108 billion is exactly the bull absolute-level threshold set pre-market; but because the actual quarter ($96.221 billion) came in above the $95.0 billion assumed when the pre-market math was done, the same guidance translates into a diluted QoQ of +12.24%, below the +13.7% threshold. Compared against the same-basis historical series (guidance-implied QoQ): the last four were +15.5% / +14.0% / +14.5% / +11.5%, so +12.24% ranks fourth of five, but is above the most recent +11.5% — a modest recovery off the low, not further deceleration.

The pre-market piece stressed repeatedly, "don't look at the headline absolute number, do the division." Having done the division today, the conclusion is: the deceleration narrative was neither confirmed nor refuted.

Criterion 2 in detail — this is the only number tonight that delivered new information: Q3 non-GAAP gross margin guidance of 74.0%, −100bp QoQ against Q2's actual 75.0%, ending the state of "pinned at 75.0–75.1% for three consecutive quarters." The pre-market reasoning was: "genuine demand expansion should let 75% hold or move up; demand self-funded off the balance sheet will show up first in gross margin via HBM4 costs and the Rubin ramp." The press release simultaneously confirmed that Vera Rubin has entered full production. 🔑 Two things hold true at the same time today: revenue and guidance were stronger than expected, and gross margin fell just as the bears expected. That is why it went flat after hours and only strengthened once the call began — the market needs management to explain whether those 100bp are ramp tuition or a trend.

Criterion 3 in detail — two of the three sub-items favor the bulls, but the magnitude of the third scraps the criterion itself:

🔴 That $279 billion figure exceeds the design range of the pre-market criterion — it satisfies both "the $130 billion+ the bulls wanted" and "the >$140 billion the bears wanted." When both sides of a criterion are triggered at once, the correct move is to admit it cannot referee that night, not to pick a side to use. Only one thing can be established: this is a supply-side statement by the company, expressed through purchase obligations, at more than double the prior scale — it is simultaneously the strongest demand signal and the largest single irrevocable risk exposure.

🔴 There is one more change in the 10-Q that the pre-market criteria did not cover at all but whose direction is clearly adverse — accounts receivable:

And the 10-Q gives the reason in the same paragraph: "for investment-grade customer purchases, we have and may in the future provide longer payment terms ranging from 90 days up to one year to assist customers with large data center builds". ⚠️ Pre-market, "DSO improving from 51.4 days to 45.4 days" had been written up as "the classic 'stuffing the channel to prop up revenue' challenge is directly falsified by the data." This quarter that falsification was reversed. Receivables concentration is rising at the same time: five direct customers account for 22%/14%/13%/11%/10% of receivables, 70% in total (the prior period was three customers at 25%/18%/13%). Revenue concentration is falling while receivables concentration is rising — the two happening together means the risk has shifted from "who is buying" to "who has not yet paid."

Other 10-Q highlights (all primary-source):

• The SB Energy guarantee cap of $105 billion is confirmed in the body of the 10-Q: 4.25GW IT load, the Pike County PORTS campus in Ohio, tenant is an OpenAI affiliate, in nine tranches, the first tranche expected to take effect in FY2029 (pre-market said "from 2028"; the 10-Q governs), with the guaranteed amount declining over each tranche's 20-year lease term. There are also land/power/facility guarantees to an AI cloud partner, with maximum total exposure of $3.5 billion.
• Inventory $31.575 billion, of which raw materials $11.341 billion; inventory turnover 119.3 days, which has not yet breached the falsification line given pre-market (about 125 days), so the "stocking up for the new architecture" explanation still holds for now.
• The balance sheet is becoming further "financialized": long-term liabilities $7.469 billion → $32.366 billion; non-marketable securities $22.251 billion → $51.157 billion; marketable equity securities $12.886 billion → $42.783 billion. Q2 other income, net $7.773 billion (year-ago period $2.766 billion) — this is also why GAAP EPS ($2.46) is higher than non-GAAP EPS ($2.22); discussing NVDA's earnings quality on a GAAP basis will distort the picture.
• Newly disclosed (not covered pre-market): an independent compute-financing platform established with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR, intending to mobilize over $500 billion of third-party capital, subject to definitive agreements. This is the most important new fact tonight on the "circular financing" thread.

4.2 CRM: $2.53 of the Headline EPS +103% Did Not Come From Operations

Original 8-K EX-99.1³.

🔴 One thing that must be unpacked: Salesforce's non-GAAP basis does not exclude gains on strategic investments. Footnote (3) of the press release states explicitly that this quarter's strategic investment gains contributed $2.43 to GAAP diluted EPS and $2.53 to non-GAAP diluted EPS (both $0.00 in the year-ago period). Restated:

And the Q3 non-GAAP EPS guidance of $3.42–3.44 is given on a "no further portfolio movement" basis — comparing it against $5.90 produces the illusion of "halving QoQ," while the correct comparison is against the clean $3.37, roughly +1.8% QoQ. ⚠️ Another dilution item: about 4pp of the +11% revenue growth comes from the Informatica acquisition (full-year guidance likewise includes a "slightly more than 3pp" Informatica contribution), so organic growth is roughly 7–8%. 🔑 Conclusion: cRPO accelerating to +14% is a real signal and deserves the +12% after hours; but the headline "non-GAAP EPS +103%" is essentially a revaluation of the Anthropic stake, not an operational improvement. If media or sell-side notes tomorrow extrapolate year-over-year off $5.90, that is wrong.

4.3 CRWD: The Single Criterion Set Pre-Market Was Beaten by a Wide Margin

Pre-market stated: "net new ARR guidance of $284–286 million is the sole criterion." Original 8-K⁴: net new ARR $333 million (an all-time record), with YoY growth accelerating to +51% — beating that threshold by about +16.8%.

• Falcon Flex cohort ending ARR over $2.29 billion, +101%
• Quarterly OCF $530 million, FCF $377 million, both records
• FY27 net new ARR growth guidance raised 630bp to a midpoint of +34% (it was +27.7% at Q1)
• Q3 guidance: revenue $1.5232–1.5292 billion, non-GAAP EPS $0.31; FY27 revenue $5.9911–6.0111 billion

✅ The criterion was set precisely and the conclusion was confirmed. The only regret is that the label stopped at "watch only," so the +9.53% after hours was not participated in.

4.4 The Other Four

• OKTA (after hours +19.04%, strongest of the day): revenue $805 million +11% (subscriptions +12%); cRPO $2.585 billion +14%, which the company explicitly calls "accelerating"; non-GAAP operating margin 28%; non-GAAP EPS $1.05 (year-ago $0.91); FCF $227 million. FY27 revenue guidance $3.216–3.226 billion (+10~11%), non-GAAP EPS $3.90–3.94. ⚠️ Two things to note: ① Q3 cRPO guidance is +11~12%, below this quarter's +14%, with the company saying it "continues to take a prudent approach"; ② the after-hours price of $160.01 is already above its 52-week high of $157.00 — so tomorrow it opens in all-time-high territory, not from a low base.
• VEEV (+8.34%): revenue $928 million +18%, subscriptions $766.8 million +16%; non-GAAP operating income $415.9 million +18%; non-GAAP EPS $2.35 (year-ago $1.99); the CFO said "every metric exceeded guidance and the full-year outlook was raised across the board." FY27 guidance: revenue $3.682–3.687 billion, non-GAAP EPS about $9.21.
• NTNX (+7.05%, FY26 Q4): ARR $2.55 billion +16%; revenue $757.1 million +16%; non-GAAP operating margin 26.2% (+790bp); FY26 FCF $840.7 million. FY27 guidance: revenue $3.180–3.230 billion, non-GAAP operating margin 24–25%, FCF $850–950 million.
• 🔴 SNPS (−2.20%, the only decliner): FY26 Q3 revenue $2.477 billion (year-ago period $1.740 billion, includes Ansys), non-GAAP EPS $3.91, above the top end of guidance, and it raised the full year to a revenue midpoint of $9.715 billion and a non-GAAP EPS midpoint of $15.07. A beat-and-raise that earned an after-hours decline. Q4 guidance: revenue $2.530–2.580 billion, non-GAAP EPS $4.10–4.16. ⚠️ This piece did not obtain sell-side consensus for SNPS Q4, and therefore cannot determine whether the after-hours decline is "Q4 guidance below consensus" or "good news out, distribution." This is a gap disclosed as such, to be filled tomorrow. The only confirmable clue is the press release's own statement that the Design IP business only just "returned to year-over-year growth" this quarter, making it the company's weakest line.

---

5. Flows and Sentiment

5.1 Sector Rotation: Today's Money Was Buying "the Physical Layer of AI"

What led was not the compute chips, but the things that connect and store the compute.

Over the same span SMH −0.01%, SOXX +0.26% — the semiconductor ETFs did not reflect this theme at all, because the theme is not at the chip-design end. This is also the structural reason why watching the AI chain through SMH/SOXX pre-market misses today's move.

Another underrated clue: the divergence inside energy. XLE +0.60%, but everything that rose was midstream pipelines (WMB +4.68, KMI +3.49, OKE +3.49), while oil services OIH was −0.20%. This is not "higher oil prices lifting energy stocks"; it is natural gas / power infrastructure being pulled along by the AI data center narrative — and it is the same leg as XLI +1.09% (FIX +3.44, EME +3.21, GEV +2.84 are all data center engineering and electrical equipment).

🔑 Put these four groups together and there is really only one theme today: the market is repricing "the physical bottlenecks of the AI build-out" — optical modules, switches, hard drives, transformers, pipelines, mechanical and electrical engineering. It is not pricing "whose chip is better."

5.2 The Declining Side

Health care was the only sector down more than 1% (XLV −1.00%): MRNA −5.77%, LLY −3.59%, REGN −2.25%, BSX −3.39%. Pre-market picked BSX at the single-stock level but did not identify health care as a theme-level weak line. Other single-stock weakness: RDDT −4.36%, GDDY −4.26%, HOOD −3.17%, COIN −2.87%, ACN −2.97%, SMCI −2.78%.

5.3 Risk-On / Risk-Off Characterization

Characterization: mildly risk-on, but not strong enough to call a trend. The more accurate description is — the whole market was waiting on one earnings report; money did not leave, it just changed seats. And the seats it changed into (networking / storage / power / engineering) are exactly the directions directly endorsed by tonight's NVDA press release — Vera Rubin at full production, the Spectrum-6 switch, the $500 billion financing platform. Intraday fund flow and after-hours fundamentals pointed to the same place today.

---

6. Next-Day Outlook (Thursday 8/27)

① Theme Continuity

② Tomorrow's Earnings and Macro Calendar

🔑 Why MRVL is tomorrow's key: it sits simultaneously on today's strongest line (AI networking) and today's most conflicted line (semiconductors), and it was already +1.97% today plus another +2.38% after hours. Its custom ASIC / optical interconnect guidance is the most direct primary-source test of whether "the AI networking theme is an independent move or an NVDA shadow."

③ Focus List (Ticker + Verification Point)

④ What to Avoid

1. Chasing NVDA's after-hours gain — the after-hours price walked an oscillating path of +0.03% → +4.78% → +3.08% → +4.11% within 20 minutes; this is a live reaction to a call in progress, not a pricing endpoint; and the stock has been negative 8/8 on next-day "open→close" over the last 8 quarters.
2. Extrapolating anything from CRM's "non-GAAP EPS +103%" — $2.53 of that is a strategic investment revaluation; the clean year-over-year figure is +16.0%.
3. Airlines (UAL/DAL/AAL/LUV) and shorting energy — today JETS −1.10%, XLE +0.60%; both legs of the pre-market pass-through chain failed at once, and it should not be restarted before new evidence appears.
4. Nuclear concepts (OKLO/SMR) — down −6.03% / −5.50% today, confirming this was a weak-stock bounce and not a new trend.
5. Bottom-fishing health care — XLV was −1.00% today, the worst of the field, with no evidence yet of stabilization.
6. Treating "a software sector reversal" as a settled conclusion — today's after-hours software strength was driven almost entirely by five reports of their own; until software names with no news of their own (NOW/ADBE/WDAY/MDB) follow suit tomorrow, this is just a collection of single-stock events.

⑤ Input Notes for the Next Pre-Market List

1. 🔴 The scan universe must add the "AI physical bottleneck" layer: optical modules (LITE/COHR/FN/AAOI), switching and networking (ANET/CIEN/FFIV/GLW), storage (WDC/STX/NTAP/MU), power and electromechanical engineering (GEV/FIX/EME/VRT/PWR), midstream pipelines (WMB/KMI/OKE). 11 of today's top 15 gainers belong to this layer, and the pre-market list had zero coverage of most of them.
2. 🔑 Do not use a single stock's pre-market quote to infer the pricing progress of an entire theme. Today SMTC turning red pre-market was used to downgrade the whole AI networking theme, while that theme was +3.92% equal-weight on the day — a theme's fund flow is not inside any single name's pre-market quote.
3. The reconciliation table must give both a "Today %" and a "vs pre-market %" column. Today the avoid group is 6/11 right on the former and only 3/11 on the latter. Giving only one column produces the opposite conclusion.
4. Tomorrow's macro risk is not tomorrow: initial jobless claims have limited impact; the real variable is Warsh's debut at 08:00 ET on 8/28. The pre-market list should write it in explicitly as a "T+1 risk," noting that the positive reaction window for earnings is only about 36 hours.
5. CME September rate-hike probabilities need to be re-pulled. Today core PCE was flat at 3.3% and durable goods orders came in far above expectations (+1.1% vs +0.5%), and the two point in opposite directions for the rate path, so the 8/17–8/20 readings carried over in the pre-market list are no longer usable.
6. NVDA's follow-up criterion should switch from "will gross margin fall" to "once it falls 100bp, when does it come back," and add one item missing from the pre-market criteria: days sales outstanding (45.4 → 59.6 days this quarter) and next-quarter changes in the extended payment-terms provision.

---

⚠️ Risk disclosure: this recap is only a post-close information review and observation, and does not constitute investment advice. Data may differ in timeliness or definition; please rely on company disclosures / SEC filings, and do not use this directly as a basis for trading. Special note: this piece went to press while the NVDA earnings call was in progress (17:20 ET); every after-hours percentage move in the text is not a final value and may already have changed significantly before tomorrow's open.

---

Data Sourcing and Quality Log (Internal)

Errors Identified and Avoided (5 this time, all in the "date/timestamp" category)

1. 🔴 Zacks/Globe's "Stock Market News for Aug 26, 2026" is in fact the 8/25 close recap. The Dow 53,577.40 / S&P 7,677.24 / Nasdaq 26,151.30 it gives, as well as "NYSE advance-decline 1.71:1, volume 14.32 billion shares," are all 8/25 data (the text itself says "traded on Tuesday"). Confirmed by three-way cross-check against CBOE prev_day_close and the official Nasdaq API. Had it been adopted directly, the entire §1 table and the §1.2 breadth section would have been off by a full day. For the same class of trap see [[recurring-column-headlines-hide-date]], [[stale-source-silent-lag]].
2. 🔴 The barchart summary "S&P −0.60%, Dow −0.39%, Nasdaq 100 −1.18%, NVDA after hours −7%, revenue +122% YoY" is an old 2024-08-28 article. How it was caught: today's original 8-K gives revenue growth of +106%, while +122% corresponds to FY25 Q2 ($30.04 billion vs $13.51 billion, August 2024). Had this been believed, it would have inverted the entire characterization of the recap (writing a flat risk-on day up as a crash day). For the same class of trap see [[phantom-earnings-from-search-summary]], [[recurring-column-headlines-hide-date]].
3. 🔴 CBOE _COMP has a last_trade_time of 2021-03-23T14:33:09, five years stale. Its price fields (close 13,344.56) are filled in like normal values; only the timestamp gives it away. Switched to the official Nasdaq API for the Nasdaq Composite (26,130.20), and cross-validated Nasdaq 100 across CBOE _NDX (29,224.52) and the Nasdaq API (29,224.52). Freshness must be checked down to symbol granularity, see [[cboe-stale-symbol-silent-trap]].
4. 🔴 CBOE _DJI's timestamp stops at 11:15:52 ET, covering only the morning; its close field of 53,467.03 is a mid-morning value, not a close. Discarded, with the Dow level gap stated explicitly in the body, and no attempt made to back out a plausibly precise level from DIA and pass it off as an index close.
5. The search summaries "VIX fell 2.52% to 15.45" and "DXY fell to a 2.5-month low" both fail to hold. 15.45 was verified against CBOE as the 8/25 close (8/26 closed 15.21); the dollar's direction was verified in both directions via UUP (+0.29%) and FXE (−0.19%) as marginally higher, not sharply lower. Both were discarded.

The Most Valuable Primary-Source Checks This Time

• All three of NVDA's verdict numbers were taken from the original SEC filings, with no reliance on any media paraphrase: 8-K/EX-99.1 (revenue, gross margin guidance) + 10-Q (customer concentration, commitments, inventory, receivables, SB Energy guarantee).
• 🔑 The 10-Q surfaced one adverse change the pre-market criteria did not cover at all: DSO rose from 45.4 days to 59.6 days (receivables +54.9% QoQ vs revenue +17.9% QoQ), and the 10-Q itself states that investment-grade customers are offered extended payment terms of "90 days up to one year." Pre-market, last quarter's DSO improvement had been written up as "the channel-stuffing challenge is directly falsified" — this quarter that falsification was reversed. This is the only substantive risk item this time found by reading the original document rather than by search.
• 🔑 CRM's $2.53/share strategic investment gain is in footnote (3) of the press release, not in any headline summary. Copying "non-GAAP EPS $5.90, +103%" would turn a revaluation of the Anthropic stake into an operational beat, and would make the Q3 guidance of $3.42–3.44 look like a halving. This is the same shape as [[one-time-refund-manufactures-eps-beat]] and [[consensus-is-modeled-on-adjusted-not-gaap]], except that this time the one-off item sits inside the non-GAAP figure rather than outside it.
• CRWD's net new ARR of $333 million was taken from the original 8-K, directly comparable to the $284–286 million threshold set pre-market, a beat of +16.8%.

Data Sourcing Failures

• yfinance was rate-limited by Yahoo again on this machine (YFRateLimitError); a direct call to query1.finance.yahoo.com/v8/finance/chart returned HTTP 429, confirmed by measurement this time. This piece used yfinance nowhere at all, even though the skill workflow lists it as the default data source.
• Other channels that failed in testing: CNBC (both article and quote endpoints 403, Akamai blocking, consistent with everything since 8/12), stooq (JS PoW anti-scraping), Yahoo Finance article pages (429), the WSJ michelangelo API (500), stockanalysis index endpoints (/api/quotes/i/* returns 400 across the board; the site has no index endpoint).
• Working channels: stockanalysis /api/quotes/s/<T> (single stocks + ETFs, both close and after-hours fine), SEC EDGAR (submissions + Archives full text, entirely fine), Nasdaq api.nasdaq.com/api/quote/<S>/info (indices COMP/NDX available; DJI/SPX/RUT/VIX not), CBOE cdn.cboe.com/api/global/delayed_quotes (requires per-symbol timestamp validation), the U.S. Treasury yield curve XML (entirely fine).
• The full-sample S&P 500 scan took three passes: the first pass at 10 concurrent got only 382/506, with the gap concentrated at the tail of index weight (i.e. the small-cap end), a one-sided bias that would systematically overstate breadth if used directly; the second pass at 6 concurrent filled it to 432; the third pass switched to serial + backoff and filled it to 501/503. This is the same shape as [[em-clist-pagination-one-sided-bias]]: what pagination/concurrency failures drop is never a random sample.

Not Obtained / Left Blank (disclosed as such)

1. The Dow Jones Industrial index close for the day was not obtained; the body states the gap explicitly and gives only DIA −0.19% as a directional reference.
2. Sell-side consensus for SNPS Q4 was not obtained, so it is impossible to determine whether the after-hours −2.20% is "guidance below consensus" or "good news out, distribution"; the body notes this as pending.
3. Official NYSE / Nasdaq advance-decline counts and volume were not obtained (the only source that hit was the stale 8/25 article, discarded). The body substitutes a self-measured full-sample S&P 500 breadth reading (501/503) and states the basis.
4. Same-day settlement prices for Brent/WTI crude were not obtained (no usable endpoint). The body substitutes the ETF proxies USO +0.95% / BNO +0.56% for direction, and fabricates no settlement price. For the reliability of that method see [[etf-proxy-detects-stale-spot-quotes]].
5. The latest reading for CME September FOMC rate-hike probability was not obtained; this piece cites no probability figure and notes in §6⑤ that it must be re-pulled tomorrow.
6. NVDA call content is not included (still in progress at press time). Management's explanation of the 74.0% gross margin is the largest information gap in this piece and is listed as the top tracking item in §6③.

Methodology Log: Two Systemic Problems Exposed by This Reconciliation

1. 🔴 The "vs pre-market price" column changed the conclusion again, and this time it changed it on the avoid side. The prior lesson ([[recap-vs-premarket-price-column]]) recorded that it overstates the hit rate of the recommended group; this time is the mirror image — it overstated the hit rate of the avoid group: 6/11 right on close-to-close moves, only 3/11 on pre-market prices, with 8 of the 11 rising from their pre-market price. This column must be given for both sides, not only on the recommendation side.
2. 🔴 Pre-market, the strongest theme of the day was downgraded on the basis that "the only name in the theme with earnings had turned red pre-market." That was a judgment that used single-point price evidence to overturn multi-point fundamental evidence. SMTC's downgrade and the downgrade of the whole AI networking theme are not two independent errors but two expressions of the same reasoning defect. Related to but distinct from [[premarket-quotes-accumulate-snapshot-is-not-fact]]: that one says a snapshot is not a fact; this one says that even when the snapshot is a fact, a single stock's snapshot cannot represent the theme's pricing progress. This deserves to be captured separately.
3. ✅ The mirror-image positive case: pre-market wrote down in advance the threshold that would overturn itself ("converging to within 0.5pp falsifies it"), and today that threshold was hit precisely (measured 0.01pp). That kept an erroneous directional lean from becoming an erroneous conclusion. Entirely consistent with [[verification-point-beats-headline-conclusion]]: write the verification point first, reason back to the conclusion second; the verification point is worth more than the conclusion itself.