US Market Recap | 2026-08-25 (ET), Tuesday

Data basis: single stocks / ETFs / indices / yields / commodities are all taken from the CNBC quote API, read at 2026-08-25 16:10–17:08 ET (regular session closed); after-hours prices are separately tagged with their snapshot time. Earnings figures follow company press releases / SEC filings, with sources noted line by line.

---

0. One-Sentence Recap

Today was a "three markets each trading their own story" day: equities bought the chip front-run ahead of NVIDIA's earnings (SMH +1.65%, NVDA ended a seven-day losing streak at +2.19%), bonds bought the growth softening implied by consumer confidence falling to a seven-month low (10Y −7.5bp), and commodities bought the Iran diplomatic de-escalation (Brent −5.47%) — the three lines do not conflict in direction, but not one of them explains the other two. At the index level it was risk-on, but the narrowest kind of risk-on there is: the S&P +0.32% while the equal-weight S&P (RSP) closed down 0.07% — the mega-caps rose, the equal weight fell. The strongest theme was neither AI nor semis, but biotech (XBI +3.00%, MRNA +14.36%), and that was Day 5 of pricing in the 8/19 Phase III data, not a new event. At the single-stock level the day was dominated by DKS −30.68% — Dick's Sporting Goods closed at the session low and dragged the entire athletic footwear & apparel chain down with it (ASO −5.80%, UAA −4.66%, NKE −3.12%). Today's pre-market list was not generated due to a system failure; this report instead reconciles against the next-day verification points laid out in the 8/24 recap: 4 of 5 verification points hit (80%), but of the 7 avoidance calls only 5 were testable and only 1 of those was right (40%) — every miss was concentrated on the "avoid" side. Next-day tone: 8/26 is the week's only pivot day — core PCE and the second estimate of Q2 GDP both land at 08:30, and NVDA reports after the close. VIX at 15.45 (−2.52%) is back at the lows: the market took its protection off right in front of a double event.

---

1. Market Overview

Indices

Cross-checked against the corresponding ETFs, consistent: SPY +0.32%, DIA +0.30%, QQQ +0.62%, IWM +0.42%.

Market Breadth — the single most important number today

The index gain came entirely from the mega-caps. Russell 2000 +0.50% looks like decent breadth, but its divergence from the equal-weight S&P says that what was bought today was the two ends — "large-cap tech + small-cap biotech" — while the layer in between (the body of the equal-weight S&P: industrials, staples, energy, regional banks) weakened across the board: XLI −0.34%, XLP −1.06%, XLE −1.66%, KRE −0.58%.

📌 NYSE / Nasdaq advance-decline (A/D) figures were not obtained for this report — the primary source was inaccessible. The RSP−SPY spread above is a proxy basis, not true breadth data. This gap is now in its second consecutive trading day.

Volatility, Rates and FX

Curve shape: the bull flattening continued. The 10Y fell 7.5bp, faster than the 2Y's 6.0bp, narrowing the 10s2s spread from 46.8bp to 45.3bp.

One call from yesterday was verified here: the 8/24 recap flagged that "the $69 billion 2-year auction at 13:00 would interrupt this week's bull flattening if the tail came in weak." It did not get interrupted — the 2Y fell 6.0bp on the day, front-end supply was absorbed smoothly, and the bull flattening stayed intact.

Sentiment Read

Risk-on on the surface, but underneath it is a "weakening growth + single-event bet" combination, not a healthy broad advance.

Three pieces of evidence:

1. The rate decline did not help the assets that should have benefited. The 10Y falling 7.5bp is not a small move, yet long-duration software IGV closed down 0.59% while semiconductors SMH rose 1.65%. If today's equity rally were driven by a lower discount rate, IGV should have outperformed SMH — instead it underperformed by 2.24pp. So today's equity gain cannot be attributed to rates; it is attributable to position adjustment ahead of NVDA's earnings.
2. Defensives split internally — this was not a flight to safety. XLU +0.21% (a mild gain) while XLP −1.06% (second-worst in the market). In a genuine risk-off tape these two move together. XLP's decline today has its own separate cause (see §3) and is unrelated to the bond rally.
3. Credit did not deteriorate. HYG +0.28%, TLT +1.10% — both up together. That is the shape of "rate-cut expectations warming up," not the shape of "risk-off selling."

Macro trigger: August consumer confidence at 89.4, below the 90.2 expected, a seven-month low. Within it, the expectations index plunged 5.8 points to 68.2 — the Conference Board itself notes that an expectations index below 80 typically corresponds to a recession within the next year; the present-situation index moved the other way, rising 6.8 points to 121.2, its first improvement in four months. At the same time, consumers' one-year inflation expectations rose from 5.6% to 5.8%.

This is a stagflationary reading: expectations deteriorating, inflation expectations rising, and the present situation improving. Bonds only bought the "expectations deteriorating" leg. Tomorrow's 08:30 core PCE will test the other leg directly — if the inflation print runs hot, today's 7.5bp gets handed back.

---

2. Pre-Market List Reconciliation

⚠️ Note on the basis of this section

The 2026-08-25 US pre-market list failed to generate; reports/us/2026-08-25.md does not exist. This section therefore cannot reconcile the day's list in the normal way.

Alternative approach: reconcile instead against the forward-looking calls that the 8/24 recap §6 made explicitly for 8/25 — namely the 5 "watch closely + verification point" entries and the 7 "avoidance calls." These are falsifiable judgments the system itself wrote down, and their reconciliation value is no lower than a list's; but they are not a pre-market list, their coverage is far narrower (5 names vs the usual 20–25), and their hit rate cannot be compared horizontally with historical list hit rates. This has to be stated plainly — a high score on a 5-sample set must not be passed off as a list hit rate.

2A. Reconciling the five "next-day verification points"

Verification-point hit rate: 4 / 5 = 80%.

2B. Reconciling the seven "avoidance calls"

Of the 5 testable calls: 1 right, 2 wrong, 2 half wrong = 40%.

Hit-Rate Summary and Post-Mortem

One line of self-criticism: 100% of today's misses were on the "avoid" side, and all three errors share the same root cause — mistaking "the narrative has run its course" for "the price will not move again."

Energy, the crypto high-beta end, and optical communications — the three lines that were avoided — delivered moves of 1.7%, 8.2% and 5.1% respectively today. The stated reason for avoiding each was "the event is fully priced / the correlation has broken down / there is no new catalyst." But "no new catalyst" does not mean "no volatility," especially when the reason for avoiding is itself "the correlation has broken down" — a breakdown is not a static state, it is a state transition, and the transition itself is tradeable.

Call 5 is the archetype: the 8/24 reasoning was "COIN/HOOD have already decoupled from the coin price → therefore avoid." Today they were still decoupled from the coin price (BTC −0.76% while all three rose), but the decoupling ran upward — because this group is no longer trading the coin price, it is trading its own fundamentals (Goldman reiterated Buy and raised its COIN target to $196 and HOOD to $124; prediction-market TAM; private-fund product lines). I wrote "no longer follows A" as "cannot be bought," skipping a step in the middle: "then what does it follow now?" That step went unasked, and a name that rose 8.17% on the day went onto the avoid list.

This and the 8/22 lesson ("can't find news" ≠ "there is no news") are two faces of the same failure mode: one mistakes missing information for a missing event, the other mistakes the end of a narrative for the end of volatility.

---

3. Theme Verification

Sector ETF panorama (descending by % change)

The ordering is itself the conclusion: biotech > semiconductors > tech hardware >> software / staples / energy.

Theme table

Dedicated section: today's biggest omission was biotech — and it wasn't even new news

XBI +3.00% was the strongest sector in the market, and MRNA +14.36% traded 46.40 million shares. Yet this line did not occupy a single slot among the 8 themes in the 8/24 recap.

What is more worth recording is why it was missed:

• The driving event was published on August 19 — Moderna and Merck announced that the individualized mRNA cancer vaccine intismeran autogene (V940 / mRNA-4157) in combination with Keytruda met both the primary endpoint (recurrence-free survival, RFS) and the key secondary endpoint (distant metastasis-free survival, DMFS) in the Phase III INTerpath-001 adjuvant melanoma trial, with 1,137 patients enrolled — the first positive Phase III result for an mRNA cancer therapy. MRNA more than doubled that day.
• Today (8/25) is not a data day, it is pricing Day 5. Today's direct trigger was Barclays analyst Eliana Merle raising the MRNA price target sharply from $48 to $125 (maintaining Equal Weight), layered on top of a bounce after the prior days' pullback and high-beta small-cap biotech absorbing the flow. MRNA is up roughly +392% year to date.
• The price structure says this was executable: MRNA opened at 143.50 (gap +3.32%) and closed at 158.83 — +10.68% of the move happened intraday — closing near the session high of 161.37. XBI likewise gapped +1.22% at the open, added 1.76% intraday, and closed near its high.

Lesson filed: when a main line is still delivering the market's biggest gain on Day 3–5 after its blow-up day, a scan window that only covers "overnight to pre-market" is structurally guaranteed to miss it. This is the same pit as the 8/18 lesson ("the scan window misses themes that ferment across days"), whose conclusion was "add 'pricing progress' as a second dimension"; 8/24 recap §6⑤ also put it on the to-do list — but no pre-market list was generated today, so that improvement never got a chance to be executed. This report puts it back into §6⑤.

Three things a pre-market list would most likely have missed

1. Biotech (reason as above: Day 5, not inside the overnight window)
2. The athletic footwear & apparel chain set off by DKS (pre-market earnings, unknowable in advance; but once the print is out, the cross-name transmission is executable in real time)
3. Crypto brokerage stocks (explicitly written onto the avoid list)

---

4. Earnings and After-Hours Moves

4.1 Intraday headline: DKS −30.68% — closed at the session low

Structural read: it gapped down 20.62% at the open, fell another 12.68% intraday, and closed near the low — no effective bid showed up all day. This is the exact opposite of the familiar "gap down on bad news, then repair intraday" pattern; it is the shape of selling pressure adding through the session.

Fundamentals (cross-checked across secondary sources; the SEC original was not retrieved — see the basis note at the end):

This decline was not caused by the earnings, it was caused by the guidance — and it forms the most valuable contrast pair of the day with INTU:

• DKS: EPS missed by 6.6%, full-year guidance midpoint 19% below → down 30.68%
• INTU: EPS beat by 12.3%, revenue guidance about 1.4% below → down 9.08% after hours

What they share is that the market priced only guidance today. What differs is that DKS's guidance cut is real, while half of INTU's "guidance blow-up" is manufactured by an accounting-basis change (see 4.2).

Revenue +53.2% has to be seen through: virtually all of that growth comes from consolidating Foot Locker, not from organic growth. The real information is that the legacy core business's comps were +4.9% (healthy) while the acquired Foot Locker's pro-forma comps were −3.6% (deteriorating). Executive Chairman Ed Stack attributed the cause to intensifying industry promotional activity and Foot Locker's over-reliance on sneaker launches, retro styles and older shoe models.

Cross-name transmission (actual today, not conjecture):

This is the only line today where "the event landed pre-market and the transmission played out fully during the session" — it does not contradict §2B call 1 (don't chase pre-market gaps >4%): you don't chase DKS itself, but the chain's cross-name transmission only developed during the session.

4.2 After-hours headline: INTU −9.08% — half the "blow-up" was engineered by an accounting-basis change

After-hours quote snapshot (2026-08-25 17:01 ET): $325.02, −9.08%, after-hours volume 1.89 million shares. The baseline is the pre-release close of $357.46 (the release came after 16:00; the −3.37% during the session was pre-positioning). Two-day cumulative: 369.92 → 325.02 = −12.14%.

⚠️ An after-hours quote is a reading accumulated up to the snapshot time, not a final value. Some media reported "down about 13%" at an earlier point and "down 7.3%" at a later point — all three numbers are correct, they just refer to different moments. This report uses its own 17:01 ET reading and tags the time.

FQ4 FY26 actuals (from the company's 8/25 press release itself):

Full-year FY26: revenue +14% to $21.4 billion; non-GAAP EPS +20% to $24.27; GAAP EPS +20% to $16.46.

Then the FY27 guidance — which must be taken apart:

The company simultaneously announced that effective August 1, 2026, stock-based compensation (SBC) will no longer be excluded from non-GAAP metrics. The company itself quantified the impact in a footnote to the guidance:

"Non-GAAP diluted earnings per share guidance includes a $5.81 impact from stock-based compensation expense" (full year); "includes a $1.48 impact" (Q1).

Restating to the old basis that sell-side models are built on (excluding SBC):

Conclusion: the widely quoted figure "Q1 guidance $2.44 vs consensus $4.02, 39% below" compares a new basis against an old-basis consensus and does not hold. Restated, Q1 misses by only 2.0%, and the full year is actually 5.5% above consensus.

This is the mirror image of the 8/21 lesson (consensus is modeled on the adjusted basis, not GAAP): last time a custom basis turned a decline into growth; this time a company changing its basis compressed a beat into a "blow-up." The test is unchanged: before comparing any EPS, confirm both sides are on the same basis.

So what actually deserves to be marked down? — Revenue, not EPS:

Three substantive negatives:

1. Growth decelerating from +14% to +9~10%, a 4–5pp slowdown — that is a hard fact.
2. Mailchimp is broken out as a separate reportable segment starting FY27, and is guided to −1% to 0% growth. For full-year FY26, GBS was +18% excluding Mailchimp and only +16% including it — breaking it out is putting the drag on the table.
3. Management explicitly said it will prioritize customer acquisition and share, even if that suppresses revenue growth and ARPU in the near term. This is a deliberate strategic choice, but the market's tolerance for "deliberately slowing down" has always been very low.

Two positives that should not be ignored either: the quarterly dividend was raised 15% to $1.38; FY26 buybacks were $5.5 billion (+96% YoY), cutting diluted share count by a net 2%.

INTU's positional context: 52-week high $705.08 (2025-09-22), 52-week low $252.84 (2026-06-22), today's close $357.46 — already down more than 44% year to date. This is not a stock sitting at the highs waiting to be knocked down; it is a stock that has already been cut in half and is being doubted over and over.

4.3 Other after-hours moves (snapshot 2026-08-25 16:50–17:06 ET)

ZM deserves its own line — it is the template for "good numbers, met with a decline":

FQ2 FY27 revenue $1.277 billion (+4.9%); enterprise revenue +7.8% to $787.5 million, the fastest in three years; online business +0.6% to $489.7 million; full-year guidance raised to $5.085–5.095 billion, non-GAAP EPS $6.08–6.12; Q3 guidance $1.275–1.280 billion. Revenue and guidance both beat, and it still fell 3.88% after hours.

⚠️ A number that must be avoided: ZM's GAAP net income this quarter was $1.542 billion with diluted EPS of $5.15, versus $358.6 million / $1.16 a year earlier. This is not operational improvement, it is a one-off item (far larger than the quarter's revenue itself; almost certainly an accounting event such as the release of a deferred-tax valuation allowance). Any reading that treats $5.15 as earnings power is wrong — use the non-GAAP basis and revenue growth instead.

4.4 Pre-market earnings (already priced today)

---

5. Flows and Sentiment

Sector rotation: buy the two ends, sell the middle

Buy side (two mutually unrelated poles):

• Small-cap biotech: XBI +3.00% > IBB +1.82%, XBI outperforming IBB by 1.18pp = the smaller the cap, the stronger, a textbook risk-appetite expansion
• Semiconductors: SMH +1.65%, driven by position rebuilding ahead of NVDA's earnings

Sell side (the body of the equal-weight S&P):

• XLE −1.66% (oil collapse), XLP −1.06% (giving back the one-day rotation), KRE −0.58%, IGV −0.59%, XLI −0.34%, XLY −0.30%

This pattern explains the 0.39pp divergence between RSP −0.07% and SPY +0.32%: both buy-side poles are narrow (SMH rests on a handful of heavyweights; XBI's gain was more than half contributed by MRNA alone), while the sell side is broad.

One attribution that can be falsified

The easiest sentence to get wrong today is "falling rates drove the equity rally." It does not hold today:

The three asset classes rates should have helped most (software, utilities, real estate) all ended flat-to-down today, while semiconductors — the group with the weakest link to rates — rose 1.65%. So today's equity rally was event-driven (positioning ahead of NVDA's earnings), while the bond rally was macro-driven (consumer confidence); the two occurred on the same day but are causally unrelated.

This test was already used once on 8/20 (then it was IGV −0.03% vs SMH −4.09% falsifying "duration de-leveraging"). Today is its positive application: the same yardstick, and on both occasions it showed that rates were not the main driver.

VIX: protection taken off right in front of a double event

VIX closed at 15.45, −2.52%, with an intraday low of 15.13.

Tomorrow has two gates to clear: core PCE + the second estimate of Q2 GDP at 08:30, and NVDA's earnings after the close. In front of a calendar like that, VIX did not rise — it fell back to just above 15. That means the market has already formed a consensus that "NVDA will give a good enough answer," and the consensus itself is the risk — pointing to the same place as the 8/21 lesson (IV crush comes from valuation, not pre-announcement; information content is measured by variance, not by headline count): when the cost of protection has been pressed to the lows, any imperfect answer gets amplified in the repricing.

risk-on / risk-off characterization

Characterized as "narrow-basis risk-on, and event-driven rather than trend-driven":

Net conclusion: today's risk-on rests on the single assumption that "NVDA delivers tomorrow," not on macro improvement — the macro actually turned worse today.

---

6. Next-Day Outlook (Wednesday, 2026-08-26)

① Theme continuity

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

Macro — tomorrow is the week's only double-event day:

Two datasets land in the same minute and their directions may cancel out — the first candle after 08:30 is often wrong; better to take the direction 30 minutes after 09:30.

Earnings:

NVDA (FQ2 FY27, after the close) — tomorrow's entire weight:

⚠️ This report makes no directional forecast on NVDA's results. A setup of a bounce after seven straight down days combined with VIX pressed to 15.4 means the odds are poor in both directions.

The other after-hours names tomorrow:

• CRM (赛富时 / Salesforce): −1.61% today. It sits in the same "large-cap SaaS guidance" group as INTU, and INTU has already given one answer — "deliberately slow down to buy share." If CRM offers a similar framing, the software divergence goes from two days to a trend.
• CRWD (CrowdStrike): −2.78% today, opening high and fading (−3.26% relative to the open) — the weakest performer today among tomorrow's after-hours names.
• SNPS (新思科技 / Synopsys): +3.62% today with nearly all of the gain made during the session. EDA is the only software asset in the semiconductor chain that strengthened in step with NVDA today.
• HPQ (惠普 / HP Inc.): +3.29% today. It is the cost-bearing side of the "memory price hikes" line — a PC maker's gross-margin guidance is independent evidence for testing the MU leg.

📌 One conflict with the 8/24 list, checked and handled conservatively: the 8/24 recap put OKTA on the 8/26 after-hours list, but CNBC event data marks OKTA's next earnings date as 12/01/2026 (est, an estimate), whereas CRM / CRWD / HPQ / SNPS / NTNX / VEEV / URBN are all marked 08/26/2026 with no estimate flag (confirmed). This report therefore does not count OKTA in tomorrow's list; defer to the company's IR announcement.

Later this week:

③ Watch closely (Ticker + tomorrow's verification point, for reconciliation the day after)

④ Avoid

1. Any directional position within 30 minutes of the 08:30 releases. Core PCE and the second estimate of Q2 GDP are released in the same minute, they may give opposite signals, and the first wave of prices is almost certain to contain mispricing.
2. Directional positions ahead of NVDA's earnings (listed for the second day running; today's opportunity cost of 2.19% has been booked). But the reason has changed: yesterday's reason was "catching a falling knife after seven down days"; today's reason is VIX 15.45 — cheap protection is precisely the sign that expectations are aligned, and surprises under aligned expectations amplify in both directions.
3. Any position built solely on "EPS beat" or "EPS miss." Today DKS (EPS missed by 6.6% → down 30.68%) and INTU (EPS beat by 12.3% → down 9.08% after hours) are two extremes of the same rule: today's market prices only guidance.
4. ⚠️ Any analysis citing INTU's "Q1 guidance 39% below consensus." That figure is an incorrect comparison of a new basis (including SBC) against an old-basis consensus; restated it is only 2.0% below, and the full year is actually about 5.5% above. If the market keeps selling tomorrow on that wrong number, that is an opportunity, not a risk — but note that the real negatives (revenue decelerating 4–5pp, Mailchimp turning negative) still stand; do not swing to the other extreme.
5. ZM's GAAP EPS of $5.15. It embeds a one-off item far larger than the quarter's revenue and is not earnings power. Any valuation based on it is wrong.
6. After-hours moves on volume below 200,000 shares never go on the list. Today's NOAH (−14.48%, 17,000 shares), HEI (+2.26%, 19,000 shares) and STRT / ELMD (volumes in the low thousands) all fall into this bucket. This is the rule-ification of the 8/24 lesson about "OLLI's thin-volume print nearly being written up as an earnings move."
7. QFIN's after-hours −13.53%. Results are scheduled for release before the Hong Kong open on 8/26; as of this report's completion there are no verifiable numbers, so take no direction.
8. The second leg in oil. After a one-day −5.47%, Iranian diplomatic progress is an unresolved event and can reverse; most of the short-side odds have already been used up today, while going long means carrying Strait-of-Hormuz tail risk.

⑤ Inputs for tomorrow's pre-market list

1. 🔴 Top priority: restore generation of the pre-market list. Today's missing list left this report unable to do a normal reconciliation, forcing a retreat to a small sample cobbled together from 5 verification points. Two consecutive days of list gaps would make the entire "pre-market → recap" loop meaningless.
2. The "pricing progress" dimension must be implemented; today we already paid for its absence. The 8/24 recap put it on the to-do list, and today's strongest sector in the market (XBI +3.00%, MRNA +14.36%) was precisely an older main line on pricing Day 5 — a window structure that only scans "overnight to pre-market" is structurally blind to it. Concretely: tag every theme with "first disclosure date + which day we are on," and set a separate search item for main lines on Day 3–7 (analyst price-target upgrades, index inclusion, large options prints), because in that phase the trigger is no longer news.
3. Upgrade "was the gain made intraday or in the gap" into a ranking variable for the list, not just after-the-fact commentary. Today's contrast is extremely clean: HOOD +8.17% entirely intraday (fully executable), AA +3.95% entirely intraday, MRNA +10.68% intraday; whereas of MU's +2.48%, +2.04% was the gap, SNDK opened up 2.72% and closed down 3.45%, and AMAT/KLAC/LRCX all three opened high and faded. The 8/24 lesson was "don't chase gaps"; today adds the positive half: prioritize finding flat-open-then-higher structures.
4. Tomorrow's list must actively answer three questions: (a) does the IGV–XLK spread widen or narrow (is software a systemic problem); (b) is the "intensifying industry promotions" DKS named a sporting-goods industry problem or a whole-consumer-discretionary problem — XLY was only −0.30% today, so the market currently treats it as the former, and Thursday's DG/DLTR/BBY/BURL will give the answer; (c) is the crypto high-beta end's strength a one-day rating-driven move or an allocation shift.
5. Introduce a new discipline for the "avoid list": any avoidance justified by "the event is fully priced / the correlation has broken down / no new catalyst" must simultaneously answer "then what does it follow now?" Today's three wrong avoidance calls (energy, crypto high-beta end, optical communications) all share this one missing step, together forfeiting moves of 1.7%–8.2%. If you cannot answer it, do not write "avoid" — write "not tracking." The two mean completely different things to the reader.
6. Holding periods can be extended across 8/26 for the first time this week, but only for two categories: names unrelated to NVDA's earnings (biotech, shorts on the athletic footwear & apparel chain), and names that already finished their gap today. Everything else stays at ≤1 day until both gates — tomorrow's 08:30 and the after-hours print — are cleared.

---

⚠️ Failure log for this report's data collection, verification and runtime environment, for internal assessment of reliability and whether a re-run is needed:

🔴 Severe: all four report slots failed for the entire day on 8/25; the missing pre-market list is this report's biggest structural defect

• reports/run.log shows 8 consecutive scheduled runs on 8/25 all failing for the same reason: Failed to authenticate: OAuth session expired and could not be refreshed.
  • ashare recap: 18:00, 18:20, 18:40 (3 runs)
  • us premarket: 20:00, 21:00, 21:20, 21:40, 22:00, 22:20, 22:40 (7 runs)
  • Each triggered a ⚠️ 研报生成失败 alert email to sk@sfg.xyz / msx@sfg.xyz, meaning the owner group has already received roughly 10 failure alerts today.
• Direct consequence: reports/us/2026-08-25.md does not exist, §2 of this report cannot do a normal list reconciliation, and degrades into reconciling the 5 verification points + 7 avoidance calls from 8/24 recap §6. That is a 5-sample reconciliation with very weak statistical meaning; the body explicitly flags that it cannot be compared horizontally with historical list hit rates.
• A separate configuration error is also printed on every run: Permission allow rule (.claude/settings.json): Write(./reports/**) is not matched by file permission checks — only Edit(path) rules are. Use Edit(./reports/**) instead. This is unrelated to OAuth; it is a malformed permission rule in .claude/settings.json, and Write(./reports/**) should be changed to Edit(./reports/**). Recommend fixing it at the same time, otherwise the write-to-disk step may still trip up once OAuth is restored.
• The 8/25 A-share market recap was likewise not generated (reports/a-share/2026-08-25-recap.md is missing).

Data pipeline

• Local yfinance failed in this run: yfinance.exceptions.YFRateLimitError: Too Many Requests, occurring at the _get_cookie_and_crumb_basic stage — it could not even fetch 5 days of history for SPY. This channel was abandoned outright.
• Single stocks / ETFs / indices / yields / commodities all went through the CNBC quote API (quote.cnbc.com/quote-html-webservice/restQuote, exthrs=1), pulling OHLC + volume + after-hours price + after-hours volume + EventData (next earnings date / halted status) in one shot.
• Symbol freshness was verified line by line: the last_time values for .SPX / .DJI / .IXIC / .RUT / .VIX / US10Y / US2Y / US30Y / .DXY / @CL.1 / @BZ.1 / @GC.1 / BTC.CM= were all 2026-08-25T16:10–17:08-0400, none hitting the _DJI/_COMP stale-data trap. Index closes were pulled twice independently, with matching results.
• ExtendedMktQuote is a nested dict, not a flat field — the first version of the script read it as flat fields, all after-hours prices came back empty, and this nearly led to writing "no after-hours moves today." Fixed. Remember this pit: a field coming back empty ≠ the data not existing.
• HIBB returned all None (delisted; acquired by JD Sports in 2024). It was removed from the athletic footwear & apparel transmission table rather than being treated as "missing data."

Channels that failed in testing

• WSJ market diary (wsj.com/market-data/stocks/marketsdiary) is inaccessible, returning Claude Code is unable to fetch from www.wsj.com. This is the primary source for advance-decline figures, so §1's market breadth had to fall back on the RSP−SPY proxy, as flagged in the body. This gap is now in its second consecutive trading day; recommend finding a stable A/D alternative (the NYSE site or Barchart) and hard-wiring it into the process.
• SEC EDGAR archive page 403: DKS's 8-K EX-99.1 (sec.gov/Archives/edgar/data/0001089063/...) could not be retrieved. All DKS figures in §4.1 come from cross-checked secondary sources (Seeking Alpha / CNBC / Motley Fool / MarketBeat) — multiple sources agree but none is primary, and the body flags "the SEC original was not retrieved." Do not quote them verbatim where legal accuracy is required.
• CNBC article page 403 (cnbc.com/2026/08/25/oil-hits-one-week-low-...), Seeking Alpha article page 403, TheStreet 403.
• Kiplinger's economic calendar page for this week returned truncated content, so a day-by-day schedule could not be obtained. Switched to checking the Bureau of Economic Analysis (BEA) official release calendar directly, confirming that 8/26 08:30 carries both the second estimate of Q2 GDP and July personal income and outlays (including core PCE). This one did clear a primary source.
• Whether July durable goods orders are also released on 8/26 could not be verified (not on the BEA calendar; the Census calendar was not checked). §6② therefore does not list durable goods orders — better to write less.

Four errors that nearly made it into the body (all caught, none used)

1. 🔴 MRNA +14.36% was nearly attributed to a piece of stale 8/19 news. Search summaries described the Moderna/Merck Phase III data as a same-day event, but the BioPharma Dive original is dated 2026-08-19 and states explicitly that "Moderna's stock more than doubled after the announcement." Had it been used, an older main line on pricing Day 5 would have been written up as "a fresh bullish catalyst today," directly inflating its weight in the next day's list. The actual trigger, confirmed on a second search, was Barclays raising the target from $48 to $125 on 8/25 (maintaining Equal Weight) + a bounce after the pullback. This is the same failure class as "AI limit-up attribution stitching in old announcements."
2. 🔴 The oil catalyst was nearly sourced from an article dated 2026-02-02. Search returned a gulfnews piece, "Oil prices plunge after Trump says hopeful over Iran talks," whose headline matched today's move closely; after fetching it, the publication date turned out to be February 2, with WTI at $62.99 / Brent at $67.09 in the text — an order of magnitude away from today's $81.08 / $87.13. The whole item was discarded.
3. 🟡 Several search summaries treated 8/24's closing figures as 8/25's. Instances included "Brent fell 35 cents to $91.82, WTI fell 41 cents to $84.60" (actually 8/24) and "the Dow at 53,540.60, up 123.44" (actually an 8/25 intraday reading, not the close). Every index and commodity figure in this report comes from self-tested CNBC closing readings; no price from any search summary was used. This is the Nth recurrence of "recurring column headlines hide the publication date."
4. 🔴 The auto-summary of Intuit's press-release PDF produced completely wrong numbers — it returned "Q4 revenue $2.96 billion, full-year $11.05 billion, non-GAAP EPS $6.80–7.10," whereas Intuit's actual full-year revenue is $21.4 billion (CNBC's revenuettm also shows 20.925B, contradicting it on the spot). Switched to downloading the PDF and extracting the full text directly with pypdf, obtaining the line-item figures and the two key footnotes ($5.81 / $1.48 SBC impact) from the original. Lesson: small models cannot be trusted to extract from earnings PDFs containing tables; key figures must be pulled down and parsed locally.

Self-check on the basis restatement (the core argument of §4.2)

• FY27: $22.88 + $5.81 = $28.69; $23.12 + $5.81 = $28.93; midpoint $28.81 vs old-basis consensus $27.32 → +5.45%.
• Q1 FY27: $2.44 + $1.48 = $3.92; $2.48 + $1.48 = $3.96; midpoint $3.94 vs consensus $4.02 → −1.99%.
• Both $5.81 and $1.48 are taken verbatim from the footnotes to the guidance table in the company's press release; they are not my own estimates from dividing total SBC by share count. The consensus figures $27.32 / $4.02 come from secondary sources and did not clear a primary source, which is why the body uses "approx." and states the conclusion as "about 5.5% above" rather than a precise value.
• Cross-check: FY26 non-GAAP EPS (old basis) $24.27 → the restated FY27 midpoint of $28.81 = +18.7%, directionally consistent with the company's "+23~24% on the new basis" (a lower base on the new basis yields a higher growth rate), with no internal contradiction.

Unfinished / open items

• NYSE / Nasdaq advance-decline figures not obtained (see above); §1 breadth uses a proxy basis.
• QFIN's after-hours −13.53% cannot be attributed: results are scheduled for release before the Hong Kong open on 8/26, and there were no figures at the time of writing. The body records the price only, gives no attribution, and lists it under §6④ avoid.
• The cross-check between BNS +7.18% and the Canadian-dollar line did not fully reconcile: secondary reporting on the Toronto line gives "+4.56% to C$125.78," which converts back at the day's FX rate to roughly $93.17, matching the New York line's close of $93.10; but converting the New York prior close of $86.86 into Canadian dollars does not match that article's implied C$120.29, off by about 2.6pp. Judged that the secondary figure is an intraday snapshot rather than a close; this report uses the self-tested New York-line close of +7.18% throughout and does not cite the Canadian-dollar figures.
• Whether OKTA reports tomorrow is unresolved: CNBC EventData gives 12/01/2026 (est), conflicting with the 8/24 list. Flagged in the body and handled conservatively (not counted), but OKTA's official IR was not checked to confirm — if it does report tomorrow, that is an omission in this report.

---

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