US Market Recap | 2026-09-04 (ET) Friday

⚠️ The reconciliation baseline of this report is a substitute, not today's pre-market watchlist.

• Today's (09-04) US pre-market watchlist was never generated. bin/generate.sh us premarket failed at launch 7 times in a row at 20:00, 21:00, 21:20, 21:40, 22:00, 22:20 and 22:40 (Beijing time), the cause being You've hit your weekly limit · resets Sep 7, 7pm (Asia/Shanghai), and each attempt triggered a failure alert email (recipients sk@sfg.xyz / msx@sfg.xyz). reports/us/2026-09-04.md does not exist; its size is 0.
• Therefore §2 of this report instead reconciles against the "Key Focus for the Next Session" list in §6③ of the 09-03 recap (6 items, each with a falsifiable verification point). That is a forward-looking conclusion we ourselves published, aimed at 09-04, and it is the strongest reconciliation baseline currently available; but it is not a pre-market watchlist, its coverage is narrow (6 items vs the usual 20+), and its hit rate cannot be compared directly with the historical pre-market reconciliation basis.
• There is also one piece of run-log noise, unrelated to the conclusions here but in need of a fix: Permission allow rule (.claude/settings.json): Write(./reports/**) is not matched by file permission checks — should be changed to Edit(./reports/**).

⚠️ Data-fetch failures and fallback channels

• Local yfinance was again rate-limited by Yahoo (YFRateLimitError, failing already at the _get_crumb_basic stage), confirmed by testing this run; yfinance was not used anywhere in this report.
• Channels actually used: the CNBC quote endpoint (extended-hours fields are readable after the close, consistent with the established conclusion "usable after hours, unusable pre-market"), the US Treasury daily yield curve CSV (used to cross-check Treasuries), and stockanalysis.com (after-hours movers list).
• Channels that failed in testing: stockanalysis.com/api/screener/s/f (404, endpoint has changed; all three parameter combinations failed) and ts-api.cnbc.com/harmony/app/charts (400). Consequently the indices' "week-to-date change" could not be obtained, and this report provides no weekly figures.
• Market breadth is computed locally: taking the 503 S&P 500 constituents (datasets/s-and-p-500-companies) and pulling each one's change via the CNBC endpoint, coverage succeeded for 499, with 4 missing (which does not affect the qualitative advance/decline read).

⚠️ Two falsifications of external information sources, written into the body; the method is recorded here

1. Zacks' Stock Market News for Sep 4, 2026 is headlined September 4, but the body describes the September 3 session (it states the Dow closed at 53,686.11 and the VIX closed at 14.32, down 5.8% — those two numbers are precisely the 09-04 prior closes read locally). Its "NYSE advance/decline ratio 1.98:1" is the breadth of 09-03, and copying it would have reversed today's actual decliner majority. All breadth figures in this report use locally computed values instead.
2. Multiple media outlets said that today "the 2-year yield hit its highest since January 2025." The Treasury's official closing par yields falsify this: the 2Y is 4.37% today, whereas both Monday (09-01) and Wednesday (09-02) of this week were 4.39% — today is not even a weekly high. This item directly changes the macro read (see §5), so it is stated explicitly in the body.

0. One-Sentence Recap

August nonfarm payrolls at +162,000 (consensus +53,000) set off a "good news is bad news" reaction, but the real story today was not rates — it was money moving inside tech, out of software and into AI hardware. The three major indices closed slightly lower (S&P −0.38%, Dow −0.51%, Nasdaq Composite −0.29%), yet the Nasdaq 100 rose against the tide, +0.21%, while inside the S&P 500 it was 177 up / 322 down with a median of −0.53% — the indices' "small decline" masked a session with clearly bearish breadth. The strongest thing on the board was semiconductors (SMH +2.61%) and the weakest was software (IGV −2.23%), a spread of −4.84pp that exactly and completely reversed yesterday's +3.02pp. Of the 6 verification points the previous day's recap set for 09-04, 4 were confirmed, 1 falsified and 1 half right, and among them "if payrolls come in strong, then software / gold miners / real estate are all three fragile legs at once" was hit precisely (IGV −2.23%, GDX −2.20%, XLRE −0.72%). But that same list completely failed to foresee that semiconductors would lead the whole market on a day of rising rates — this is today's biggest cognitive gap. The VIX was only 14.53 (+1.47%), the 10Y 4.784% and the 2Y 4.368%, with the front end up 3bp and the 30-year actually down 1bp — the shape of a rate-hike repricing, but in magnitude it did not recover the dovish move caused by Waller's remarks on Wednesday (see §5, the most counterintuitive item of the week). The next trading day is 09-08 (Tuesday) — Monday 09-07 is closed for Labor Day.

---

1. Market Overview

Indices and Turnover

Basis note: The indices above are CNBC official closing reads (read at 16:33–17:01 ET). The Russell 2000 is represented by the IWM ETF (no read of the Russell index itself was obtained locally). The indices' "week-to-date" changes were not obtained this run, so this report provides no weekly figures.

There is a structure here that must be spelled out: the Dow fell 0.51%, the Nasdaq 100 rose 0.21%, and the Russell 2000 also rose 0.28%. What rose was "AI hardware" and "small caps"; what fell was "large-cap value + software + defensives." This is not a session that can be summarized with a single "risk-on / risk-off" label.

Market Breadth (computed locally, S&P 500)

Breadth is more bearish than the index. The median of −0.53% is weaker than the cap-weighted −0.38%, showing that the index was held up by a handful of heavyweight AI hardware names. On an equal-weighted basis, today was unambiguously a down day.

Rates, FX, Volatility, Commodities

Treasury data was taken from both CNBC and the US Treasury's daily par yield curve CSV, and the two agree. The Treasury's official basis is used (bp is authoritative). Note: CNBC's 2Y change_pct field contradicts its own last/prior close (it shows −0.0664% while the change amount is +0.034); that field is confirmed unusable this run, and this table states everything in bp.

Sentiment read: the rate-hike repricing holds, but panic did not show up. The curve shape is textbook hike pricing — front end +3bp, long end −1bp, 2s10s narrowing 2bp — the dollar strengthened 0.25% and gold fell 1.41%; four indicators corroborate one another with not a single conflict. But the VIX is only 14.53, and on a day when "the September hike probability jumped from 49% to 60%" that level is near this year's low range. The market is repricing the rate path but refuses to pay an insurance premium for it.

---

2. Reconciliation: the 6 Verification Points the 09-03 Recap Set for Today

⚠️ Baseline note (must read): today's US pre-market watchlist failed to generate, so this section is not the usual pre-market watchlist reconciliation.

The substitute baseline is the 6 forward-looking conclusions in §6③ "Watch Closely (Ticker + Falsifiable Verification Point)" of the 09-03 market recap. These are falsifiable judgments we published ourselves, explicitly aimed at 09-04, so they can be reconciled strictly; but there are only 6 of them, far narrower than a normal pre-market watchlist (usually 20+ names), and the "hit rate" below cannot be compared laterally with the historical pre-market reconciliation basis.

Hit Rate and Self-Critique

4 confirmed, 1 half right, 1 falsified — roughly 4.5 / 6 ≈ 75%.

Three macro-level judgments are settled separately; they matter more than the single-stock list:

One line of self-critique (the single most important thing to write down in this report):

We bet the whole table on one variable — rates — and the rates line was entirely right, yet we missed the day's real protagonist.

§6 of the 09-03 recap was built end to end around "payrolls → rates → duration assets," and five of the six verification points (GWRE, LULU, IGV/SMH, NTAP/HPE, XLRE/KRE) were directly or indirectly expressions of rate sensitivity. That framework was proven right today — every one of the three fragile directions named duly fell.

But semiconductors rose 2.61% on that same day, with SNDK +11.90%, KLAC +7.32% and MRVL +7.05%, and not a single item in the entire list mentioned semiconductors. Yesterday even stated explicitly that "SMH +0.39% shows the spillover did not happen; this theme can only be used for single-stock screening" — semiconductors were actively removed from the watch list.

The lesson is not "the rates framework was wrong," but that "a correct framework makes you stop looking for a second driver." When all six verification points come from the same variable, they are not six independent judgments but six copies of one judgment — the high hit rate on the rates line is exactly what masked the fact that the whole list had only one dimension. §3 and §6 accordingly add a mandatory "non-rate driver" scan item.

---

3. Today's Theme Verification

The Core Disagreement: today was not a rates day, it was a "reallocation inside tech" day

Falsifying evidence first, attribution second:

The 10-year Treasury moved only +1bp today (4.77% → 4.78%), while IGV fell 2.23%.

A 1bp move at the long end cannot explain a −2.23% decline in software. Yesterday attributed software's +3.41% to "buying the discount rate"; if rates are still used today to explain software's −2.23%, one must accept an absurd elasticity (2.23% ÷ 1bp). The rates attribution fails on the spot today.

Stronger counter-evidence: within the same high-valuation tech complex, semiconductors were +2.61% and software −2.23%, and both face the same yield curve. If discount rates were the driver, the two should move together; in reality they differ by 4.84pp. The driving variable is not rates, it is money relocating inside tech.

Theme Scorecard

Sector ETF Panorama (only 3 of the 11 SPDR sectors rose)

XLK's +0.70% contains both SMH's +2.61% and IGV's −2.23%. Looking only at XLK yields the conclusion "tech up a little," and completely hides the 4.84pp tear inside the sector. This is a session that can only be read correctly by drilling down to the sub-industry level.

Dedicated Section One: what exactly rose in semiconductors? — not a broad rally

Leading must be distinguished from broad-based. Breaking today's semiconductors down by sub-segment:

This was not a broad semiconductor rally but a highly directional "memory + equipment" move. Analog/MCU underperformed the sector across the board by 1–2.5pp, and even NVDA (+0.84%) and AVGO (+0.21%) lagged far behind. Any conclusion that reads today as "the AI main line is fully back" is falsified by these two data sets — the analog names and NVDA.

On the catalyst — honesty is required: no new company-level catalyst could be confirmed today.

• Two independent 24/7 Wall St. reports (one on semiconductors, one on memory) both explicitly stated there was no company-specific news that day, with SanDisk's noted as "no company announcement this morning; the intraday move is sector fund flow rather than a new headline."
• The "Susquehanna expects DRAM contract prices to rise more than 50% and NAND about 60% this quarter" that surfaced in searches was not published today; it comes from an existing view cited by a 09-04 commentary piece. Treating it as today's catalyst would commit the error of "taking an old disclosure for today's good news," and this report does not adopt it.
• The mechanism hypothesis put forward by the media is: memory and equipment names have finished reporting, their near-term cash-flow certainty is high, and when the discount rate turns unfavorable they instead attract money looking for a place to park. This is only a hypothesis; this report does not treat it as a conclusion, and it has been converted into a falsifiable verification point in §6.

⚠️ At the same time, a reminder of an established directional trap: the strength in semiconductor equipment (WFE) speaks to "future supply," not "current demand." The gains in KLAC/LRCX/AMAT/ASML reflect fab capex expectations; for memory-cycle strength it is a lagging confirmation, not a leading one. Using equipment strength to infer that memory price increases can persist gets the direction backwards.

Dedicated Section Two: software's problem may not be rates at all

Listing all 20 software names one by one, breadth is 5 up / 15 down (yesterday IGV was 13/13, all up):

Three pieces of evidence point to a "structural de-rating" rather than "rates":

1. The 10Y moved only 1bp (as noted); the rate elasticity cannot hold.
2. The only big gainer, ORCL (+3.08%), is precisely the software name most treated as "AI infrastructure." What rose was not the software attribute but the hardware attribute.
3. Background data: the S&P 500 information technology index is −3.5% year to date, and within it the software & services sub-sector has collapsed 23% year to date; INTU's and CRM's valuations have been compressed by roughly 46% and 30% year to date. The market's pricing of "generative AI will eat software business models" is a trend line running through the whole year, and yesterday's +3.41% is the anomaly.

Conclusion: yesterday's recap attributed software's +3.41% to "buying the discount rate," and today that needs correcting — it was more likely a short squeeze, and the sign of the IGV-vs-SMH spread (the verification point we set ourselves) has already given the answer. This is consistent with established experience: the software/hardware fork is the fastest yardstick for falsifying a rates attribution.

Dedicated Section Three: two event drivers missed entirely today

① FICO −16.68%: regulation punches straight through the business model (a genuinely new catalyst that day)

• Event: FHFA Director Bill Pulte instructed Fannie Mae and Freddie Mac to immediately approve the use of VantageScore by all lenders, ending FICO's long-standing monopoly in mortgage credit scoring.
• This is a new event, not old news reissued. Pulte already pushed one round in April 2026 (FICO fell then too), but that round was a pilot capped at 50 institutions; today's action removes the cap and opens it up fully — the nature is upgraded from "pilot" to "monopoly removal."
• Pricing rationale: Pulte said FICO has raised the cost of a single-person credit score by 1,800% since 2020; VantageScore 4.0 is quoted as low as $1 per pull, roughly 90% cheaper than FICO's approximately $10 price.
• An intraday divergence that must be pointed out (execution information):

All three were smashed indiscriminately by 10–16% at the open, but parted ways by the close: FICO did not gain a cent, while the two credit bureaus each recovered 4.5–5.4pp.

This divergence makes sense, and the direction matters: VantageScore is itself a joint venture of the three credit bureaus Equifax, Experian and TransUnion. A rising VantageScore share is not purely negative for EFX/TRU — it may even be positive. That move at the open treated all three as "the credit bureau sector" and sold them indiscriminately; intraday money corrected the error. FICO is the only net loser, and the market priced it without any hesitation.

② The design / engineering software cluster was hit hard collectively, and the catalyst could not be confirmed

ADSK −8.26%, PTC −6.04%, SNPS −5.40%, CDNS −3.99%, plus ADBE −6.73%, form a clearly bounded down-cluster.

• ADBE has a clear catalyst: days before its earnings the board announced that insider Anil Chakravarthy would become CEO, which the market read as "choosing continuity over change"; against a backdrop of OpenAI (DALL·E / Sora), Midjourney, Canva, Figma and Alphabet (Imagen / Veo / Gemini) continuously eroding its core creative business, doubts about the AI strategy outweighed everything else. ADBE is already −37% year to date, at roughly 13 times earnings.
• ADSK's catalyst cannot be confirmed by this report. The explanation returned by searches is "sell the news after the August 27 earnings" — but an August 27 print does not explain a −8.26% on September 4, a week apart. ADSK opened at $232.08 (only −2.3%) and closed at $217.90, at the low of the day, grinding down all session rather than gapping, a shape that looks more like incremental information or a large reallocation appearing intraday than overnight news.
• This report states plainly "cannot be confirmed for now" and fabricates no attribution. It must be emphasized: "unable to find news" does not equal "there is no news" — SNPS/CDNS/PTC/ADSK fell sharply in the same direction on the same day, and the fact that all four fell together is itself evidence that "some common driver exists"; we simply could not locate it locally within the hour after the close. This is logged as a mandatory check item for the 09-08 pre-market.

---

4. After-Hours Earnings Moves

① After today's close (after the 09-04 close): no significant earnings

Friday after the close is an earnings vacuum; there are no company results tonight worth treating as a catalyst for the next session. Checked against the stockanalysis after-hours movers list, the top 5 gainers (IPEX +26.97%, BRNX +21.88%, ATER +12.24%, ALTI +12.00%, ISPC +10.71%) and top 5 decliners (FCUV −44.18%, BTAI −20.16%, NCT −14.05%, ONCO −13.65%, JAGX −10.81%) are all micro-cap companies, and not one can be confirmed as earnings-driven.

After-hours prices for the major names essentially hug their closes, with no unusual moves: NVDA −0.13%, AVGO −0.08%, SMH −0.18%, IGV +0.03%, GWRE −0.25%, LULU −0.28%.

② During today's session: the final pricing of last night's earnings (this is today's real earnings story)

This fulfills a methodological commitment made in the 09-03 recap. Yesterday explicitly recorded that those three after-hours reads were taken at 17:01–17:03 ET, i.e. right at the start of the calls, and required that "they must be re-read the next day with the difference between the two moments noted." Here it is:

Fundamentals review (data taken from the 09-03 recap, all on verified bases):

• GWRE (Guidewire): FY26Q4 EPS $0.99 vs consensus $0.93; revenue $411.1 million, +15% vs consensus $402.7 million; ARR $1.237 billion, +19% (constant currency). The FY27 full-year revenue guidance midpoint is slightly above consensus, but Q1 FY27 revenue guidance of $372–378 million vs consensus $387.1 million is about 3% low at the midpoint.
• LULU (Lululemon): adjusted EPS $2.92 vs consensus $1.82, but $0.86 of that came from a tariff refund and interest; restated it is $2.06 (still a beat); revenue $2.40 billion, −4%, missing consensus, with comparables −9%; full-year revenue guidance midpoint $10.425 billion vs consensus $11.03 billion (a 5.3% cut).
• DOCU (DocuSign): FY27Q2 EPS $1.16 vs consensus $0.96 (a 20.8% beat); revenue $875.7 million, +9%; revenue, ARR and IAM share guidance were all raised.

The three samples give the same rule, confirmed today by closing prices: the fork is not the current quarter's results (both GWRE and DOCU beat), it is next quarter's guidance. GWRE's next-quarter guidance 3% low → −19.93%; LULU's full-year guidance 5.3% low → −17.38%; DOCU raising all three guidance lines → +3.70%.

One new lesson learned only today: the deviation between "the after-hours read" and "after the call" is asymmetric. GWRE, with negative guidance, kept worsening by 5.2pp after the call, while the positive DOCU gave back 1.6pp. Both directions moved toward the worse side — the first-instant after-hours read underestimates negatives and overestimates positives.

---

5. Flows and Sentiment

Sector Rotation: a clear money path

Inflows: memory / semiconductor equipment / AI interconnect / AI power (SNDK +11.90, KLAC +7.32, MRVL +7.05, NRG +6.42, CEG +4.88) Outflows: software / design software / credit bureaus / gold miners / consumer discretionary (IGV −2.23, GDX −2.20, XLY −1.33, FICO −16.68)

This is a "software to hardware" reallocation inside tech, overlaid with a rate-driven reduction in duration assets. The two lines are independent of each other:

• The rates line (moderate intensity): XLRE −0.72%, XLF −0.79%, GDX −2.20%, DXY +0.25%, gold −1.41%. All directions are consistent, but none of the magnitudes are large.
• The AI hardware line (extremely high intensity): SMH +2.61%, SOXX +3.52%, single stocks 5–12%. This line has nothing to do with rates, the evidence being that it led the market higher against the tide on a rising-rate day.

The Most Counterintuitive Item of the Week: payrolls could not push rates back to where they were

Media today widely reported that "the 2-year yield touched its highest since January 2025." This claim does not match the Treasury's official closing par yields, and this report corrects it:

Today's 2Y at 4.37% is not even a weekly high — both Monday and Wednesday of this week were 4.39%.

Doing the arithmetic: Waller's remarks knocked the 2Y from 4.39% down to 4.34% (−5bp); a payrolls print three times consensus pushed it back only to 4.37% (+3bp), recovering about 60%.

The hike probability gives an entirely consistent signal: roughly 66% on 08-31 → 49.4% on 09-03 (post-Waller) → 60.2% on 09-04 (post-payrolls). It likewise did not return to the level before Waller's remarks.

Two independent indicators cross-validate the same conclusion: even handed a jobs print that beat by a wide margin, the market is unwilling to revise its rate expectations back to where they were a week ago. That means the weight assigned to Waller's stance is greater than that of a single payrolls report, and it also means the real adjudication point is still the 09-11 CPI, not today's payrolls.

⚠️ Basis note: the 49.4% → 60.2% hike probabilities above are taken from the same source (24/7 Wall St. citing CME FedWatch) to keep the yardstick comparable; the 66% for 08-31 comes from Forbes citing CME FedWatch and was read at a different moment, so it serves only as a trend reference. The 09-03 recap recorded 54.6%, which is a read at a different point in time from this report's 49.4%; do not mix them when comparing across sources.

August Payrolls Details

This print is strong on three dimensions: the absolute number is three times consensus, the prior two months were revised up in aggregate (July flipping from negative to positive), and it is far above the +31,000 monthly average of the past 12 months. By industry, bars and restaurants led, while information fell — possibly related to AI investment substitution, a detail that echoes today's tape of "software down, AI hardware up."

Risk-on / Risk-off Characterization

Conclusion: neither; this was a "theme reallocation day."

Basis for the call:

The decisive evidence is the last row: if today were risk-off, money should have flowed into utilities and staples — but XLU was only +0.12% and XLP −0.80%; the defensive sectors received no money. Meanwhile high-beta semiconductors rose 2.61%. This is not reducing risk, it is moving risk from one theme to another.

---

6. Next-Session Outlook

🚨 Time-base reminder: the next trading day is 2026-09-08 (Tuesday). 2026-09-07 (Monday) is Labor Day; US equity and bond markets are closed all day, with no early close scheduled beforehand. All events in this section are labeled with absolute dates, and "next session" always means 09-08 (Tuesday); with 3 days of closure in between, the time risk of holding through the holiday is significantly higher than a normal weekend.

① Theme Continuity

② Earnings and Macro Calendar (all with absolute dates)

⚠️ One date correction: the 09-03 recap labeled the August CPI as "09-11, next Thursday," which is wrong — 2026-09-11 is a Friday. Some financial media likewise carried the erroneous "Thursday, Sept. 11" wording. Correct wording: September 11 (Friday) 08:30 ET. Because of the Labor Day shift, that week has only 4 trading days (09-08 through 09-11).

📌 The 09-10 after-close is the most concentrated single-point risk of this round: ORCL and ADBE report the same evening. One is today's only winner in software (the AI infrastructure narrative), the other today's disaster stock (the AI disruption narrative). These two prints will, on the same night, give a directional answer to "is AI software's enemy or its friend," and their pricing impact on the sector may well exceed CPI's impact on software.

③ Watch Closely (Ticker + Falsifiable Verification Point)

This edition adds a mandatory "non-rate driver" item (see the self-critique in §2). Of the 6 items below, only 2 come from the rates line; the other 4 come respectively from thematic fund flow, event drivers and the earnings pricing rule — to avoid another case of "six verification points that are really six copies of one judgment."

④ What to Avoid

1. Avoid treating today's semiconductor gains as the start of a trend. With no catalyst confirmable, a single-day gain of 5–12% can only be treated as a flow event. Two independent sources both said there was no company news that day; treating "sector fund flow" as a fundamental inflection is the same error as yesterday treating "a short squeeze" as "the end of software's washout" — and that error happened just yesterday and has already been falsified today by our own verification point.
2. Avoid using today's analog/MCU performance to characterize the semiconductor sector. QCOM +0.10%, NXPI +1.26% and TXN +1.82% all underperformed the sector by 1–2.5pp, and NVDA (+0.84%) and AVGO (+0.21%) also lagged far behind. What rose today was "memory + equipment," not "semiconductors," and certainly not "the AI bellwethers."
3. Avoid buying ORCL or ADBE on a "beat for the quarter" after the close on 09-10. This week's three independent samples (GWRE −19.93%, LULU −17.38%, DOCU +3.70%) have nailed the rule down: the fork is next quarter's guidance, not the current quarter's results. And two kinds of accounting noise must be restated: a tariff refund booked into the adjusted basis (LULU is the fifth such case) and a change in the non-GAAP basis.
4. Avoid making overnight decisions on the first-instant after-hours read following an earnings release. New evidence today: GWRE worsened from −14.72% (start of the call) to −19.93% (close), a deviation of 5.2pp; DOCU decayed from +5.28% to +3.70%. Both directions moved toward the worse side. The 09-10 ORCL/ADBE after-hours reads must be labeled as read before or after the call.
5. Avoid treating XLRE and KRE as the same leg. Falsified today: XLRE −0.72% (hurt on duration), KRE +0.53% (helped on spread) — opposite directions on a rising-rate day. Yesterday's pairing of the two as "pure rate beta" was wrong.
6. Avoid positions based on "a big payrolls beat = a big move up in rates." The fact is the opposite: the 2Y is 4.37% today, below Monday's and Wednesday's 4.39%; the hike probability is 60.2%, below roughly 66% on 08-31. A jobs print three times consensus recovered only about 60% of the impact of Waller's remarks. The real adjudication is the 09-11 CPI.
7. Avoid carrying heavy positions through this 3-day long weekend. 09-07 is closed, and the first day back on 09-08 faces 3 days of accumulated information, so the probability of an opening gap is higher than usual; and the memory/equipment sector with the largest gains this week is precisely the one whose driver is least clear.

⑤ Input Notes for the 09-08 Pre-Market Watchlist

1. [Highest priority] Fix the generation failure. The 09-04 US pre-market watchlist failed 7 times in a row because the weekly quota was exhausted and was never produced, which cost this report its standard reconciliation baseline. The quota resets at 09-07 19:00 (Asia/Shanghai), only about 1 hour before the 09-08 US pre-market slot (20:00 Beijing time) — with just 1 hour of margin, the pipeline must be verified as working on the evening of 09-07; we cannot wait until 20:00 on 09-08 to discover a problem.
2. The attribution for ADSK/PTC/SNPS/CDNS must be completed. Four design/engineering software names fell 3.99%–8.26% on the same day, and no common driver could be located locally. This is the biggest unresolved item in this report. Note: "unable to find news" does not equal "there is no news" — four names falling together in the same direction is itself evidence that "a common driver exists."
3. The pre-market scan must explicitly add a "non-rate driver" item. The self-critique in §2 of this report shows that when all of a list's verification points come from the same macro variable, a high hit rate masks the fact that "the whole list has only one dimension." The 09-08 list must include at least one theme line unrelated to rates.
4. Semiconductors must be brought back into the routine scan. The 09-03 recap actively removed them from the watch list ("SMH +0.39% shows the spillover did not happen"), and the very next day they led the entire market at +2.61%. A single day of sector lag is not a reason for removal.
5. Model the 09-10 double earnings night (ORCL + ADBE) in advance. The 09-08 and 09-09 pre-market watchlists should give both companies' next-quarter guidance consensus values in advance rather than hunting for them on the night of the print. Note that consensus may be revised up ahead of earnings, so the read date must be labeled.
6. The credit bureau line (FICO/EFX/TRU) needs one piece of fundamental homework: confirm the equity relationship and revenue-sharing structure between VantageScore and the three credit bureaus, in order to judge whether today's intraday recovery in EFX/TRU is an "oversold bounce" or a "business model re-rating." That determines whether it is a trade or a theme.
7. Any call on "the strongest theme" must come with a sub-segment breakdown. Reporting only "SMH +2.61%" today would hide the fact that analog/MCU did not rise at all and that NVDA/AVGO underperformed badly. A sector-level gain must be drilled down to the sub-segment level before it can serve as a conclusion.

---

Appendix: Today's Key Data Quick Reference

Top gainers (S&P 500): SNDK +11.90%, KLAC +7.32%, MRVL +7.05%, COHR +6.60%, NRG +6.42%, STX +6.34%, MU +6.10%, WDC +5.86%, GLW +5.68%, TER +5.49%, LRCX +5.12%, CEG +4.88%, AMD +4.69%, SMCI +4.54%, INTC +4.51% (non-S&P 500 constituent: ALAB +9.75%)

Top decliners (S&P 500): LULU −17.38%, FICO −16.68%, ADSK −8.26%, ADBE −6.73%, EFX −6.37%, PTC −6.04%, TSLA −5.92%, SNPS −5.40%, WDAY −5.38%, NFLX −5.35%, DASH −4.63%, IT −4.62%, PLTR −4.49%, HPE −4.48%, ALB −4.45% (non-S&P 500 constituent: GWRE −19.93%)

Data basis: Single-stock and ETF changes are taken from the CNBC quote endpoint on a 16:00 ET regular-session close basis (read between 16:00 and 16:15 ET); after-hours prices are labeled separately. Treasury yields are taken from the US Treasury's daily par yield curve (cross-checked as consistent with CNBC). Market breadth is computed locally for each S&P 500 constituent, covering 499/503. The indices' week-to-date changes were not obtained this run and are therefore not provided.

---

Sources: CNBC · August nonfarm payrolls +162,000¹ · BLS · Employment Situation 2026 M08² · UPI · Payrolls beat expectations³ · US Treasury · Daily yield curve⁴ · 24/7 Wall St. · Chip stocks shrug off rising rate-hike odds⁵ · 24/7 Wall St. · SanDisk soars and the NAND cycle⁶ · 24/7 Wall St. · Adobe's internal CEO pick⁷ · Yahoo Finance · FICO falls as Pulte ends the mortgage scoring monopoly⁸ · Seeking Alpha · Fair Isaac slumps on the VantageScore directive⁹ · FHFA · Credit Scores policy page¹⁰ · Business Wire · Adobe to report FY26Q3 on 09-10¹¹ · Nasdaq · Oracle earnings calendar¹² · Morningstar · August CPI forecasts¹³ · Forbes · CME FedWatch hike probability (08-31 read)¹⁴ · Yahoo Finance · Labor Day market closure schedule¹⁵ · NYSE · 2026–2028 holiday calendar¹⁶

---

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