US Market Recap | 2026-09-02 (ET) Wednesday

Coverage window: 2026-09-02 09:30–16:00 ET regular session + 16:00–18:05 ET after-hours. Read-timestamp note: regular-session prices are on the 16:00 ET closing basis; after-hours prices are uniformly read at 18:05 ET and carry an after-hours volume tag — the credibility of an after-hours price depends on volume, and this report draws no conclusions from thin-volume reads. Price sources: single stocks/ETFs from the stockanalysis.com quote endpoint (including after-hours fields), cross-checked name by name against CNBC extended-hours quotes; indices from CNBC; Treasuries from the US Treasury's official Daily Treasury Yield Curve CSV (the 09-02 same-day data has been published); earnings figures per the original SEC 8-K EX-99.1 text.

---

0. One-Line Recap

1. Today was risk-on, with one precise exception: software. S&P +0.46%, Dow +0.56%, Nasdaq +0.45%, Russell 2000 +1.13% (IWM +1.18%); the three major indices ended a three-day losing streak, roughly 2/3 of single stocks rose, and VIX fell to 15.20 (−6.98%). Meanwhile the software ETF IGV −2.60% was the weakest in the market, while semis SMH +0.96% — a 3.56 percentage-point divergence between the two.
2. The pre-market list got the directional calls almost entirely right, but its "executable value" was far smaller than the hit rate suggests. The avoid group was 10/10 down on the day (mean −7.62%), the bullish group 9/10 up (mean +4.92%). But executed at the 10:13 ET prices at which the list was actually written, only 4/10 of the bullish group closed higher; excluding DELL, which this list downgraded to "watch only," the group averaged −0.72%, with 6 of 9 names in the red.
3. The single biggest misjudgment today was DELL. The pre-market list read it at 10:13 at +2.65%, already having given back 5.58% from its gap-up high, and on that basis called it "watch only (do not chase)." DELL closed the day +15.81%, with an intraday high of $497.99, while that 10:13 read ($436.26) was very nearly the day's lowest zone (intraday low $432.27). Buying at 10:13 and holding to the close would have returned +12.82%.
4. The AVGO verdict: AI hardware demand has not peaked — it is accelerating — but total revenue guidance came in slightly below consensus, and the after-hours path was "sold first, then recovered." Q3 AI semiconductor revenue $16.7 billion (+221% YoY, +54% QoQ), Q4 AI guidance $21.7 billion (+236% YoY), with growth rates stepping up quarter by quarter. After the 8-K was released, media reported the stock down 5%–6% after hours; yet at 18:03 ET the measured after-hours prints were $367.15/$367.33, essentially flat versus the $367.24 close (−0.02%/+0.02%), on after-hours volume of 13.14 million shares (equivalent to 35% of regular-session volume).
5. All four companies reporting tonight beat, yet the reactions were +22% / flat / −4.6% / −8.7%. SNOW beat thickly (EPS beat by 37.8%) and raised guidance → +21.8%; AVGO → flat; HPE beat and simultaneously raised revenue, EPS and free cash flow for both FY26 and FY27 → −4.6%; NTAP beat every one of its own guidance items and raised the full year → −8.7%. "Beat + raise" added two more failure samples tonight.

The single most important structural fact of the day: XLK closed −0.02%, essentially dead flat — it netted IGV (−2.60%) against SMH (+0.96%). Using XLK to read today's tech sector produces the false conclusion that "nothing happened."

---

1. Market Overview

Market breadth: for more than half of the session, roughly 2/3 of US single stocks were up and 33% down (source: CNBC same-day intraday broadcast). ⚠️ That read is on an intraday, not a closing, basis; this report did not obtain precise closing advance/decline counts and makes no further quantitative inference from it.

Treasuries (official Treasury CSV, 09-02 published)

⚠️ The flat close masks violent intraday swings: the 10-year touched 4.818% intraday, the highest since November 2023, before falling back to close flat at 4.79%. Today's equity rally lines up, in time, precisely with the retreat in yields from that high. TLT +0.10% and HYG +0.01% are consistent with "flat at the close."

Sentiment read: risk-on — and specifically the "rate pressure pauses + small caps/cyclicals lead" flavor of risk-on, not the "mega-cap tech leads" flavor. The ranking was IWM (+1.18%) > DIA (+0.54%) > SPY (+0.44%) > QQQ (+0.23%), consistent with and reinforcing yesterday's (09-01) direction. VIX broke below 15.5; credit (HYG) shows no stress.

---

2. Pre-Market List Reconciliation

How to read: "vs 10:13" = the close relative to the moment the pre-market list was actually written (10:13 ET). That column is what a reader executing the list could actually have obtained; "Close change%" is relative to the 09-01 close, a large part of which had already happened before the list was written.

2A Bullish Side (list §3A)

Bullish group: 9/10 closed up, mean +4.92%; but only 4/10 closed above the 10:13 price, mean +0.64%. Excluding DELL it is −0.72%, with 6 of 9 names below their 10:13 price.

2B Bearish/Avoid Side (list §3B)

Avoid group: 10/10 closed down, mean −7.62%. But from 10:13 to the close it fell only another 0.37%, and 4/10 actually rallied.

Hit Rate and Self-Critique

Self-critique in one line (three items, by severity):

1. The most serious is DELL, and the error mode was systematic. The pre-market list took a single intraday fragment — "gave back 5.58% after the open" — and from it derived "the opening price is the day's high," a conclusion about the whole day, at a point just 43 minutes after the open. 43 minutes of drift is not enough to define a 6.5-hour path. More worth recording: in §8 the list elevated this into a universal rule ("do not chase gaps on earnings names today, in either direction"), and that rule failed the very same day on its own listed sample — GTLB/CRDO/PANW fit it, DELL ran the other way, and DELL happened to be the largest-magnitude name. Inducing a same-day rule from a same-day sample that has not finished playing out is the methodological problem most in need of fixing this time.
2. The avoid group's 10/10 is real, but its value was overstated. These declines had largely already happened before the list was written: after 10:13 the avoid group fell only another 0.37% on average, and 4/10 rallied. The list going out 43 minutes late ate up precisely its most valuable portion.
3. The OKTA ranking error is informative. The list used "52-week percentile 92%, most expensive in the group" as the primary basis for its bearish ranking, placing it at the front of the software avoid group; in fact it was the smallest decliner in the avoid group (−1.97%), while CRDO, at only the 54th percentile before earnings, fell 20%. (Both percentiles are taken on the pre-earnings basis used by the list; after this drop CRDO's current percentile has fallen to 35%.) This is consistent with the self-reminder the list itself wrote in §9③ ("the simplification that only high-percentile names get killed is wrong") — but the reminder was written into the conclusion while the ranking was still done by percentile. The criterion was written correctly and then not used in the ranking.

---

3. Same-Day Theme Verification

① Did the pre-market call the strongest theme? — It called the bearish main line, and missed the bullish one

What it got right: "post-earnings valuation flush" really was the strongest single theme today; IGV −2.60% was the weakest sector ETF in the market, and PANW/MDB/CRDO, all named by the list, ranked among the biggest decliners. This item is hard.

What it missed: the list offered no long-side lead whatsoever corresponding to today's strongest sectors. Today's closing sector-ETF ranking:

The pre-market list got the logic right on the "weakening labor → front-end rates → small caps benefit" chain (listed as theme 6), but landed it only on IWM and gold, never on banks. And KRE +2.23% was the strongest today, with XLF +0.80% also beating the S&P. Weak jobs → rate pressure eases → regional banks is the most direct transmission link in that chain, and the list missed it entirely.

② Unexpected theme one: steel and precious metals moved in opposite directions on the same day

This was the most informative internal divergence of the day, and it is adjudicated entirely by this report's own data:

The pre-market list put all seven under the same "metals" theme, explained by the same macro driver (weak ADP → weaker dollar → repricing of non-yielding assets). This afternoon they split into two groups moving in opposite directions. The dollar fell only 0.11% on the day (DXY) — a magnitude that explains neither group, let alone the divergence between them.

So the "metals" classification was falsified today: precious metals/copper were trading macro (and faded as yields retreated from their intraday high), while steel was trading something else.

On steel's same-day catalyst — this report still could not confirm one, and refuses to stitch one together. Searches returned only out-of-window content such as the 8/24 breakdown in US–Canada trade talks and the 8/30 CNBC tariff analysis, consistent with the pre-market list's search results yesterday. These are late-August old news and cannot be recorded as 09-02 catalysts.

But this report offers one piece of evidence the pre-market list did not have: the same-day divergence between steel and precious metals is itself a rejection of the "macro driver" explanation. The pre-market list downgraded CLF to "watch only" because it "could not find a same-day catalyst"; the correct handling today would have been: no catalyst found ≠ no driver, and cross-sectional divergence can prove a driver exists and is not macro. This item is handled per the memory entry "no news found ≠ no news" — the rationale for the downgrade (no catalyst) is not sufficient to support the action of the downgrade (do not participate), because the price action is itself evidence.

⚠️ At the same time this must be stated explicitly: the above reasoning can only prove that "a driver exists and is not macro"; it cannot prove that driver is tariff/trade policy. The latter remains unconfirmed and is not stated as a conclusion.

③ Unexpected theme two: oil reversed intraday, and airlines did not follow

The pre-market list's 4th theme was "oil pullback → improving airline costs," evidenced by WTI −0.72, USO −1.67%, BNO −1.30%, XLE −0.34% at 10:13, and described as "four independent reads jointly confirming the oil pullback."

By the close, all four reads had reversed:

Yet airlines/cruises gave nothing back: UAL closed +3.56% (+3.79% at 10:13), LUV +2.47% (+2.44%), CCL +2.20% (+2.30%).

Conclusion: the causal chain the pre-market list gave for airlines was falsified by its own data that same afternoon, but the conclusion (airlines strengthen) still holds. The more likely explanation is that airlines/cruises were bought today as "high-beta small-cap cyclicals" (in sync with IWM +1.18%, KRE +2.23%, XLB +1.69%), not as "fuel cost improvement."

The lesson here is actionable: the verification point the pre-market list set for UAL was "whether Brent stays below $95; if it moves back above $95 this line is immediately invalidated." Oil did rebound today, so by that verification point UAL should have been judged invalid and exited — and UAL closed +3.56%. The verification point was set on the wrong instrument. Same family as the memory entry "verification point beats headline conclusion," but this time it is the negative case: the verification point was set on a variable that is not in a genuine causal relationship with the conclusion, so it produced a false sell signal.

④ CRDO idiosyncrasy: the pre-market's core call was confirmed tonight

Pre-market list §5② made a high-risk call: CRDO −17.57% is an idiosyncratic single-stock event and should not be taken as evidence that AI networking demand has peaked, and set a falsification condition — "if AVGO's networking/custom-ASIC guidance disappoints, this conclusion must be overturned."

AVGO's answer tonight is decisive: AI semiconductor revenue Q3 +221% YoY / +54% QoQ, Q4 guidance +236% YoY — the growth rate is stepping up. Demand-side falsification is complete; the pre-market call stands.

⚠️ But the other half of the cross-check has not yet delivered, and must be recorded honestly: the list required that "ALAB/MRVL recover tomorrow the excess decline they took relative to SMH today." Today they not only failed to recover it, they underperformed again:

So the conclusion "CRDO is idiosyncratic" currently sits in a state of "fundamentally confirmed, price not yet confirmed." Real confirmation requires watching whether these three close the gap versus SMH on 09-03 after AVGO's print — this is tomorrow's first-priority observation item.

---

4. After-Hours Earnings Moves (next-day catalysts)

⚠️ Every after-hours move in this section carries an after-hours volume tag. No conclusions are drawn from thin-volume reads — this is a hard rule of this report (the pre-market list already paid for this once on PANW yesterday: a thin-volume after-hours read of −0.02%, followed by an actual −8.24% the next day).

① AVGO: the total-revenue guidance "miss" comes entirely from the non-AI business

This is the number most easily misread tonight, and it must be broken apart. Media headlines read "Broadcom forecasts quarterly revenue below estimates" (Reuters) and "weak guidance overshadows earnings beat" (CNBC). But once Q4 guidance is broken out by business, the direction is the opposite:

That is: the $230 million total-revenue shortfall is the net of a $370 million AI beat against a $600 million non-AI shortfall. The line the market is buying is beating; the line it is not buying is missing. Corroboration: Q3 infrastructure software revenue $8.752 billion (+29% YoY), below the $8.82 billion consensus; while semiconductor solutions came in at $20.839 billion (+127% YoY). Q4 guidance implies non-AI business growth of only +1.6% QoQ — that is the entire source of the "guidance miss."

Reading AVGO's guidance as "AI demand slowing" is therefore a directional error. In the same 8-K, Hock Tan's own words are: "In Q4 the momentum continues, and we expect AI semiconductor revenue to accelerate to $21.7 billion, up 236% year-over-year."

② AVGO's after-hours path: the 8-K sold it off, the call took it back — a precise mirror image of CRDO yesterday

This is the most important sequencing observation of the night.

• 16:05 ET 8-K released → immediate media reports of after-hours −5% to −6% (CNBC, Bloomberg and tradingkey all consistent), attributed to "guidance below consensus."
• 17:00 ET earnings call. Management disclosed material completely absent from the 8-K: Anthropic is expected to become the largest XPU customer in 2027 and continue into 2028, corresponding to more than 10GW of incremental silicon; AI orders $30 billion vs $11 billion already delivered; FY27 AI revenue guidance $100 billion.
• 18:03 ET measured after-hours $367.15 (CNBC) / $367.33 (stockanalysis), versus the $367.24 close essentially flat, on cumulative after-hours volume of 13.14 million shares.

Two independent quote sources gave the same price in the same minute, on volume equal to 35% of the regular session — this is not a thin-volume illusion, it is real price discovery. Combined with the −5% to −6% the media reported around 16:15, the most reasonable reading is: the 8-K's headline numbers knocked out a 5–6% decline, and the call took all of it back.

This is a precise mirror image of CRDO last night: CRDO's first after-hours read was only −4.8%, and the decline widened to −17.6% during the call (because Q&A could not provide a second-half breakdown). Both cases point to the same thing: for companies like these, the 8-K does not contain the decisive information — the call does.

⚠️ Three reservations that must be stated:

1. The $30 billion in orders, the $11 billion already delivered, and the $100 billion FY27 AI revenue figure all come from secondhand accounts of the call; this report did not obtain the original call transcript or 10-Q cross-verification.
2. All three are cumulative/annual figures, not single-quarter numbers. The verification point the pre-market list set for itself reads verbatim: "what is the single-quarter guidance number for next quarter's AI revenue (not cumulative, not annualized, not backlog)" — the answer to that question is $21.7 billion, confirmed in the 8-K; the three figures above cannot substitute for it and should not be read as single-quarter increments.
3. This report does not have the tick-by-tick after-hours sequence between 16:05 and 18:03; "sold first, then recovered" is an inference derived from three things (media timestamps + the 18:03 measurement + the large volume), not a direct observation.

③ HPE: tonight's purest "beat + across-the-board raise = sold off" sample

One search-side error must be corrected first: an Investing.com report returned by search said HPE fell because "FY26 adjusted EPS guidance of $2.20–2.40 came in below the $2.41 consensus." That is an old article. The original SEC 8-K EX-99.1 text gives FY26 non-GAAP EPS guidance of $3.75–3.85, and that is the raised number, differing from $2.20–2.40 by about 60%. The company's own numbers falsify it on the spot: Q3 actual non-GAAP EPS $1.11 + Q4 guidance $1.20–1.30 = $2.31–2.41 from just two quarters, which cannot equal a full year of $2.20–2.40. (Same type as the memory entry "recurring column headlines hide the publication date.")

What HPE actually delivered was a nearly flawless report (all from the original 8-K):

• Revenue $12.2 billion, +34% YoY, a record; non-GAAP EPS $1.11, beating consensus by 19.4%
• GAAP gross margin 40.1%, +1,090bp YoY and +360bp QoQ; non-GAAP operating margin 16.2%, +770bp YoY
• Networking revenue $2.9 billion, +74.9%, operating margin 22.0%; within it data center networking $382 million, +112.2%
• Revenue, EPS and free cash flow all raised for both FY26 and FY27; FY26 FCF ≥$3.75 billion, FY27 FCF ≥$5 billion; the CFO described order backlog as "at record levels"

Result: −4.59% after hours, on 10.77 million shares.

This report could not confirm a specific attribution for the decline, and explicitly declines to adopt that old article's explanation. Only two facts can be confirmed: ① the report itself was a beat plus an across-the-board raise; ② HPE had already closed the regular session up 1.89%, at the 72nd 52-week percentile. Under the "repriced ≠ realized" framework this looks more like a valuation event than a fundamentals event — but that is an inference, and this report does not write it as a conclusion.

④ Tonight's cross-section: beat magnitude and reaction still are not proportional

The mechanism the pre-market list proposed was "beat magnitude ÷ ex-ante implied volatility." Tonight SNOW (+37.8% → +21.8%) fits it perfectly, and AVGO (thin beat → flat) fits too. But HPE beat by 19.4% with all three items raised and still fell 4.6%, which the mechanism cannot explain.

The honest conclusion: tonight the mechanism went 2 wins, 1 loss and 1 draw — it is a tendency, not a law. The pre-market list already promoted a conclusion too early once yesterday on the basis of "three self-consistent samples" (the 52-week percentile); this report does not repeat that error — tonight's sample is not enough to write it up as a law.

---

5. Flows and Sentiment

① Sector rotation: a clean "low-valuation cyclicals ⇄ high-valuation software" switch

Leaders: KRE +2.23% > XLB +1.69% > XLC +1.39% > XBI +1.21% > SMH +0.96% > XLF +0.80% > XLV +0.75% Laggards: IGV −2.60% > XLRE −0.70% > XLK −0.02%

Three reads make today's structure clear:

1. IGV (−2.60%) and SMH (+0.96%) diverged by 3.56pp. Software was killed, semis were bought. This directly rejects the reading that "high-valuation growth was killed across the board" — semiconductor valuations are not low, and they rose. What was killed was software, a specific asset, not "growth" the factor.
2. XLK closed −0.02%, essentially dead flat. Because it holds software and semis simultaneously, the two hedge each other out. Using XLK to judge tech today produces the false conclusion that "nothing happened." You must break it down to IGV/SMH granularity.
3. KRE (+2.23%) was the strongest in the market. Weak ADP → front-end rate pressure eases → regional banks is the most direct macro transmission today, and the pre-market list did not cover it at all.

② Rates and volatility

• The 10Y closed at 4.79%, unchanged from 09-01; but it touched 4.818% intraday, the highest since November 2023. The equity rally corresponds in time to the retreat in yields from that high. The "flat at the close" read completely masks today's driver — you have to use the intraday high.
• 2Y 4.39% unchanged, 30Y 5.27% unchanged, 5Y −1bp. No change in curve shape today.
• VIX 15.20 (−6.98%), back into the low range seen since mid-August.
• DXY −0.11%, UUP −0.14% — the dollar barely moved. This matters: the pre-market list attributed the metals rally to "a weaker dollar," yet the dollar fell only 0.11% on the day, which cannot explain GDX +3.13%, still less the opposite-direction split between steel and precious metals (see §3②).
• TLT +0.10%, HYG +0.01%: no stress at either the bond or credit end, consistent with risk-on.

③ Risk-on / risk-off characterization

Characterization: risk-on, and specifically "a rate-pause-driven, low-valuation cyclical risk-on."

Evidence chain: the IWM (+1.18%) > DIA > SPY > QQQ (+0.23%) ranking, KRE/XLB/XBI leading, VIX −7%, roughly 2/3 of single stocks up, no credit stress.

But one exception must be flagged: software (IGV −2.60%) fell hard against the tape on a day when 2/3 of stocks rose and VIX fell 7%. This is not risk-off, it is an intra-sector valuation flush proceeding independently, decoupled from broad-market sentiment. This is also why today's "index up 0.46%" number carries almost no information for execution.

---

6. Next-Day Outlook (09-03, Thursday)

① Theme continuation

② Tomorrow's earnings and macro calendar

Macro (ET)

⚠️ The real adjudication this week remains the 09-04 (Friday) payrolls. Today's ADP print of 38,000 has already pushed expectations down; ISM services is the only release tomorrow that can materially change rate pricing — if it comes in clearly above 54.5, today's small-cap/bank/metals line gets taken back as a whole.

Earnings: AVGO/SNOW/HPE/NTAP already reported tonight, so tomorrow is their next-day pricing session, not a new earnings day. This report did not obtain a list of key 09-03 pre-market/after-hours earnings, and does not list a speculative one.

③ Key watch list (ticker + verification point)

④ What to avoid

1. Avoid treating AVGO's call figures as single-quarter reads. The $30 billion in orders and the $100 billion FY27 AI revenue are both cumulative/annual figures, and both are secondhand accounts. The only single-quarter number confirmed by the 8-K is the Q4 AI guidance of $21.7 billion.
2. Avoid bottom-fishing in software. IGV has fallen hard for two straight days, and tonight NTAP/HPE added two more "beat+raise sold off" cases. Until a "beat and then rise" sample appears, the burden of proof in this sector still sits with the bulls.
3. Avoid chasing SNOW's opening price. GTLB's full script yesterday (big after-hours gain → the open is the day's high → give it back) is less than 24 hours old.
4. Avoid reading today's Treasuries off the "flat close." The 10Y touched 4.818% intraday (the highest since November 2023) before retreating; the driver is intraday, not at the close.
5. Avoid extrapolating today's risk-on into a broad rally. Even as 2/3 of stocks rose today, IGV fell 2.60%. The index-level +0.46% carries almost no information for stock selection.
6. Avoid adding to small caps/banks/metals before tomorrow's 10:00 ISM. All three lines share the same premise (rate pressure easing), and one data point can knock out all three at once.

⑤ Input notes for tomorrow's pre-market list

1. The banks/financials line must be added. Today's list missed the strongest sector (KRE +2.23%), even though its logic (weak jobs → rate relief) was already written correctly by the list itself. Getting the logic right without landing it on a name is today's biggest structural gap in the list.
2. Verification points must be set on genuinely causal variables. The "Brent $95" verification point set for UAL today produced a false sell signal that very day (oil rebounded while UAL rose). Before setting a verification point, ask: have I verified the causal chain between this variable and the conclusion?
3. Do not define a whole day from 43 minutes of drift after the open. The DELL misjudgment came from exactly this, and was then elevated into a universal rule. An intraday fragment can only describe a fragment.
4. For any earnings name, the 8-K and the call must be distinguished. AVGO tonight (the call recovered 5–6%) and CRDO last night (the call knocked 13% off) are two sides of the same law. Reading only the 8-K will systematically misread the after-hours session.
5. Sector calls must be broken down to IGV/SMH granularity; XLK cannot be used. Today's XLK −0.02% completely masked a 3.56pp internal divergence.
6. Cross-sectional divergence can substitute for catalyst search. Steel versus precious metals moving in opposite directions proved, in the complete absence of any same-day news, that a driver exists and is not macro. "No catalyst found" is not grounds for a downgrade.

---

Internal Notes (not sent to clients)

Data channels

This report did not use yfinance (consistent with recent practice). All single stocks/ETFs went through api.stockanalysis.com/api/quotes/s/<ticker>; 77 of 83 targets were obtained, and all 6 failures were indices (/api/quotes/i/<spx|dji|comp|rut|vix|tnx> returns HTTP 400 across the board). Conclusion: stockanalysis's index endpoint path differs from the single-stock one or has been retired; indices switched to CNBC. Recommend locking this in.

After-hours fields confirmed available and reliable: the ep/ecp/eu fields all returned live values in the after-hours window (read here at 18:03–18:05 ET), matching CNBC's ExtendedMktQuote name by name (AVGO 367.33 vs 367.15, SNOW 372.5 vs 372.74, NTAP 164.94 vs 164.79, HPE 49.45 vs 49.51). Complementary to the memory entry "CNBC extended-hours fields unavailable pre-market": in the after-hours window both channels work and can cross-check each other, while pre-market only stockanalysis works.

⚠️ stockanalysis's epv (after-hours volume) is always null; after-hours volume can only be taken from CNBC's ExtendedMktQuote.volume. Every volume-based criterion in §4 of this report depends on that field — without it there is no way to distinguish AVGO's flat print (13.14 million shares) from DDOG's +4.27% (0.31 million shares). Recommend locking in: after-hours moves must be pulled on two channels, price from stockanalysis and volume from CNBC.

Treasuries: the official Treasury CSV already contained the 09-02 same-day data at 18:00 ET, so a primary source was available and CNBC's Treasury fields were not used (memory entry "CNBC Treasury change_pct can break field by field"). But the CSV carries only 2 decimals, and today 2Y/10Y/30Y are all "unchanged," making sub-1bp changes indistinguishable — the 4.818% intraday high comes from CNBC's same-day broadcast, not the CSV.

CNBC futures fields were unavailable today: @CL.1/@BZ.1/@GC.1 all returned last == previous_day_closing, change == UNCH, last_time == 01:00 ET. This is textbook silent staleness (same type as the memory entry "CBOE's _DJI/_COMP are dead data," with the same tell in last_time). All oil/gold prices in this report were switched to ETF proxies (USO/BNO/GLD/SLV) — the second direct hit for the memory entry "ETF proxies can detect silently stale spot quotes."

SEC primary sources: both AVGO's and HPE's 8-K EX-99.1 were retrieved with Bash + curl (with a User-Agent), HTTP 200. ⚠️ Hit one new trap: https://www.sec.gov/Archives/edgar/data/0001645590/...(CIK with leading zeros) returns 200 but with empty content; dropping the leading zeros, .../data/1645590/..., returns the real 679KB file. The former produced an empty file rather than an error, making it a silent failure. Recommend locking in: EDGAR Archives paths must use the CIK without leading zeros.

Search-side traps caught this time (three, all adjudicated by primary sources or in-house data)

1. The attribution for HPE's decline came from an old article, and the number was off by 60%. Investing.com's "FY26 adjusted EPS guidance $2.20–2.40 below the $2.41 consensus" — the original 8-K says $3.75–3.85, and that is the raised figure. There are two ways to catch it, neither requiring outside information: ① back-compute from the company's own numbers — Q3 actual $1.11 + Q4 guidance $1.20–1.30 = $2.31–2.41, meaning two quarters alone already hit the top of the supposed "full year," which cannot hold; ② the article's headline pattern coincides with HPE's fixed rhythm of reporting Q3 results every September. Same type as the memory entry "recurring column headlines hide the publication date." Had it been copied, §4③ would have stated a conclusion directly opposite to the 8-K — and been internally self-consistent throughout.

2. AVGO's "semiconductor revenue $16.7 billion beat the $15.2 billion consensus" is a definitional mix-up. The top search result wrote AI semiconductor revenue ($16.7 billion) as total semiconductor solutions revenue. Original 8-K: semiconductor solutions $20.839 billion, infrastructure software $8.752 billion, total $29.591 billion. Addition falsifies it on the spot ($16.7 + $8.752 ≠ $29.591). Without catching this error, the entire decomposition in §4① (that the non-AI business is the source of the shortfall) would have been impossible.

3. HPE's Server segment revenue was written up in a summary as $5.5 billion / +32.7%, while the original 8-K says $6.8 billion / +35.3%. The "Cloud & AI $7.7 billion +22.9%" in the same summary also fails to line up with the 8-K's segment definitions (the 8-K's Cloud & AI includes Server/Storage/Financial Services). The summary appears to have used the prior quarter's data.

⭐ The most important methodological finding this time: the media's after-hours move differed from the measured one by 5–6 percentage points, and both may be right

AVGO's after-hours price was the check this report spent the most time on, and also the most valuable one.

• Three search-side outlets (CNBC / Bloomberg / tradingkey) agreed: down 5%–6% after hours.
• Two independent measured channels (stockanalysis / CNBC extended quotes) both gave, at 18:03: essentially flat, on cumulative after-hours volume of 13.14 million shares = 35% of the regular session.

The first instinct is "the media were wrong" — that conclusion would itself be wrong. The correct reading is that the two have different timestamps: the media wrote a dozen-odd minutes after the 8-K was released; the measurement was read after the call ended. What happened in between was the call.

This is the same item as the memory entry "the 8-K has no guidance, the call does," but today it produced the price-pattern version of it, and in both directions:

• CRDO last night: −4.8% after the 8-K → widening to −17.6% during the call (Q&A could not provide a second-half breakdown)
• AVGO tonight: −5 to −6% after the 8-K → recovering to flat after the call (disclosing Anthropic / $30 billion in orders / FY27 $100 billion)

Recommend recording this in memory under the title "an after-hours price must be asked whether it was read before or after the call." Criterion: for any after-hours move, record both the read timestamp and the call time (usually 16:30 or 17:00 ET); any read between 16:05 and 17:00 is only a reaction to the 8-K, not that evening's final pricing. This also explains the loss the list took on PANW yesterday (thin-volume after-hours read −0.02%, next day −8.24%) — but the mechanism differs: PANW was too thin, AVGO was too early. Both distort after-hours reads, but the remedies are opposite (wait for volume in the first case, wait for the call in the second).

⚠️ This report's honest boundary: I do not have the tick-by-tick after-hours sequence from 16:05 to 18:03, so "sold first, then recovered" is an inference. If AVGO gaps up tomorrow, the inference is confirmed; if it gaps down sharply, then that flat 18:03 read is itself suspect and this item needs to be overturned and rewritten.

Reconciliation against the pre-market list's unclosed gaps

Data not obtained / left blank in this report

• Precise closing advance/decline counts: CNBC's .ADV.NYSE/.DECL.NYSE and similar symbols all returned null, and the stockanalysis screener endpoint 404s. The main text uses only CNBC's intraday broadcast of "roughly 2/3 up," flagged as an intraday basis, and fabricates no closing figure.
• AVGO call transcript: only secondhand accounts were obtained; the $30 billion / $11 billion / $100 billion figures were not verified against a primary source, and the main text says so.
• Specific attribution for HPE's after-hours decline: not obtained. The main text explicitly says "could not confirm" and does not adopt the old article's explanation found in search.
• NTAP's specific free cash flow figure: search says a "decline in FCF" caused the drop; no primary source was checked, and the main text only records the media attribution without endorsing it.

---

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