U.S. Pre-Market Brief | 2026-08-28 (ET) Friday

Coverage window: 2026-08-27 16:00 ET regular-session close → 2026-08-28 08:29 ET (includes 8/27 after-hours earnings, the overnight Asian/European sessions, and this morning's pre-market). All data in this piece is as of 08:29 ET; price moves after that are not included. Three quote snapshots were taken: ① 08:01–08:08 (breadth scan, full-market pool), ② 08:17–08:19 (re-check of key names), ③ 08:24–08:29 (second re-check of key names) — all three are labeled column by column in §1.6. Source: the stockanalysis.com quote API (pre-market price / percent change / cumulative pre-market volume all from the same source at the same moment). ⚠️ Clock discipline: the semiconductor breadth table in §5⑩ uses snapshot ① throughout, with no mixing of clocks; the key names in §3.1/§3.2/§5/§7/§9 uniformly use snapshot ③. For every pre-market number cited, the snapshot it belongs to can be cross-checked in §1.6. Read timestamps for rates, VIX, crude, gold, the dollar and the Asian/European indices: 08:03–08:08 ET, source CNBC restQuote API. Earnings-figure basis: the core figures and guidance for eight companies — MRVL / AFRM / GAP / ULTA / RBRK / ADSK / S / WDAY — have been verified line by line against SEC primary filings (8-K EX-99.1, 10-K, 10-Q, XBRL); these are labeled "primary source" in the text, while anything relayed only by media or the earnings call is labeled "media-sourced" / "secondary transcript". All consensus figures come from third-party relays and have not been checked against a terminal — this piece therefore switches in several places to "new guidance vs the company's own prior guidance" as the criterion; see §5③ for why. ⚠️ Pre-market volume is cumulative from 04:00 ET, not an instantaneous reading; every volume-based criterion in this piece is converted to a notional amount (shares × pre-market price), and the same threshold is applied to both the long and short sides.

---

0. Today in One Line

1. Today's protagonist is not a single stock — it is the podium at 10:00 ET. Fed Chair Kevin Warsh delivers his first Jackson Hole keynote since taking office (10:00 ET), colliding head-on with the final UMich consumer sentiment reading (10:00 ET), with the Chicago PMI (09:45 ET) just ahead of it. The backdrop is stagflationary: inflation 3.4% (media-sourced; this piece has not verified which measure or which month), the 30-year Treasury at 5.21%, and UMich's August preliminary reading already collapsed to 51.0 (prior 55.2, expected 54.5). And this year's Jackson Hole theme is "Financial Innovation: Implications for Payments and Policy," not monetary policy — "he said nothing at all" is a far-from-negligible probability.

2. The cleanest single event is the collapse of an M&A premium. PYPL is −14.31% pre-market to $52.68, on roughly $240 million of pre-market notional — the Advent + Stripe consortium has walked away from the $53 billion acquisition (offer price $60.50/share). But the premium has not been fully given back: the close before the deal rumor (7/14) was $47.37, and the current price is still 11.2% above that. Meanwhile yesterday's close of $61.47 was itself above the $60.50 offer, which means the market was pricing in "a raised bid or a second buyer" — two layers of expectation broke today.

3. The bar for AI hardware has been raised a notch: a revenue beat is no longer worth anything; the direction of gross margin is. MRVL beat on both lines and guided Q3 revenue to $3.15 billion, above consensus, yet it is still −7.36% (snapshot ③, 08:29) pre-market on roughly $489 million of notional — the largest in the market. Primary source confirms: Q2 non-GAAP gross margin 58.9%, Q3 guidance 57.5%–58.5%, midpoint 58.0%, precisely −90bp sequentially, which the CFO attributes to custom ASICs ramping, at gross margins below standard products. Management is also reported to have said FY28 gross margin will be "similar to this year's exit rate" — i.e. zero gross-margin improvement while revenue goes from $12B to $18B. ⚠️ Note: the 90bp for Q3 is confirmed by the SEC 8-K primary source, but the FY28 remark appears only in secondary transcripts; this piece did not obtain a verbatim transcript, so that evidence sits one grade lower.

4. Five software prints: the only one whose margin guidance went up — ESTC — rose 22%, while the ranking of this morning's declines among the other four is almost exactly the inverse of the ranking of their guidance quality. ESTC +22.24% to $102.36, a fresh 52-week high (operating margin 16.2% → Q2 guidance 19.0% → full-year 19.4%). And after comparing each company's new guidance against its own prior guidance line by line (consensus is merely an echo of guidance and gets revised up within 12 hours of the print, so it cannot serve as the baseline):

   • RBRK, down the most (−6.28%), raised every single guidance line, and the size of the raise ($46 million) exceeds the size of the quarter's beat ($31 million) — even the second half was raised.
   • ADSK's (−4.65%) "free-cash-flow cut": the primary source shows the low end of guidance was not moved at all, only the top end was trimmed, and the new guidance includes roughly $45 million of MaintainX acquisition transaction costs; strip that out and the underlying cash-flow guidance was actually raised.
   • S (−2.69%) did have its EPS guidance cut by 11.4%, but in the same 8-K operating income guidance was raised 5.0% — the entire gap sits below the operating line (about $19 million of non-operating income assumptions plus 11 million shares of dilution).
   • And WDAY — the only one to cut its implied second half, with total backlog growth down to +8.0% (already below current revenue growth) — went from −0.29% to +1.22% this morning.

   But this must be stated smaller, not bigger. RBRK went into the print up +11.33% on the day, closing at the 98.6th percentile of its 52-week range; its decline is fully explained by "position." ADSK's −4.65% is on just 8,750 shares ($2.26 million), explainable by "thin volume." S was +10.73% yesterday and is giving back only 2.69% today. All three have ready-made explanations that do not require invoking "the market is wrong." The only one that cannot be explained by position, thin volume or profit-taking is WDAY: its fundamental evidence chain points most completely to deterioration, and it is up this morning. So this piece's claim is singular — "the price of this one stock, WDAY, is moving opposite to its fundamentals" — not "the software sector is mispriced."

5. Three consumer prints, three kinds of "illusion," each pointing a different way — this is the passage that most needs slow reading today.

   • AFRM (+13.16%): of the reported EPS of $4.62, $4.14 (90%) is a non-cash $1.465 billion reversal of a deferred tax asset valuation allowance (primary source: FY26 10-K). Stripped clean, operating EPS is about $0.43–0.48, and RLTC +39% is faster than GMV +36% (unit economics are improving; this is not subsidizing for volume). The headline is definitely fake; whether the beat is real depends on which consensus you take — ⚠️ two mutually exclusive versions were found, $0.33–0.35 and $0.85. Take the former and it is a 25%–45% beat; take the latter and it is a large miss, and the conclusion flips entirely. This piece uses the former (which has an explicit Zacks source), but it is the single most fragile link in the whole piece; see §5⑤.
   • GAP (+17.99%): adjusted EPS of $0.52 did beat consensus, but GAAP EPS of $1.38 includes a $417 million tariff refund credited against COGS ($0.86/share); net sales −2.0% YoY, adjusted operating income −11.3% YoY; the $0.05 raise to full-year EPS guidance is, once decomposed, 100% attributable to the share count falling from 375 million to 367 million, while the top end of the sales growth range was actually cut. The raised guidance implies second-half adjusted EPS of −10.4% to −0.9% YoY.
   • ULTA (−1.21%): it raised guidance across the board, it did not maintain it (EPS $28.36–28.80 → $28.70–29.00; the press release headline literally says "AND RAISES"). The reason it fell can be quantified precisely: the quarter beat by $0.38, but the midpoint of full-year guidance was raised by only $0.27 — the $0.11 difference was taken out of second-half expectations.

6. Pre-market looks like a "breadth repair," but that description has to be discounted. Yesterday only 1 of the 11 SPDR sector ETFs closed higher (XLK +3.16%); this morning it is the reverse: within this piece's semiconductor watch pool, 13 of the 15 names with notional ≥$2 million are down and 1 is up, while a sample of 14 large-cap blue chips that fell yesterday is 14 up, 1 flat. The indices have barely moved: SPY 0.00%, DIA +0.08%, QQQ −0.28%, IWM −0.13%, VIX 14.48.

   ⚠️ But neither statistic is full-market breadth: the blue-chip group was selected on the condition "fell yesterday" — that is sampling on the dependent variable, and bid-ask mean reversion plus pre-market noise alone will produce "down yesterday, up today"; 0 of the 15 being down looks more like thin, positively-biased pre-market quotes than genuine breadth. The semiconductor group is likewise this piece's self-selected watch pool, not a full SOX constituent scan. The correct statement is "the blue-chip sample that fell yesterday is broadly rebounding this morning while the semiconductor sample is broadly giving back," and no full-market breadth can be inferred from it. Real breadth has to wait for per-stock advance/decline counts after 09:30.

---

1. News Overview

---

1.5 Pre-Market Index and Cross-Asset Readings (08:03–08:08 ET)

Overnight Asia/Europe: Nikkei 225 66,405.56 (+0.41%), Hang Seng 25,584.79 (+0.07%), FTSE 100 10,809.58 (+0.16%), German DAX 26,499.14 (+0.50%) — all modestly higher; the overnight session gave no direction, so today's direction has to be set by the U.S. itself.

How to read the curve: 3M 3.783% → 2Y 4.234% → 10Y 4.692% → 30Y 5.210%. The front end has barely moved (2Y +0.2bp) while the long end is being sold (10Y +2.0bp, 30Y +1.9bp). That is term premium moving, not rate-cut expectations moving — ahead of Warsh's speech, the market is not pricing "dovish."

1.6 Three-Snapshot Comparison for Key Names (08:01–08:08 → 08:17–08:19 → 08:24–08:29 ET)

Why this table exists: pre-market quotes are cumulative, and a single-point snapshot is not a fact. Reading the same set of names again 12 minutes later shows whether the gap is widening or narrowing — that is more informative than any static number.

Three readings:

1. Everything that is down is narrowing, everything that is up is strengthening, and nothing is deteriorating. There is no panic selling pre-market; it looks more like orderly turnover — this is the same phenomenon as item 6 in §0 ("breadth repair"), seen from another angle.
2. WDAY going from −0.29% to +1.22% is the biggest directional change in the table, and its notional ($20.55 million) is the best of the four software earnings names. And it happens to be the only one of the four whose fundamentals genuinely deteriorated (see §5⑦).
3. AVGO went from +0.12% to +0.38%.

⚠️ The most important use of this table is self-restraint, not evidence.

• The same stock can read 1.5 percentage points apart within 28 minutes (WDAY). No single-point pre-market snapshot is a fact — including every one cited in this piece.
• Therefore the "magnitude of change" in readings ② and ③ is not itself evidence: AVGO's 26bp is within the pre-market bid-ask spread; WDAY's 1.5pp is far larger and backed by $20.55 million of notional, but it is also the very example this table uses to show that snapshots cannot be trusted.
• Rule: pre-market directional changes are treated as leads only, never as arguments. Every falsifier in §8 starts reading at 09:30.

---

2. Strongest Themes, Descending

---

3. Overall Single-Stock Strength Board (long-side scoring + short-side watch list)

Ranking basis: descending by a composite of "news quality × degree of primary-source verification × pre-market notional × executability." Pre-market notional = cumulative pre-market volume × pre-market price. ⚠️ Clock: this section and §3.2 uniformly use snapshot ③ (08:24–08:29 ET); where a name only has an earlier reading, the timestamp is noted in the cell (e.g. MSTR is 08:02). The semiconductor breadth table in §5⑩ uses snapshot ①; the two tables cannot be compared directly.

3.1 Long / Watch Side

3.2 Short / Avoid Side (descending by pre-market notional — the larger the notional, the more credible the signal)

No total scores in this section. The scoring model in §4 was designed for "long candidates you could actually enter" (moat, earnings elasticity, catalyst persistence and similar dimensions are not comparable on the short side); forcing it onto short candidates would produce a misleading ranking. The short side gets direction, evidence and risk only.

3.3 Breadth Evidence (evidence only, not entry candidates)

---

4. Single-Stock Scoring Model (100 points)

Note that ULTA (59) scores higher than GAP (45), while today ULTA is falling and GAP is up 18%. This is not a scoring error — the score measures fundamental quality and information content, not today's direction. These two things are orthogonal, and §9④ will say so again.

---

5. Detailed Analysis of Top Names

① PYPL — PayPal | The $53 billion offer is withdrawn, and the premium has not been fully given back

• Related news (8/27 after hours): Advent International and the Stripe consortium have abandoned their acquisition of PayPal. The offer was $60.50/share, over $53 billion, made in mid-July; the board rejected it as "inadequate, with regulatory and financing risk," after which the consortium withdrew.
• ⚠️ The evidence grade has to be stated first: both links obtained for this piece (Benzinga, Tradingpedia) are themselves relaying a Bloomberg report; this piece did not obtain the original Bloomberg/Reuters text, there is no PYPL 8-K, and there is no public statement from PayPal or from either side of the consortium. The "board's wording" and "the consortium fully withdrew" above are media relays, not company-confirmed facts.

  This implies a specific risk: deal-break news being denied, corrected, or followed by a second buyer is a common script. If any party issues a formal statement during today's session, −14.4% can reverse within minutes. This piece has no primary source that can rule that out. That is also why the impact level for this item was cut from S to A (media-sourced).

• Catalyst logic: it does not affect any quarter's earnings; what it affects is the "acquisition option" embedded in the valuation. The transmission is pure: arb positions unwind unconditionally.
• Price archaeology (daily bars pulled for this piece, source: stockanalysis history page):

• Two non-consensus readings:
  1. Yesterday's close of $61.47 was itself above the $60.50 offer. What the market was pricing then was not only "the deal will close" but also "there will be a raised bid or a second buyer." Two layers of expectation broke today.
  2. The premium has not been fully given back. $52.675 is still 11.2% above the pre-deal $47.37.
• But $47.37 is not a clean floor: since the rumor, PayPal has reported Q2, beating expectations and raising full-year profit guidance; new CEO Enrique Lores (who took office in March 2026) has announced a strategic reorganization and will for the first time break out the three main businesses in disclosure with individual revenue targets. How much of that 11.2% is residual premium and how much is fundamental improvement since July cannot be separated by this piece — that is the biggest uncertainty in this name today, and the reason not to chase it short.
• Pre-market: −14.40% to $52.62 (08:29), notional about $250 million; the decline went from −15.54% at 08:08 → −14.31% at 08:19 → −14.40% at 08:29, i.e. narrowed and stabilized, which indicates buyers are absorbing. 52-week range $38.46–$79.215.
• Final judgment: Avoid / short watch, do not chase. Observation points in §8.

② MRVL — Marvell | This is not a sell-the-news top at a high; it is a print that arrived after four weeks of +48%

• Primary source (SEC 8-K EX-99.1⁵):

• Q3 guidance (primary source): revenue $3.150B ±5%; non-GAAP gross margin 57.5%–58.5% (midpoint 58.0%); non-GAAP EPS $1.10 ±$0.05; opex ~$655M (+7.2% QoQ); share count 921M.

  58.0% − 58.9% = −90bp, an exact match. This most widely circulated number is confirmable from a primary source.

• Catalyst logic — this piece's core argument, now in a harder version:

  1. Short term: CFO Dan Durn attributes it explicitly to product mix — "the expected acceleration in the custom business constitutes a sequential headwind in the third fiscal quarter." Custom ASIC gross margins are below Marvell's standard products (optical DSP, switching, broadband analog). It is not demand, it is not price cuts; it is that "the business being won carries a lower gross margin."
  2. The long-term layer sits one evidence grade lower and must be flagged first:

  ⚠️ The three figures below appear only in media relays of the earnings call — this piece did not obtain a call transcript, Marvell IR's quarterly-earnings page returns HTTP 404, and the 8-K EX-99.1 contains no FY28 figures at all. They are secondary transcripts, unverified against a primary source.

  Management reportedly gave three figures at the same time: FY28 revenue of about $18B (+50%), FY28 gross margin "similar to this year's exit rate" (≈58%), and operating margin heading to 38–40%. Using $18B × 58% = $10.4 billion of gross profit to back out an opex ceiling:

  • if OpM is taken at the top of the range, 40% → opex of about $3.2 billion, only 28% above FY27's roughly $2.5 billion;
  • if OpM is taken at the bottom of the range, 38% → opex of about $3.6 billion, allowing a 44% increase. The two ends of the range differ greatly and the conclusion is highly sensitive to the endpoint, so this piece claims direction only, not magnitude.

  What has to be said clearly is that "if gross margin is flat, all the leverage comes from the expense line" is an identity, not a discovery. What actually carries information is the forward statement itself: if FY28 gross margin really is flat versus this year, then the market's default assumption that "AI semiconductor gross margins will rise as volume ramps" simply fails — this round of growth was bought with gross margin, and it has to keep being bought all the way to FY2028. At 20x P/S, that confirmation is far more expensive than the 90bp itself. And that forward statement is precisely the weakest link of evidence in this piece.

• Why it fell despite a beat and raise — "expectations fully priced" has a precise price trajectory this time:

From the 7/29 low to the 8/27 close, +47.8% in four weeks, almost entirely driven by the Google deal. The current price of $223.67 is −32.2% from the 52-week high and sits around the 60th percentile of the 52-week range ($61.44–$329.88) — this stock is not at a high at all. What it is falling on is not this print; it is that this print cannot catch up with the price paid in advance for it over the past four weeks.

• The decisive mathematical blow: third-party consensus (still the pre-print stale values as of writing) is FY27 revenue $11.56B / EPS $4.05, FY28 revenue $16.88B / EPS $6.32. Last night the company guided FY27 to about $12B (= the top of the consensus range) and FY28 to about $18B (above the top of the consensus range).

  But backing out $18B of revenue × 58% gross margin × an enlarged share count gives FY28 non-GAAP EPS of about $5.85–6.68, whose center essentially coincides with the existing consensus of $6.32. Revenue consensus needs a 6.6% upward revision while EPS consensus barely needs to move — that is the mathematical reason the "raise" did not help.

• Two risks not previously discussed (primary source):

  1. The cost of equity is deteriorating: SBC as a share of revenue rose from 7.66% to 11.91% (SBC +112% YoY, three times the revenue growth rate); diluted share count +5.84% YoY; H1 buybacks of $400 million covered only 75% of H1 SBC of $534 million. non-GAAP EPS growth (+40.3%) is 7.6pp below non-GAAP net income growth (+47.9%) — dilution ate roughly one sixth of the profit growth.
  2. Potential dilution from the Google warrant (8-K primary source¹⁹): 58.97 million shares, strike $206.58, expiring 2033; 57.61 million of those vest on performance, in 240 tranches, one tranche per $500 million of custom product revenue, measured from FY27Q3 (i.e. the current quarter). Full vesting requires cumulative custom revenue of $120 billion. 58.97 million shares ≈ 6.72% of the existing share count; adding the Celestial AI earnout, potential dilution is about 9.8%. And the strike of $206.58 is only 7.6% below the pre-market $223.67.

• Segments (primary source; note: since FY26 the company no longer gives quantitative disclosure for the four-way split of enterprise networking / carrier / consumer / auto-industrial):

Data center contributed 92.9% of the YoY increment, while the communications business has now declined sequentially for two consecutive quarters and is a net drag. Marvell is in fact already a single-end-market company, and the 20% communications business not only fails to smooth the cycle, it is bleeding at the same time.

• Final judgment: Avoid / short watch (do not chase). The position (52-week range $61.44–$329.88, about the 60th percentile) does not support chasing it short, and the long-term gross-margin conclusion does not support buying the dip. Today is not the day to take a directional view on this name.

③ ESTC — Elastic | The only software print last night where margin guidance went up

• Primary source (Elastic IR 8-K⁶): FY27Q1 (ended 7/31) revenue $478.1 million (+15%); subscription revenue $449 million (+15%); cloud revenue $235 million (49% of the total, +20%); non-GAAP EPS $0.70; non-GAAP operating margin 16.2%; adj FCF $143 million; cRPO +21%; a record net addition of customers with >$100K ACV; cash and equivalents $881.8 million.

• Guidance (the real reason for the stock move): Q2 revenue $486–487 million (+14.9%), EPS $0.80–0.82, operating margin about 19.0%; full-year revenue $1.998–2.010 billion (+15.2%), EPS $3.29–3.37, operating margin about 19.4%, adj FCF margin about 21.5%.

• Comparison with the other four (the most informative table in the piece)

  The criterion was switched once, and that matters: instead of "vs consensus," this uses "new guidance vs the company's own prior guidance." The reason is that consensus is merely an echo of guidance — RBRK's third-party FY27 consensus ($1.64 billion / $0.31 / FCF $298 million) happens to equal, line by line, the midpoints of its prior guidance, while ADSK's consensus has already been revised up to equal the midpoint of the new guidance. Mixing the two produces conclusions pointing in opposite directions. The company's own old guidance is the only baseline with an unambiguous timestamp and a consistent basis.

① ESTC is the clean positive case — the only one of the five whose growth, margin and cash-flow guidance all moved up together; the 22% gain is justified. ② The price ranking of the other four is almost exactly the inverse of the guidance-quality ranking. That is the factual layer, verified line by line against primary sources.

③ But "inverse" does not equal "mispriced," and this step has to be taken very slowly. All three names have ready-made explanations that do not require invoking "the market is wrong":

• RBRK: went into the print up +11.33% on the day, closing at the 98.6th percentile of its 52-week range — position alone explains the decline.
• ADSK: the −4.65% is on only 8,750 shares / $2.26 million — thin volume explains it, and the magnitude was never credible.
• S: +10.73% yesterday, giving back only 2.69% today — profit-taking explains it.

④ And there is an even more fundamental counter-explanation, with the counter-evidence sitting inside this very table: RBRK's own Q2 guidance was non-GAAP EPS of $0.03–0.05 and it printed $0.20, i.e. it sandbagged itself by 4–5x. For a company that habitually guides far too conservatively, the information content of "all four guidance lines raised by $46 million" may be close to zero — the market not buying it is rational, not mispricing. This piece was unable to obtain RBRK's "own guidance vs actual" deviation over the past four quarters to calibrate this, therefore: "all four lines raised" is used in this piece as a statement of fact only, not as evidence of mispricing.

⑤ The claim therefore narrows to one stock: WDAY. It has no position explanation (only +1.48% yesterday), no thin-volume explanation ($20.55 million is the best of the four), and no sandbag explanation (what it cut was its own implied second half). "WDAY's price is moving opposite to its fundamental evidence" is a singular claim; "the software sector is mispriced" is a universal one — this piece asserts only the former.

• Fundamentals: three lines moving the same way — growth (revenue +15%, cloud +20%), leading demand indicator (cRPO +21%, above revenue growth), and earnings quality (operating margin expansion + FCF margin of 21.5%).
• ⚠️ Position risk — the single most important sentence about this stock: the pre-market price of $102.36 (08:27) is already above every published pre-print sell-side price target — Cantor Fitzgerald raised to $91 on 8/24 (still Neutral), Barclays to $94 (OW), JPMorgan to $93 (OW) — all three are below the current price. There will almost certainly be a new round of raises today, but until new targets land, buyers are taking stock at a level with no sell-side support. At the same time, $102.36 is already above the 52-week high of $96.065, so there is neither trapped supply nor any reference point above.
• Valuation: at the FY27 EPS guidance midpoint of $3.33, about 30.7x forward P/E — not outrageous for a software company with 15% growth, a 19.4% operating margin and a 21.5% FCF margin, but no longer cheap.
• Final judgment: Watch closely, do not chase. The cleanest fundamentals in the market today, but the price has run beyond every known anchor.

④ GAP — Gap Inc. | The most important downgrade in this piece: beneath an +18% gap, the core business is shrinking

This section is the one place in the piece where I changed the draft's conclusion from "watch closely" to "watch only." The reason is that primary-source verification diverged too far from the media version.

• First, a correction to a widely quoted basis (SEC 8-K EX-99.1, primary source⁸): the circulating "Q2 EPS $0.52" is the adjusted figure; GAAP diluted EPS is $1.38. The $0.86 difference comes entirely from the IEEPA tariff refund: Q2 booked a $417 million credit to COGS ($512 million of refunds − $95 million of committed supplier compensation), of which only $95 million was actually received in Q2, with the rest expected in Q3.

• The real operating picture after removing one-time items (primary source):

A textbook case of "beating guidance ≠ beating last year": versus consensus ($0.48–$0.50) it is a beat; versus last year, adjusted operating income is −11.3%, net income −12.0%, EPS −8.8%. The buyback effect can be quantified: had the share count stayed at 379M, adjusted EPS would be $0.501, so buybacks contributed +$0.024, cushioning a −12.0% decline in net income into an −8.8% decline in EPS.

The source of the margin decline is equally clear: adjusted gross margin improved only 20bp while the operating expense ratio worsened by 92bp. Expense deleveraging under negative revenue growth — that is the core problem with this quarter's operating quality.

• Comparable sales by brand — a 23% engine cannot rescue a 58% drag:

Old Navy flipped from +2% to −4%, which the company attributes to pressure in seasonal women's categories plus an "unanticipated slowdown in traffic"; a leadership change at Old Navy was announced at the same time (Michael Francis taking over as CEO) — the change itself confirms how serious the brand's problem is. Athleta is in its second consecutive year of double-digit declines.

• The $0.05 full-year guidance raise decomposes 100% to buybacks:

Recomputing with all of the company's own parameters (accurate to one cent, reproducible): if you take every operating parameter from the new guidance but keep the old share count of 375M, EPS is $2.36 — exactly the midpoint of the old guidance. In other words, the $0.05 raise comes 100% from an 8 million share reduction; the +10bp of operating margin improvement is almost entirely offset by the cut to the top of the sales growth range and the higher tax rate.

• The most critical layer: even after the raise, guidance still implies a YoY decline in the second half. Backing out from the GAAP guidance midpoint of $3.82 − the adjusted midpoint of $2.40 = $1.42 of one-time tariff EPS (of which $0.86 was booked in Q2, leaving $0.56 in Q3), H1'26 adjusted EPS = $1.40 (vs H1'25 $1.07, +30.8%), which makes implied H2'26 adjusted EPS = $0.95–$1.05, vs H2'25's $1.06 = −10.4% ~ −0.9%.

  The raised full-year guidance implies a second half that is flat to down 10% year over year — and that is with a 4.4% smaller share count.

  ⚠️ But this item has to carry its own warning, because it is the pivot of this section's downgrade: the H2 implied figure above is backed out, and its implicit assumption is that "the full-year gap between GAAP and adjusted EPS ($1.42) comes entirely from the tariff refund" — if it also contains restructuring, impairment or Athleta-related charges, the H2 range has to be recomputed. One hint that this assumption may be missing items: on the same back-out, Q1'26 adjusted EPS ≈ $0.88, which against Q1'25's $0.50 is +76%, while Q2 is −8.8% — a +76% / −8.8% swing between two adjacent quarters at the same company is not natural. The correct verification path is to take the already disclosed actual Q1 adjusted EPS and add Q2's $0.52 to get H1 directly, with no back-out needed; this piece was unable to obtain the Q1 figure before publication to complete that step. So the correct statement of this conclusion is: "under the assumption that the full-year GAAP-minus-adjusted difference is entirely tariff-related, the back-out implies H2 of −10.4%~−0.9% YoY; that assumption has not been cross-checked against Q1 data." The rest of the downgrade evidence (net sales −2.0%, adjusted operating income −11.3%, the guidance raise being 100% from buybacks, Old Navy flipping from +2% to −4%) are all numbers disclosed directly by primary sources and do not depend on this back-out.

• The tariff itself is noise: after the July Section 301 announcement, the company cut its incremental tariff rate assumption from 7/24 onward from the high teens to about 10%, for a full-year net benefit to gross/operating profit of only about $15 million (about 10bp), concentrated in Q4. What actually moves the financials is that $512 million one-time refund, which is not repeatable.

• Financial health (it is not without merits): cash + short-term investments $2.485 billion, long-term debt $1.493 billion, net cash of +$992 million excluding leases; H1 operating cash flow $550 million (versus $308 million a year ago); inventory flat YoY (healthy); valuation only 10.2x (on the adjusted guidance midpoint of $2.40 and the pre-market price of $24.53).

• Pre-market: +17.99% to $24.53, notional about $10.52 million; its 52-week percentile jumps from 24% to 54% — that single jump takes GAP from the low end of its range straight to the middle.

• Final judgment: Watch only.

  ⚠️ This must be said clearly: this is a judgment about "quality," not a prediction about "today's direction." A company at 10x P/E with $992 million of net cash that just beat expectations can perfectly well keep going up; CRM yesterday was a live example of "the fundamental decomposition was entirely right and the ranking entirely wrong." This piece's claim is falsifiable and consists of exactly one sentence: GAP's +18% is built on a tariff refund and a buyback, not on the business getting better. Verification point in §9①.

⑤ AFRM — Affirm | The headline is fake, the beat is real, and the magnitude is 25%–45%, not 13x

• The precise amount of the one-time item (primary source: FY2026 10-K⁷):

Contribution to EPS: $4.14, or 90% of the reported $4.62. In the shareholder letter, CEO Levchin himself calls this net income "(currently) ludicrous $1.6 billion" — the company is not trying to dress it up as an operating result.

• The real beat after stripping it out:

• Once stripped clean, the operating layer is actually more worth looking at (primary source: shareholder letter EX-99.1):

The most solid piece of evidence: RLTC growth (+39%) is faster than GMV growth (+36%), and RLTC/GMV rose from 4.09% to 4.18%. Every dollar of GMV earns more, and it is not being bought with subsidies. And this has nothing whatsoever to do with the tax item.

• FY2027 guidance (primary source): GMV >$64.0B, adjusted operating margin >30.5%; FQ1 revenue $1.19–1.22 billion. ⚠️ But the implied growth rate is decelerating: $64.0B vs FY26's $50.2B = +27.5%, a clear slowdown from FY26's +37%. The company names two headwinds itself: the large merchant promotion in FQ1'26 will not repeat, and one enterprise merchant is migrating Pay Later volume to its own wallet (a drag expected to fade only in the final month of FQ1'27).
• Two quality flags that must be raised:
  1. 26.8% of shareholders' equity is now a deferred tax asset ($1.467 billion / $5.484 billion), whose value depends on being able to use roughly $3.2 billion of federal NOLs in the future. This is paper equity, not cash.
  2. Credit metrics are drifting mildly: the credit loss provision rate went 5.65% → 5.89% (+24bp); the loan modification ratio 0.17% → 0.25%. Neither move is large, but the direction is downward, in mild tension with the "steady credit outcomes" language.
• Valuation (you have to compute it yourself; third-party data is unusable): third-party sites show a TTM P/E of 13.54 and a margin of 45.29%, all contaminated by that $1.465 billion tax item; their "net cash of −$8.36 billion" treats the $9.99 billion of funding liabilities used for lending as corporate leverage — for a balance-sheet lender that is simply the wrong basis; the company's own basis is net cash of +$1.5 billion. Cleaned up: TTM P/E about 63x (on the pre-market price); forward P/E about 53x. FY26 operating cash flow $1.231 billion (+55%).
• Pre-market: +13.16% to $87.69, notional about $29.92 million; 52-week range $42.095–$100.00, percentile up from 61% to 78%.
• Final judgment: Watch closely. Stripped clean, this is the best operating quality of the three consumer prints. But anyone buying it must know they are buying $0.43–0.48 of operating EPS, not $4.62.

  ⚠️ The most fragile link in this section: the consensus figures retrieved come in two versions, $0.33/$0.35 and $0.85. If the true consensus is $0.85, then $0.43–0.48 after cleaning is instead a large miss and the whole conclusion of this section reverses. This piece uses $0.33–0.35 (with an explicit Zacks source), but that dependency is explicitly registered here.

⑥ ULTA — Ulta Beauty | It raised guidance; it just did not raise it enough

Correcting a fact I got wrong in the draft: ULTA did not "maintain" guidance, it raised it across the board. The press release headline itself carries "AND RAISES FISCAL 2026 GUIDANCE."

• Q2 actuals (primary source): revenue $3.0357 billion (+8.9%); comparable sales +3.8% (prior year +6.7%, a 2.9pp deceleration); gross margin 39.1% (−10bp); operating income $379.6 million (+10.1%); net income $282.0 million (+8.1%); diluted EPS $6.55 (+13.3%), consensus $6.17 (a 6.2% beat).
• Why it fell despite a beat and a raise — two quantifiable explanations:
  1. The raise is smaller than the beat. The quarter beat by $6.55 − $6.17 = +$0.38; the midpoint of full-year guidance was raised by only $28.85 − $28.58 = +$0.27. The −$0.11 difference amounts to taking it out of second-half expectations. Looking at the endpoints makes it clearer: the low end rose $0.34 (more than the beat), but the top end rose only $0.20 — the ceiling was lifted by less than the beat.
  2. The raised guidance implies a clear second-half deceleration: H1 operating income +10.9% → implied H2 +5.8%~+7.8%; H1 EPS +14.6% → implied H2 +9.4%~+11.7%.
• Three earnings-quality flags:
  • About 5.2pp of the +13.3% EPS growth comes from buybacks (operating income +10.1% → net income only +8.1%, because interest swung from net income to net expense).
  • Revenue +8.9% but organic comparable sales only +3.8%, with the difference coming from the Space NK acquisition and new stores; and that acquisition is precisely the reason gross margin was −10bp (per the company). Revenue growth has now decelerated for four consecutive quarters: +12.95% → +11.78% → +11.08% → +8.87%.
  • H1 buybacks were $791 million while H1 free cash flow was only $242 million; the gap was covered by a $278 million net increase in short-term debt and a drawdown of cash (from $424 million to $158.5 million), and the buyback authorization was raised to $1.8 billion. This is doing buybacks with leverage and cash reserves — it is both the mechanism behind EPS growth persistently exceeding operating income growth and a question mark over sustainability.
• Valuation and position: P/E 18.5x (on the guidance midpoint of $28.85); 52-week percentile 36% → 33%, the only one of the three consumer names whose position did not improve. Net debt (ex-leases) $126 million.
• Final judgment: Watch only. It is the only one of the three consumer prints with no one-time contamination, and its fundamental score (59) is above GAP's (45), yet it is the only one falling. There is no contradiction — what the market is buying today is the rate of change, not quality.

⑦ WDAY — Workday | The only one of the four whose fundamentals genuinely deteriorated, and the only one that is green

In the draft I wrote this one up as "the negative is being digested / watch only." After primary-source verification the characterization has to be reversed: it is not "the negative is digested," it is "the negative is not yet priced."

• Three independent primary-source lines of evidence, all pointing the same way:

1. Subscription revenue growth has decelerated monotonically for five consecutive quarters, and the slope is steepening — from about 16% down to about 10%. The Q4 implied figure computed off the two ends of full-year guidance is +10.17% ~ +10.59%, a range only $10 million wide; there is no escaping it.
2. Total subscription backlog growth fell 4.2pp over two quarters to +8.0%, now clearly below current revenue growth of 13.9% — this is a bookings-side problem rather than a revenue-recognition one, and it is the most leading of the three. Moreover the FY26 backlog base still includes the Paradox and Sana acquisitions, so the organic figure is weaker still.
3. H1 free cash flow was only +6.6% while H1 revenue was +13.1%; Q2 capex was $59 million, +111% YoY. This press release gives no FY27 full-year free-cash-flow guidance at all.

• It is the only one of the four to cut its implied second half: Q2 subscription revenue beat its own guidance by $16 million, while the midpoint of full-year guidance was raised by only $7.5 million — only 47% of the quarter's beat flowed through to the full year; the other 53% was taken out of the second half. (For comparison: RBRK 147%, ADSK 375%, S 227%.)
• ⚠️ Basis trap: GAAP EPS of $2.57 includes a $1.52/share one-time tax benefit (an intra-group IP transfer as part of an internal legal-entity reorganization), and GAAP net income being +177% YoY is entirely non-comparable. Any YoY computed off GAAP net income is wrong.
• The revenue contribution from AI agents has not been quantified: the press release offers only "AI contributed over 25% of new ACV" and "more than 5,500 customers are using at least one in-house agent (+35% QoQ)" — a ratio (with the denominator undisclosed) and a customer count, with no dollar amount or ARR whatsoever.
• In fairness, the other side: non-GAAP operating margin of 31.1% (+2.1pp), with full-year guidance raised from 30.5% to 31.0%; a forward P/E of 17.9x, the cheapest of the four. But the margin improvement comes from efficiency and buybacks (Q2 repurchased 9.80 million shares / $1.3 billion, with a new $4 billion authorization), which changes the per-share basis, not the growth trajectory. Cash and marketable securities are −58.4% YoY, leaving net cash of only +$414 million.
• Pre-market: from −0.29% at 08:07 all the way to +1.22% ($195.94) at 08:28, notional about $20.55 million — the best pre-market liquidity of the four software earnings names, so this move into the green has volume behind it and is not thin-quote drift.
• ⚠️ The counter-explanation (weighted equally with this judgment; read them together):

  Five consecutive quarters of decelerating subscription revenue has been publicly visible for over a year; it is not new information. And this piece itself writes that WDAY's forward P/E of 17.9x is the lowest of the four — the cheapest valuation is itself direct evidence of "already priced in." This is precisely the mirror image of "pre-disclosed ≠ repriced": a long-visible deceleration is more likely already repriced than unpriced. So the only thing this piece can claim is one narrow sentence: "total subscription backlog growth falling to +8.0% and dropping below current revenue growth for the first time is new information this quarter, and whether it is already in the price is uncertain." This piece cannot claim "the entire deceleration trajectory is unpriced."

• Final judgment: A short-thesis watch item (not a trading recommendation). It is the only name today that is "fundamentally deteriorating with a complete evidence chain, moving the opposite way in price, and lacking any alternative explanation such as position, thin volume or sandbagging" — which makes it worth watching, but does not make it a trade. The 17.9x valuation, the margin guidance raised to 31.0% and the $4 billion buyback authorization are all real cushions; a decelerating stock is not the same as a stock that will fall. Verification point in §9①.

⑧ RBRK / ADSK / S — three names whose reasons for selling do not survive primary-source checking

RBRK (−6.28%, notional $12.89 million) — the purest sell-the-news

• Q2 non-GAAP EPS $0.20 versus the company's own Q2 guidance of just $0.03–0.05 (the so-called "consensus of $0.04" is precisely the midpoint of that guidance — consensus was simply copying); revenue $427.26 million (+37.9%); subscription ARR $1.6609 billion (+33%, up rather than down from Q1's +32%).
• All four FY27 guidance lines were raised: revenue +2.80%, EPS +66.7%, free cash flow +10.1%, ARR +1.3%. And the size of the raise ($46 million) exceeds the size of the quarter's beat ($31 million) — the second half alone was raised by about $14.70 million.
• There is no "non-recurring" add-back anywhere in the non-GAAP EPS reconciliation; the $0.20 is clean.
• It went into the print up +11.33% on the day, closing at the 98.6th percentile of its 52-week range; four sell-side firms have already raised price targets this morning (GS $106→120, Stephens $90→120, BTIG $109→125, KeyBanc $112→120).
• ⚠️ But this piece cannot calibrate the information content of "all four lines raised," so it is not used as an argument: RBRK's own Q2 guidance was non-GAAP EPS of $0.03–0.05 and it printed $0.20 — sandbagging itself by 4–5x. For a company that habitually guides far too conservatively, a $46 million raise may just be putting back what was deliberately held down, rather than a genuine upward revision of expectations. This piece was unable to obtain its "own guidance vs actual" deviations over the past four quarters, so: the "all four lines raised" point is used as a statement of fact only, not as evidence that "the market is wrong."
• Two genuine blemishes (neither of which is new information from this print): non-GAAP gross margin has declined two quarters running, 83.7% → 82.9% → 81.0%; SBC is 23.6% of revenue, 2.26x non-GAAP net income; and there is a 19pp gap between "adjusted net new Cloud ARR of only +20%" and installed Cloud ARR of +39%.
• Judgment: watch only. Shorting means betting against a company that raised all four guidance lines; buying means catching a stock that has just come off the 98.6th percentile of its 52-week range. Neither side has an edge.

ADSK (−4.65%, notional only $2.26 million) — the draft's attribution for this section is refuted and must be corrected

In the draft I wrote that "ADSK and MRVL are the same story: revenue guidance raised + cash-flow guidance cut." After primary-source checking, that analogy does not hold.

• What the FCF guidance actually did: old $2.725–2.800 billion (the 8-K explicitly notes "Does not include MaintainX") → new $2.725–2.750 billion (noted "Includes MaintainX"). The low end did not move at all; only the top was trimmed, midpoint −0.90%. And the new guidance explicitly includes roughly $45 million of MaintainX transaction costs — strip that out and the underlying cash-flow guidance actually rose by about $20 million.
• The widely cited "real estate–related spending" is a misreading: "real estate" appears 0 times in the entire EX-99.1; the real-estate reference on the call is a positive customer anecdote (a large North American retailer using Forma + Tandem to build a digital twin of its real-estate portfolio). A positive anecdote got spliced onto a different data point.
• The MaintainX consideration was about $3.575 billion all cash, announced 5/28 and closed 8/3 — fully disclosed three months before the print; it is pre-disclosed information.
• But ADSK does have one real problem that the media has essentially not mentioned: billings only +10% against revenue +16%; total RPO only +2%; unbilled deferred revenue −8%. The company attributes this to deliberately reducing multi-year discounting (an intentional action), but on the books, growth in forward bookings has gone to nearly zero. And Q3 guidance is +15.1% with Q4 implied at +12.8%, and those are the numbers after MaintainX is consolidated.
• ⚠️ Volume warning: 8,750 shares / $2.26 million for a company with a $54.4 billion market cap is only enough to confirm direction, not magnitude — within the few minutes I was pulling quotes it drifted from −4.02% to −4.65%.
• Judgment: watch only (no longer a short thesis). The surface reason for selling does not hold up, while the real problem (bookings) needs the 10-Q to separate organic from acquired before it can be characterized.
• ⚠️ Open item: of the $135 million raise to FY27 revenue guidance, how much is organic and how much is MaintainX consolidation is not disclosed in the 8-K. If the consolidation accounts for nearly all of it, then organic guidance was in fact raised by zero or even cut, and ADSK's characterization would have to change from "accounting noise" to "organic deceleration masked by an acquisition."

S (−2.69%, notional $3.50 million) — the cut is real, but both its location and its attribution have been mis-relayed

• The so-called "below consensus of $0.35" — $0.35 is exactly the midpoint of the company's own old guidance. So this is not "missing consensus," it is the company cutting itself.
• But operating income guidance was raised. Backing it out shows the EPS cut comes almost entirely from an approximately $19 million downward revision to non-operating (interest/investment) income assumptions plus 11 million shares of dilution, with nothing to do with operations. The specific trigger is not explained in the press release and cannot be confirmed from a primary source.
• "Pricing pressure" has no primary-source support: the phrase appears 0 times in EX-99.1; the 10-Q filed today attributes the gross-margin decline to "higher cost of revenue from scaling operations to meet growing customer demand," and lists $12.30 million of customer support costs + $7.90 million of acquired intangible amortization + $3.10 million of capitalized software amortization. The causal chain "pricing pressure from CRWD/PANW" could not be confirmed in any primary source by this piece.
• Gross margin has in fact stabilized: GAAP 72% → 72%, non-GAAP 77% → 77%, flat sequentially for two straight quarters (down 2–3pp YoY). This is the shape of "a step-down happened a year ago and has since stabilized," not ongoing deterioration. The revenue trajectory is +20.8% → +20.6% → guidance +19.7% → implied +20.2%, the only one of the four holding a plateau around 20%.
• The negatives that actually deserve attention are elsewhere: the growth rate of $100K+ customers fell from +17% to +13% in a single quarter; total deferred revenue is up only +5.6% YoY against ARR +22% (collection duration is shortening, and this kind of divergence usually leads ARR deceleration); Q2 operating cash flow turned negative at −$6.55 million, and the company deleted the entire "cash flow margin" line from the Q1 press release in Q2. There is also a $180 million settlement with the Israeli tax authority, with a remaining liability of $162.3 million as of 7/31 accruing interest at 7.0%.
• The sell-side stance is the worst of the four: the average price target is $22.06, below the 8/27 close of $22.71; August saw the first Sell rating appear.
• Judgment: avoid. The reason not to recommend shorting is position — it was +10.73% yesterday and is giving back only 2.69% today, on a notional of just $3.50 million.

⑨ IREN | The revenue basis conflicts, but the loss and the "old news as new news" are certain

• Primary source (8-K): FY26Q4 revenue $137.2 million; full-year FY26 net loss of about $703 million (including the exit from mining); AI Cloud $70.50 million for the quarter; AI Cloud Services $128.8 million for the year vs $16.40 million a year ago; $4 billion of contracted ARR, $6.5 billion of GPU financing.
• ⚠️ Conflicting consensus definitions; this piece declines to characterize it: two mutually exclusive consensus revenue figures were found — $132.29 million (pre-print preview) and $157.14 million (post-print reporting). The actual $137.2 million is simultaneously above the former and below the latter. Until it can be confirmed which one is the latest pre-print consensus, any "beat/miss" characterization would be fabrication. This piece states only: the market voted with −5.67%.
• ⚠️ The "$4 billion ARR" is old news: the figure had already appeared in media preview articles before the print. It is a cumulative contracted figure, not this quarter's new additions, and does not constitute an expectation gap.
• Pre-market: −5.67% to $38.23, notional about $101 million; 52-week range $22.04–$76.87.
• Final judgment: Avoid / short watch. But the $6.5 billion GPU financing and Microsoft's acceptance of Horizon 1 are genuine business progress; this is not a pure short story.

⑩ Semiconductor breadth — what really deserves attention today is not any single name

Semiconductor names with pre-market notional ≥$2 million (the same threshold applied to both sides; readings 08:01–08:08 ET):

13 down, 1 up. The only one green is AVGO — and AVGO happens to be the largest player in custom ASICs. If MRVL's "ASICs dilute gross margin" were an industry-wide problem, AVGO should not be the strongest name.

⚠️ Two pieces of self-restraint: ① All 15 names in this table use the 08:01–08:08 snapshot; no clocks are mixed. AVGO's readings at 08:19/08:28 (+0.36% / +0.38%) are in §1.6, but that 26bp of change is within the pre-market bid-ask spread and does not constitute evidence of "strengthening," so it is neither written into this table nor used as an argument. ② This table is not a full-market scan; it is this piece's self-selected semiconductor watch pool (15 names left after filtering for pre-market notional ≥$2 million). TXN / ADI / ON / MCHP / NXPI were excluded by the same threshold for insufficient notional (their pre-market readings are also all lower, so the direction does not change the conclusion, but they are not counted). "13 down, 1 up" is a count within this pool and cannot be quoted as an industry breadth statistic.

The moment this falsifier actually starts producing readings is 09:30, not now.

---

6. Negative / Avoid List

⚠️ False signal: every "biggest pre-market decliners" list today will put VISN at the top

VISN (Vistance Networks) is −41.62% pre-market to $6.79, market cap about $2.68 billion — this is not a decline, it is an ex-distribution adjustment.

The board declared a $5.00/share special cash distribution (paid 8/27, record date 8/17); because the distribution is ≥25% of the share price, under Nasdaq Rule 11140(b)(2) the ex-date is set to the trading day after the payment date, i.e. 8/28. The funding came from the cash consideration of the 7/1 sale of Ruckus Networks to Belden.

After the ex-distribution adjustment: $11.63 − $5.00 = $6.63, so the pre-market $6.79 is actually up about 2.4%.

Any list generated automatically by "pre-market decline" will treat it as a crash today. It is recorded here so that it stays out of everyone's short candidate list.

---

7. Intra-Theme Rankings

Theme 2: AI hardware · the gross-margin bar (direction: bearish)

A detail that is easy to get backwards: what MRVL said is that optical interconnect and standard products carry higher gross margins than custom ASICs. So MRVL's gross-margin problem should not be read as a negative for optical modules. This morning's declines in LITE (−2.39%), COHR (−2.30%) and AAOI (−2.57%) are more likely semiconductor beta than transmission of this logic.

Theme 4: software (direction: divergent, with an identifiable mispricing)

This table is sorted descending by "guidance quality," with this morning's prices alongside. Note that the prices of #1 and #5 are the inverse of their ranks.

Theme 5: consumer · three kinds of "illusion" (direction: divergent)

The ordering of these three rows is exactly the inverse of today's price direction (AFRM +13%, ULTA −1.2%, GAP +18%). This is not a scoring error; it is a reminder: what the market is buying today is "the rate of change versus yesterday," not "quality versus peers."

Theme 1: macro · Warsh (direction: TBD)

---

8. Opening Verification Signals

① Pre-market (now to 09:30)

• Whether MRVL can hold $223.67 (the snapshot ③ pre-market price). A notional of $489 million means this price has real transacted support behind it, not thin quotes. A break below $220 → gross margin is being treated as a structural problem; a recovery above $230 → sentiment overshot.
• Whether PYPL can hold $52.62 (the snapshot ③ pre-market price). The decline has narrowed from −15.54% and stabilized at −14.4%. Losing it and moving toward $47–48 = pricing a "full return to pre-deal"; holding above $53 = the market is giving credit for the fundamental improvement since July.
• Whether ESTC is gap-and-go or gap-fill. The current price of $102.36 (08:27) is above every pre-print price target ($91–$94) and already above the 52-week high of $96.065 — there is no reference point above; the first reference below is $96.065.

② The first 30 minutes — the most important verification in this piece

Yesterday's broad software repricing among the names that have not reported (CRM, NOW, OKTA, CRWD, PANW, ZS, NET) is almost entirely within −0.9% ~ +0.7% pre-market, with IGV at only −0.10%.

The criterion: if IGV still does not fall relative to SPY in the first 30 minutes, yesterday's software repricing had real buying behind it and the declines in the four earnings names are single-stock events; if IGV weakens along with ADSK/S/RBRK, then yesterday's +7.74% was a squeeze with no foundation. The two cases have completely opposite implications for next week.

And one addition today, aimed at the §5⑦ judgment that "WDAY's price is moving opposite to its fundamentals":

The criterion is asymmetric: it can only falsify, never confirm:

• If RBRK recovers over the full day and WDAY weakens → the market read the guidance correctly on the day, and this piece's judgment is void on the spot.
• If WDAY outperforms RBRK → that is not confirmation. The prior on single-day relative strength is close to 50/50, and both names are in the high-noise state of the day after earnings. Confirmation would require 3 consecutive trading days in the same direction plus evidence that the sell side is starting to restate models on backlog rather than subscription revenue.

The other two:

• Semiconductor breadth: SMH's strength relative to QQQ; watch AVGO in particular (already strengthened from +0.12% to +0.36% pre-market) — if AVGO turns from up to down and follows MRVL, theme 2 is upgraded to industry-wide; if it stays relatively strong all day, MRVL is a single-stock event.
• Whether the breadth repair persists: all 14 blue chips that fell yesterday are green pre-market. If DIA / IWM strengthen relative to QQQ after the open, the breadth repair is real; if QQQ overtakes them right at the open, it was just low-liquidity pre-market drift.

③ 09:45 and 10:00 — today's real watershed

• 09:45 Chicago PMI — what is released today is the August figure. The prior is July's 57.6 (published 7/31, above the roughly 55–56 expected at the time); 56.7 is the June figure — do not use it as the baseline. (TradingView/MNI²³ · CNBC²⁴)
• 10:00, two events collide: the UMich final (August preliminary 51.0, prior 55.2) and Warsh's first Jackson Hole keynote.

  A genuine "vacuum risk": this year's Jackson Hole theme is "Financial Innovation: Implications for Payments and Policy", not monetary policy. Sell-side economists have already explicitly said Warsh may give no rate signal at all. If so, positions built pre-market on the assumption that "there will be guidance" lose their rationale collectively around 10:05 — this kind of "non-event event" is often harder to trade than either a hawkish or a dovish outcome.

④ Options and volatility

• VIX 14.48 (−0.21%), vs yesterday's close of 14.51. The factual statement stops there: this piece did not obtain any implied volatility levels, term structure or realized volatility, so it makes no qualitative judgment on whether "VIX is cheap or expensive." A reader wanting to judge that would need to compare VIX against trailing 20-day realized volatility themselves.
• MRVL / ESTC / RBRK / ADSK / S / WDAY / AFRM / GAP / ULTA have all passed their earnings events, and historically such names generally see implied volatility fall quickly the day after (IV crush). This piece did not obtain actual IV data for these names, so it only flags this common phenomenon and makes no pricing judgment.

⑤ State-change observation items (if they appear, the premise has changed and needs re-evaluation)

• Reversal after the gap: if ESTC falls back below $96 (the prior 52-week high) within the first 15 minutes, the +22% has no absorption behind it.
• A lone riser: GAP rising while the rest of apparel retail (ANF +0.59%, LULU +0.26%) does not follow means this is a single-stock event, not a consumer-sector recovery. Note that this only changes "how to interpret GAP's rise"; it is not a bearish signal on GAP — §6 has already stated explicitly that shorting GAP is not recommended.
• Divergence: if SPY rises while the 30Y yield simultaneously breaks above 5.25%, that is the most dangerous combination — equities ignoring rates, and yesterday already staged one round of "index up, ten sectors down."
• Waiting ahead of the event: if volume between 09:00 and 10:00 is clearly below normal, everyone is waiting for 10:00 and prices carry no information at that point.

---

9. Final Conclusions

① The 6 stocks most worth watching today

Uniform verification basis: each name is given three numbers — "prior close / pre-market price (snapshot ③, 08:24–08:29 ET) / criterion line" — so that everything can be reconciled with the same ruler after the close. Note that the nature of the criterion line differs by row (pre-market price / 52-week high / prior close), as labeled.

② The 3 strongest themes today

③ What to avoid today, and why

1. The "first chase order" in any name that reported last night — in either direction. IV crush has already happened, the gaps are complete, and the two largest by notional (MRVL $461 million, PYPL $240 million) are already down 7.9% and 14.3% respectively, with both declines narrowing.
2. "Beats" propped up by one-time items — GAP's $417 million tariff refund and AFRM's $1.465 billion tax reversal together turn both headlines into figures that cannot be used directly. Once stripped, GAP shows net sales −2.0% and adjusted operating income −11.3%; once stripped, AFRM still beats. Same act of "stripping," opposite conclusions — which is exactly why you have to strip item by item rather than doubt everything uniformly.
3. The four memory names (MU / SNDK / WDC / STX) — yesterday's thesis was falsified on the spot by its own criteria, and today they are being sold with volume on no new news whatsoever. A decline with no news can neither be chased short nor bought as a dip.
4. Pure beta vehicles (INTC / SMCI / optical modules / neoclouds) — on a day with a certain event at 10:00, holding names that only amplify direction without providing independent information is handing your position to the podium.
5. Thin-volume gappers (ADSK $2.26 million / S $3.50 million / ULTA $7.13 million) — the direction is readable, the magnitude is not credible, and they will be repriced at the open.
6. The instinct that "whatever fell the most should be shorted the most" — in software today that instinct is systematically wrong: the biggest decliner RBRK raised all four guidance lines, the second-biggest decliner ADSK's reason for selling does not survive primary-source checking, and the only genuine deterioration, WDAY, is up.
7. VISN — the −41.62% is the ex-date of a $5.00/share special distribution, not a decline; adjusted for it, the stock is actually up about 2.4%.

④ The final one-line judgment

Today there are three leads nailed down by primary sources, one orthogonality reminder, and one moment that can overturn all of them.

Lead one (semiconductors): a revenue beat is no longer worth anything; the direction of gross margin is. MRVL's Q3 revenue guidance is above consensus and its EPS guidance barely beats; what is missing is those 90bp — and that layer is confirmed by the SEC 8-K primary source. The more serious layer is that management reportedly extended the gross-margin plateau all the way to FY2028, but that layer appears only in secondary transcripts, this piece did not obtain a verbatim transcript, its evidence grade is one notch lower, and it should not be treated as equal to the Q3 guidance. Moreover the stock has already been halved once from its June high of $329.88 and now sits at only about the 60th percentile of its 52-week range ($61.44–$329.88) — this is not a sell-the-news top at a high, so it should not be chased short at the pre-market price either.

Lead two (payments): PYPL's M&A premium collapse is clean and unambiguous, it is already down 14.3% and narrowing, and 11.2% of the premium has yet to be given back.

Lead three (software) is the most valuable today, and also the one that has to be stated most carefully: after comparing all five companies' new guidance against their own old guidance line by line, this morning's ranking of declines is almost exactly the inverse of the ranking of guidance quality. The biggest decliner, RBRK, raised all four guidance lines and the raise exceeds the quarter's beat; ADSK's "cash-flow cut" is $45 million of acquisition costs with the low end of guidance untouched; S's EPS cut occurs below the operating line while operating income guidance over the same period was raised. But "inverse" does not mean "mispriced": RBRK entered the print at the 98.6th percentile of its 52-week range (position explains it), ADSK traded only 8,750 shares (thin volume explains it), S was +10.73% yesterday (profit-taking explains it); worse still, RBRK's own Q2 guidance sandbagged itself by 4–5x ($0.03–0.05 vs an actual $0.20), and for a company that habitually sandbags, the information content of "raised across the board" may be close to zero. So this piece narrows the claim to one stock: WDAY — the only one to cut its implied second half, with total backlog growth down to +8.0% (already below current revenue growth) and no alternative explanation of any kind — went from −0.29% to +1.22% this morning. That is a singular claim, not the universal claim "the sector is mispriced." And even for this one name the counter-explanation holds: five quarters of deceleration is old information that has been publicly visible for more than a year, and the lowest valuation of the four at 17.9x may itself be evidence that it is already priced.

An orthogonality reminder: after stripping one-time items, the quality ranking of the three consumer prints is AFRM > ULTA > GAP, while today's price ranking is exactly the reverse, GAP > AFRM > ULTA. Fundamental quality and same-day direction are two independent questions; do not use the former to trade the latter — CRM already taught that lesson yesterday with +22.58%.

And Warsh's debut at 10:00 ET will reshuffle all of it, with particular attention to the fact that this year's Jackson Hole theme is "Financial Innovation: Payments and Policy" rather than monetary policy — "he said nothing at all" is a real and far-from-negligible outcome.

This piece gives no position advice. What it can offer is five criteria that settle at today's close — MRVL vs $223.67, PYPL vs $52.62, ESTC vs $96.065, WDAY vs $193.57, AVGO relative to SMH — plus one reminder: not one of those five can be read out before 10:00 ET.

---

10. Data Sourcing and Verification Status (internal)

Quote pipeline

• The pre-market trio for individual stocks (pre-market price / percent change / cumulative volume) all comes from stockanalysis.com api/quotes/s/<TICKER>/, fields ep/ecp/epv/eu, with the eu timestamp recorded for every entry. Two snapshots were taken (08:01–08:08, 08:17–08:19) and the comparison table is in §1.6 — this is the direct response to "pre-market quotes are cumulative and a single-point snapshot is not a fact."
• Per the memory entry "CNBC extended-hours fields are unusable pre-market," the pre-market slot did not use CNBC's ep/ecp fields; CNBC restQuote was used only for rates, VIX, crude, gold, DXY and the Asian/European indices (non-extended-hours fields, with last_time checked).
• CNBC futures symbols @ES.1 / @NQ.1 / @YM.1 / @RTY.1 returned empty on all four attempts. This piece has no direct index futures quotes; §0 and §1.5 use SPY/QQQ/DIA/IWM pre-market quotes as substitutes, and the text is written on an ETF basis throughout (ETF percentage changes are never described as "futures").
• PYPL daily history (7/1–8/27) was scraped and parsed from the stockanalysis history page, used to back out the $47.37 starting point of the M&A premium.

Sub-agent status

• Three fundamentals-analyst runs: (a) MRVL deep dive, (b) AFRM/GAP/ULTA, (c) a cross-comparison of the four software earnings names. All three returned, and their conclusions have been fully incorporated into the text.
• All three agents reported that yfinance was rate-limited by Yahoo (YFRateLimitError / HTTP 429), failing on the very first call; all three degraded per their preset instructions to SEC EDGAR (primary) + stockanalysis and obtained the data successfully. Affirm's IR site returned 000 on curl (domain unreachable), so the image version of the shareholder letter on EDGAR was used instead. Marvell IR's quarterly-earnings page returns HTTP 404; the IR sites of the four software companies returned 403 to curl, so no earnings call transcripts were obtained at all, and any statement that exists only on a call has been labeled in the text as a secondary transcript or "cannot be confirmed from a primary source."

Five draft errors corrected by agents this round (important, recorded for the file)

1. I wrote in the draft that ULTA "maintained guidance" — wrong. It raised across the board (EPS $28.36–28.80 → $28.70–29.00), and the press release headline is "AND RAISES FISCAL 2026 GUIDANCE." The draft was based on second-hand media wording, which the primary source refuted directly.
2. GAP was labeled "watch closely / bullish" in the draft — downgraded to "watch only" (score 62→45). Primary source: GAAP EPS of $1.38 includes a $417 million tariff refund; net sales −2.0%, adjusted operating income −11.3%; recomputing with the company's parameters shows the $0.05 guidance raise is 100% attributable to the share count going 375M→367M; implied H2 adjusted EPS is −10.4%~−0.9% YoY.
3. AFRM was "unscored" in the draft — now scoreable (67). The one-time amount was pinned precisely from the 10-K ($1.4648 billion, 90% of EPS), and after stripping it operating EPS of $0.43–0.48 still beats consensus by 25%–45%, with RLTC +39% faster than GMV +36%. The draft's cautious direction was right, but stopping at "unscored" would have missed a genuine bullish conclusion.
4. ⚠️ The most serious error in the draft: I drew an analogy between ADSK and MRVL as the same story of "revenue guidance raised + cash-flow guidance cut," and made it a pillar of the second theme — that analogy does not hold. The primary source shows ADSK's FCF guidance low end did not move at all (old $2.725–2.800 billion noted "Does not include MaintainX" → new $2.725–2.750 billion noted "Includes MaintainX"), only the top was trimmed; and the new guidance includes roughly $45 million of MaintainX transaction costs, so stripped out the underlying guidance actually rose by about $20 million. The widely cited "real-estate related spending" is also a misreading — the phrase appears 0 times in EX-99.1, and the real-estate reference on the call is a positive customer anecdote. §5⑧ has been fully rewritten and the text explicitly states that this is a correction.
5. WDAY was labeled "the negative is being digested / watch only" in the draft — changed to "short watch." Three independent primary-source lines all point the same way: subscription revenue decelerating monotonically for five quarters (15.69%→14.33%→13.92%→12.08% guidance→10.38% implied); total subscription backlog +12.2%→+10.9%→+8.0%, now below current revenue growth; H1 FCF only +6.6% versus revenue +13.1%. And it is the only one of the four to cut its implied H2 (only 47% of the quarter's beat flowed to the full year). And this morning it went from −0.29% to +1.22% — price moving opposite to fundamentals.

The resulting third theme-level revision: the draft's software theme was "the direction of margin inside the guidance decides life or death"; after three-way primary-source comparison it was upgraded to "the market prices the apparent direction of guidance, not the reason behind it," and it identified one nameable, same-day-settleable mispricing (criterion: the same-day relative strength of WDAY vs RBRK, see §8②). The key methodological change: the criterion was switched from "vs consensus" to "vs the company's own prior guidance" — because RBRK's third-party consensus equals its old guidance midpoints line by line (consensus is merely an echo of guidance) while ADSK's consensus has already been revised up to equal the new guidance midpoint, and mixing the two produces conclusions pointing in opposite directions.

Unresolved basis conflicts (disclosed in the text, not characterized)

1. AFRM consensus EPS: $0.33 (Zacks) / $0.35 vs $0.85. This piece uses $0.33–0.35. If the true consensus is $0.85, the conclusion of §5⑤ reverses entirely — the dependency is explicitly registered in the text. This is the most fragile link in the whole piece.
2. IREN Q4 consensus revenue: $132.29 million (pre-print preview) vs $157.14 million (post-print reporting), with the actual $137.2 million in between. No "beat/miss" characterization given.
3. ESTC consensus revenue: $469.75 million vs $479.1 million. This piece uses $469.75 million (same source as the EPS consensus of $0.58). Using the other version would make revenue a slight miss. The text states only the actual figure of $478.1 million.
4. MRVL consensus: third-party sites are labeled "Last updated Aug 28, 2026" but the FY27 $11.56B / FY28 $16.88B they display are clearly still pre-print stale values (company guidance is already above the top of their ranges). The text therefore avoids phrasing like "guidance beats consensus by X%" and uses only the absolute values of company guidance and self-consistent back-outs.
5. September FOMC probabilities: three mutually exclusive versions were found — "66% hold / 34% hike," "58.6% hold," and "about 60% hike." The text cites no specific probability, stating only verifiable facts: the 7/29 meeting held at 3.50–3.75% by a 9–3 vote, plus inflation of 3.4% (media-sourced; this piece has not verified which measure or which month) and the 30Y at 5.21%.
6. ULTA fiscal-year labeling: the company calls it "fiscal 2026" (ending 2027-01-30) while third parties label it "FY2027." The text uses the company's convention throughout and notes this.
7. The accounting treatment of MRVL's Google warrant is unknown: if it is netted against revenue as "consideration payable to a customer," its scale could explain a substantial part of those 90bp, whereas management attributes everything to mix. The agent ran a reverse check (there is no place for the warrant in the Q3 GAAP→non-GAAP gross margin bridge, and Q3 SBC guidance is below the Q2 actual) and leans toward "it is not in the Q3 guidance," but this cannot be confirmed and requires the 10-Q (expected in early September). The text states only the primary-source terms of the warrant and the dilution ratio, and does not write this hypothesis up as a conclusion.

Traps actively excluded

• DELL did not report last night. DELL appeared among after-hours movers in the initial search (pre-market −1.69%); verification confirmed Dell's FY27Q2 call is scheduled for 2026-09-01, and it has been removed from the earnings list entirely. Without verification this would have manufactured an earnings event that does not exist.
• VISN −41.62% was verified as the ex-date of a $5.00/share special distribution (Nasdaq Rule 11140(b)(2)) and is listed separately as a false signal in §6.
• The "Thursday's Top Wall Street Analyst Calls" column: headline and URL structure are near-identical day to day, and the search returned the ratings from 8/27 (Thursday). These ratings were not treated as today's catalysts; the price targets cited in the ESTC section all carry explicit dates (Cantor 8/24, Barclays, JPM) and are explicitly labeled as published before the print.
• The first search result for ESTC claimed "revenue below consensus, a loss of $0.16 per share," conflicting with non-GAAP EPS of $0.70. This was identified as mixing GAAP and non-GAAP, and the text labels both bases separately.
• IREN's $4 billion ARR had already appeared in pre-print preview articles and has been handled under "cumulative-period order figures are old news," not counted as an expectation gap.
• MRVL's "YTD +187%": the agent verified with daily bars, 2025-12-31 close $84.98 → 8/27 $241.45 = +184.1%; the media figure is broadly accurate (the difference comes from the quote timing), and the text uses it as background only, not as an argument.

Handling of pre-market volume: everything is converted to notional, with the same threshold ($2 million) applied to both the long and short sides. ADSK ($2.26 million) clears the line but is extremely thin relative to its market cap, and is separately labeled "direction credible, magnitude not."

---

10.1 risk-auditor QA results and disposition (run 08:35–08:43 ET, recorded item by item)

The auditor's verdict was "not publishable as it stands," with 16 red flags. Below is the item-by-item disposition:

The two items not dispositioned, and why:

• F9 (GAP's implied H2 EPS should be cross-checked by summing the actual Q1 figure rather than backed out from the full-year GAAP-minus-adjusted difference) — I agree with this one, but there was no time before publication to obtain GAP's Q1 earnings data for the summation. The auditor pointed out that the back-out implies Q1'26 adjusted EPS ≈ $0.88 (a +76% move against Q1'25's $0.50), forming an abnormal swing against Q2's −8.8% and suggesting the bridging assumption may be missing items (restructuring/impairment/Athleta charges). §5④ already states the assumption behind that conclusion in the text, but the sentence "not cross-checked against Q1 data" has not yet been added to the body — this is the largest known open loop at publication time and must be closed in the recap.
• Whether AFRM's $0.85 consensus is on a different period basis (full fiscal year vs the single quarter FQ4) — the auditor suggested spending 5 minutes confirming it, which would most likely eliminate this uncertainty outright. Not completed before publication; retained as an explicit dependency.

Parts the auditor explicitly endorsed and that need no change (recorded to prevent later mis-edits): the basis-conflict register and the five self-reported corrections in §10; the item-by-item removal of one-time items (AFRM $1.465 billion / GAP $417 million / WDAY $1.52 per share); the methodology itself of switching the criterion from "vs consensus" to "vs the company's own old guidance"; the interception of the VISN ex-distribution false signal; the four trap interceptions (DELL / analyst-column dating / ESTC GAAP mixing / IREN cumulative ARR); the symmetric application of the notional threshold to both sides; and §7's correction of the reverse transmission, i.e. "MRVL's gross-margin problem should not be read as a negative for optical modules."

The auditor's overall verdict is worth leaving here verbatim: "The strongest part of this piece (the basis-conflict register and the five self-reported corrections in §10) and the weakest part (§0/§9 writing those uncertainties up as settled values) are two presentations of the same material. The problem is not that the research was insufficient; it is that the honesty of the inside pages did not carry through to the only two pages the client will actually read." — most of this round's revision work went into that one sentence.

---

⚠️ Risk disclaimer: this list is a pre-market information review and watchlist only and does not constitute investment advice. U.S. equities carry high volatility and pre-market gap risk, and post-earnings IV crush and guidance reversals are possible; automatically generated content may contain stale information or factual errors. Company disclosures and SEC filings prevail, and this piece must not be used directly as a basis for trading.