U.S. Pre-Market Brief | 2026-08-13 (ET) Thursday

Coverage window: 2026-08-12 16:00 ET regular-session close → 2026-08-13 08:15 ET (includes 8/12 after-hours earnings, the overnight Asia/Europe sessions, this morning's pre-market, and earnings released pre-market). Quote conventions:

• Previous close / prior-day change / pre-market price / pre-market change / pre-market volume / 52-week high-low for single stocks and ETFs come from the stockanalysis.com quote API, snapshots 07:33–08:11 ET (labeled per ticker). "Previous close" is the 8/12 16:00 ET regular-session close; "prior-day %" is the 8/12 regular-session change for that day; "today's pre-market %" is the pre-market change relative to the 8/12 close. These two columns are two different things from two different trading days; this brief lists them separately throughout and never mixes them.
• Treasury yields, VIX, crude, gold, the dollar index and bitcoin come from the CNBC quote API, snapshots 08:03–08:05 ET.
• Earnings figures are taken first from SEC EDGAR primary documents and company press releases (each filing is cited); where the primary document could not be obtained, authoritative media reprints are used and marked "media-sourced". What this brief verified: every pre-market mover named by the media this morning was re-checked one by one against the quote API to establish "which trading day that move actually happened on". The re-check overturned two circulating claims (SMCI, MQ); see section 1.3 — this is not nitpicking, it decides whether you chase a move today that was already over yesterday. Data this brief did not obtain: ① July PPI was released at 08:30 ET, later than this brief's main-text snapshot time; the scenario mapping written before the release was set down at 08:15 ET, and the actual result plus a conventions trap have been added in section 9.0, with the two passages left in their original order; this brief only observed the "market reaction" to PPI for roughly 20–60 seconds after the release, and section 9.0 states explicitly that this window is not enough to represent pricing; ② option implied volatility (IV) was not obtained, so every judgment about IV crush in this text is inferred backwards from price action and historical earnings reactions, not from options data — discount accordingly; ③ the FY27 sell-side consensus conventions for CSCO and COHR are in doubt: numbers were obtained for both (CSCO FY27 EPS $4.80, COHR FY27 EPS $8.37, both from the stockanalysis aggregate), but that source page was found to contain a base-year error (see section 5.1), so this brief does not treat "guidance vs consensus" as a settled conclusion, and every place it appears is tagged "to be verified"; the sole exception is COHR's forward PE percentile (42.5x / 100th percentile), which uses $8.37 — if that figure is wrong the percentile must be recomputed, and this is flagged here; ④ the difference between Cerebras' "core revenue" and "total revenue" conventions has been obtained first-hand (line by line from the 8-K reconciliation table, see section 5.3), but its customer early-release lockup terms and post-IPO normalized stock-compensation run rate are still not obtained, and are tagged to be verified; ⑤ index futures (@ES/@NQ) quotes were not obtained this time; the pre-market direction for the broad market is proxied by the pre-market prices of SPY/QQQ/IWM/DIA, as noted in section 1.2.

---

0. Today in One Sentence

First, let's tamp down a claim that is easy to overstate. The first draft of this brief said "today is yesterday's mirror image": in a 35-stock tech sample, 19 of the 24 that rose more than 1% yesterday were falling this morning, and all 6 that fell more than 1% yesterday were rising. That statistic did not pass quality control — it used an asymmetric threshold (yesterday required |>1%|, this morning only looked at the sign), which counted a great deal of noise as "reversal". Recomputed on symmetric conventions (see the full sample table in section 1.4):

The "mirror-image rotation" conclusion essentially does not hold under strict conventions. The six software names that fell more than 1% yesterday were indeed all green this morning, but the magnitude was +0.40% ~ +0.64%, and only CRM rose more than 1% (and it had its own catalyst: a JPMorgan initiation at Overweight). The correct statement is: at the sector level there was only a very mild give-back and rebound; today's genuinely large moves came almost entirely from names with their own earnings. This brief keeps the observation that "software is relatively stronger than semiconductors" (IGV pre-market +0.57% vs SMH −0.14%), but no longer calls it "a decisive turn in money flow".

So today's real main line is not rotation but earnings — and the dividing line is drawn on gross margin and cash flow, not on AI. Both of last night's reports "beat", and both stocks are down more than 6% this morning, but what they are falling on is not the same thing at all:

• Cisco (CSCO, pre-market −6.09%) has a problem that is in the numbers, and it is more than just gross margin. Revenue $17.3 billion (+18%), total product orders +35%, AI infrastructure orders of $9.3 billion cumulative in FY26 — all above the top end of guidance — but non-GAAP total gross margin has now fallen sequentially for four straight quarters (68.4→68.1→67.5→66.0→66.3), and product gross margin was 64.8% vs 67.5% a year ago (−270bp); Q1 FY27 guidance puts non-GAAP gross margin at 65%–66%, a midpoint of 65.5% that is another 80bp below the 66.3% just delivered. Less often mentioned is cash flow: FY26 revenue +12% and non-GAAP EPS +14%, yet free cash flow was −3.9%; FCF/net income fell from 87% to 74%, inventory rose +80% year over year, and days of inventory went from 53 days to 84 days. This is "trading gross margin and cash flow for AI revenue" written into the financial statements; the story itself has changed.
• Coherent (COHR, pre-market −6.68%) has a problem that is precisely not about this report's impact on the share price — because that impact is roughly zero. The key arithmetic: the 8/12 report came after the close, and the stock had already risen 8.24% that day ($328.57 → $355.64). This morning's pre-market $331.03 is still +0.7% against the 8/11 close of $328.57. In other words, this morning's −6.68% almost exactly gives back the bar bought a day early ahead of the print, and the net pricing impact of this report is close to zero, not negative. Any "good news is out" narrative has to clear that arithmetic first. The report itself was a double beat: revenue $2.0455 billion vs $1.99 billion expected, non-GAAP EPS $1.74 vs $1.61 expected; FQ1'27 guidance of EPS $1.85–2.05 has a midpoint 12% above the quarter just delivered; orders are booked out to calendar 2028.

The two look identical on the tape (both "sold off after a beat" by a bit more than 6%), but they are opposites in nature: CSCO is the market repricing fundamentals; COHR is merely flattening its own one-day head start. The former's decline may only be starting; the latter has not really fallen today. That is the single most valuable takeaway today — and the one the tape most easily fools you on.

But COHR does have one page of genuinely new negative news that almost nobody mentioned this morning: cash flow. FY26 free cash flow was −$1.023 billion and operating cash flow only $79.5 million, while GAAP net income was $805 million — every $1 of accounting profit turned into just $0.10 of operating cash; inventory rose +79.5% year over year (revenue only +33.8%), the diluted share count rose +30% year over year, and capex was 15.5% of revenue (27.2% in FQ4 alone). Note that this is the same disease as CSCO's +80% inventory and negative FCF. So today's real theme might better be put this way: the accounting profits of AI hardware are looking less and less like cash — gross margin is the first layer, working capital is the second.

The one exception group happens to prove the rule: the AI server trio DELL (yesterday +9.87%, this morning +2.99%), HPE (yesterday +8.11%, this morning +2.50%) and SMCI (yesterday +19.02%, this morning +1.09%) are the only group in the field that "rose yesterday and is still clearly rising today" (snapshot 08:33 ET). They have the worst gross margins in the field, yet they are up, because SMCI's FY27 revenue guidance ($65–72 billion vs FY26's $39.1 billion) made the revision to "volume" large enough to overwhelm gross margin. The market today is paying for "volume" and docking points for "gross margin" — whoever's volume is still being revised up gets an exemption; whoever's volume is already priced in and has only gross margin left gets hit. One reminder: DELL's pre-market $498.99 is running right up against its 52-week high of $485.70 (intraday basis) — but note that the 8/12 close of $484.50 was itself only 0.25% below that high, so this is not "a new high" but chasing near the high, driven by the spillover from someone else's guidance.

The driver type is "earnings + guidance", not macro — but today's first-order event lands at 08:30 ET: July PPI (expected +0.2% m/m and +4.9% y/y, prior −0.3%/+5.5%; core +0.3% m/m and +4.1% y/y) and initial jobless claims (202,000 expected, 199,000 prior). Note the gap between the two conventions: yesterday's CPI at 3.4% y/y counts as mild, while PPI is still expected at 4.9% y/y — price pressure on the production side is far greater than on the consumption side, and that gap is itself the macro version of gross-margin compression, the same thing happening at the single-stock level today, just at a different scale. Pricing for the September FOMC is still "roughly fifty-fifty between a hike and no change" (CME FedWatch, as of 8/13 pre-market; note this is a fast-moving variable — at the end of July it was about six in ten for a hike, so it moved more than 15 percentage points in ten days; always cite it with a timestamp), not a cut, which runs counter to the instincts of the past two years.

The only big cross-asset move is oil: WTI $81.55 (−2.07%), Brent $87.40 (−1.78%), ending a six-session winning streak. Two causes: reported progress in the talks to reopen Hormuz (the Oman–Iran consultations are said to have entered a late stage), plus an EIA report of a 17.4 million barrel jump in U.S. crude inventories last week (that figure and the phrasing "largest since early 2023" are both media-sourced; this brief did not verify them against the original EIA weekly report and tags them to be verified; also that weekly report was published 8/12 10:30 ET, making it information from yesterday's session rather than a fresh catalyst this morning — XLE still closed +0.16% yesterday). The 10-year Treasury is at 4.672% (−2.0bp) and VIX at 14.57 (+0.14%); risk appetite itself has not deteriorated.

---

1. News Overview

1.1 Event Table

1.2 Pre-Market Broad Market and Cross-Asset

ETFs (stockanalysis quotes, snapshots 07:48–08:01 ET; the two columns belong to two different trading days — do not read them together)

How to read this: all four broad-market ETFs are green pre-market but by tiny amounts (+0.02% ~ +0.29%), while at the sector level SMH/XLK went straight from yesterday's +2.08%/+1.49% to a stalled +0.06%/+0.01%, and IGV flipped from −0.81% to +0.50%. The index has not moved, but the hands underneath have changed — this is classic intra-market sector rotation rather than a change in risk appetite, and VIX at just 14.57 corroborates it.

Cross-asset (CNBC quotes, snapshots 08:03–08:05 ET)

The curve shifted slightly lower across the board with 2Y and 10Y moving together (−2.1bp / −2.0bp); this is not rising rate-cut expectations, it is wait-and-see buying ahead of PPI. VIX at 14.57 is at the low end, showing the market does not treat today's single-stock declines as a systemic event.

1.3 ⚠️ One circulating claim overturned by re-checking, plus one this brief got wrong itself and corrected

(A) Overturned: SMCI's "pre-market +19.02%". Several media pre-market roundups this morning mixed yesterday's (8/12) regular-session gain and this morning's pre-market gain into the same table. After re-checking each ticker split by trading day:

(B) What this brief itself got wrong: MQ (Marqeta). The first draft, based on a single 07:41 ET snapshot, said "MQ is only +0.77% pre-market this morning, so the media's +4% does not hold". That was wrong: the stock quoted $16.20 / +3.32% at 08:33 ET, and the media figure was broadly correct. The cause of the error was using a single snapshot taken before 08:00 to refute an entire pre-market session — pre-market prices and volumes both accumulate over time, and a single snapshot cannot support a conclusion about "the facts". The judgment has been withdrawn. (One thing to add: MQ's quote timestamp stops at 07:54 ET and its pre-market volume is only 4,945 shares, so the right verdict is "volume too thin to conclude", not "the media got it wrong".)

(C) On "pre-market volume too low to be meaningful": this brief has corrected its own use of the rule. The first draft listed a "not credible on low volume" roster, but made two mistakes — ① pre-market volume is cumulative, so an early-morning snapshot is necessarily low, and without a timestamp the call is simply wrong; ② the standard was not applied symmetrically, being used only to question quotes inconvenient to this brief's argument. Measured comparison:

At the same time, honesty requires pointing out: several readings this brief uses to support its own views are likewise below the 50,000-share threshold — SMH (45,512), XLK (36,985), ANET (30,395), VRT (30,379), CRDO (34,295), ADBE (14,499), DDOG (12,689), STX (19,065), WDAY (2,629). To avoid a double standard, this brief's approach is: every low-volume name is tagged in the tables with its measured volume and timestamp so readers can discount for themselves, rather than having this brief apply the threshold selectively to only some rows.

1.4 Full sample table (so readers can re-check this brief's reversal statistics)

Below are all 35 names used for the three rows of statistics in section 1.0, sorted by the 8/12 daily change in descending order. The reason for publishing it: a statistic that cannot be re-checked should not be used to support a conclusion. Snapshots 08:30–08:34 ET; * = pre-market volume < 50,000 shares.

The one thing to notice most in this table: only three names are down more than 2% this morning — COHR, CSCO, AAOI — and two of them reported last night; while the six supposed "yesterday's losers all bouncing" (MSFT/CRM/NOW/DDOG/ADBE/WDAY) are all between +0.40% ~ +0.64%, with only CRM reaching +1.46%. Calling displacement of that magnitude "a turn in money flow" is over-reading the data — section 1.0 of this brief has downgraded the conclusion accordingly.

2. Strongest Themes, Descending

---

3. Master Single-Stock Strength Ranking

Grouped by direction, sorted by total score descending within each group. The scoring model is in section 4. "Prior-day %" and "today's pre-market %" are data from two different trading days — read them separately.

3.A Positive Direction

3.B Negative Direction

---

4. Single-Stock Scoring Model (100 points total)

Scoring example (HLIT, total 74): source 13 (company earnings) + catalyst directness 18 (own results) + earnings elasticity 14 (EPS double the estimate, broadband revenue above the top end of guidance) + moat 8 (small cap, a single business after the divestiture) + expectation gap 9 (not previously priced by the market) + persistence 6 (one quarter of confirmation; a second is still needed) + trading characteristics 10 (240,000 shares pre-market is ample for this market cap) − risk −4 (already gapped 23%, chase risk) = 74.

Scoring example (DELL, total 60): source 9 (mostly a media read) + catalyst directness 10 (not its own disclosure) + earnings elasticity 12 (rapid AI server revenue growth) + moat 12 (scale and supply chain) + expectation gap 5 (SMCI guidance has been public for a day) + persistence 8 (a long AI capex cycle) + trading characteristics 10 (excellent liquidity) − risk −6 (close to the 52-week high + the gross-margin tax) = 60.

⚠️ Correction: in the first draft of this brief the components of these two examples did not add up to the totals given (HLIT's components summed to 70 but 80 was written; DELL's summed to 51 but 71 was written). They have been recomputed and the components adjusted, and both examples now add up. In the section 3 tables, HLIT is changed from 80 to 74 and DELL from 71 to 60, with the ordering adjusted accordingly. An arithmetic error in the very place that shows the calculation would render the whole scoring system meaningless — hence this explicit correction rather than a silent edit.

---

5. Detailed Analysis of Top Names

5.1 CSCO (Cisco) — today's most important earnings report, and the easiest one to read backwards

Related news: 8/12 16:05 ET, SEC 8-K EX-99.1¹.

First-hand figures (all from the 8-K itself, not media relays):

• FQ4'26 revenue $17.252 billion (+18%), product revenue +24%, services flat; by region Americas +18%, EMEA +19%, APJC +14%
• Product lines: networking $9.791 billion (+28%), security $2.226 billion (+14%), collaboration $1.167 billion (+12%), observability $275 million (+6%)
• Q4 total product orders +35% yoy; +25% excluding hyperscale customers; networking product orders +40%, the eighth consecutive quarter of double-digit growth
• AI infrastructure orders: $4 billion in Q4, $9.3 billion cumulative in FY26; AI-related revenue about $4 billion in FY26, expected $7.5 billion in FY27
• Non-GAAP EPS $1.22 (consensus $1.17), GAAP EPS $0.97 (+52%)
• FY26 revenue $63.325 billion (+12%), non-GAAP EPS $4.33
• Q1 FY27 guidance: revenue $18.0–18.2 billion, non-GAAP gross margin 65%–66%, non-GAAP operating margin 35.5%–36.5%, non-GAAP EPS $1.32–1.34
• FY27 guidance: revenue $72.2–73.4 billion, non-GAAP EPS $5.05–5.11

Catalyst logic — why revenue and guidance both look good yet the stock is down 6%:

First, let's take the "looks great" part at face value: FY27 revenue guidance of $72.2–73.4 billion against FY26's $63.325 billion is +14%~+16%, an acceleration from FY26's +12%; non-GAAP EPS of $5.05–5.11 against $4.33 is +17%~+18%. On those two lines alone, this is guidance for accelerating growth.

The problem is in the gross margin line, and it is spelled out plainly in the original document:

That is the whole answer. Product revenue +24%, networking +28%, orders +35% — the volume side is beyond reproach; but every dollar of product revenue carries 2.7 points less gross profit than a year ago, and the company's own next-quarter guidance sets gross margin below what it just achieved. This is not a one-off seasonal wobble — the cause is structural: Q4 total orders were +35%, but only +25% excluding hyperscale customers, meaning the extra 10 percentage points of the increment came from hyperscale customers, and those customers have enormous bargaining power and hand out low-margin orders. In other words, the fastest-growing part of Cisco's business is precisely the least profitable part.

The company offset it with opex discipline — non-GAAP operating expenses rose only +5% (far below revenue's +18%) — so the non-GAAP operating margin actually rose from 34.3% to 35.9%. But expense discipline has a floor, while a gross-margin decline does not; once revenue growth slows while gross margin is still compressing, operating margin falls fast. What the market is selling today is that second-order problem, not the first-order beat.

Three additions from the fundamentals check (absent from the first draft, but they change how much weight the conclusion carries):

(1) Gross margin did not "drop this quarter"; it has been dropping for four consecutive quarters. Non-GAAP total gross margin, extracted quarter by quarter from each period's 8-K:

FY26 full year 66.9% vs FY25 68.74% = −181bp, and the year-over-year decline is widening (Q1 −120bp → Q2 −120bp → Q3 −260bp → Q4 −210bp). The guidance midpoint of 65.5% is another 80bp below the 66.3% just delivered and 260bp below the 68.1% of Q1'26 a year earlier. This is no longer a fluctuation; it is a trend line.

(2) Cash flow: FY26 was a year of "revenue +12%, non-GAAP EPS +14%, free cash flow −3.9%".

Inventory +80% against revenue +17.6%, with days of inventory stretching from 53 to 84. It is the same shape as COHR's +79.5%. It could be pre-building for AI orders and tariff risk (neutral to positive), or a mismatch in shipment timing — the 8-K does not break out the composition of inventory; that requires the 10-K, and this brief will not infer it. But whichever explanation holds, "profit is rising and cash is not keeping up" is established.

(3) Valuation is at a five-year high, and most of the past year's gain did not come from earnings.

EV/EBITDA is 28.03x, versus a FY22–FY24 range of 11.1–14.0x (currently 2.0–2.5x that); P/FCF 38.24x (11.13x in FY23); P/S 7.71x. No year in the past five comes close to the current level — this is the high point of visible history, and there is no "percentile cushion". More important is the decomposition of the past year's +75.29%: TTM non-GAAP EPS went from $3.81 to $4.33 (+13.6%), while the TTM non-GAAP P/E went from 18.5x to 28.6x (+54.2%). About 54pp of the gain came from multiple expansion and only about 14pp from earnings. A stock three-quarters driven by multiple, being repriced while gross-margin guidance points down, is arithmetically very straightforward.

Consensus (tagged to be verified): the stockanalysis aggregate of 24 analysts gives FY27 revenue of $68.8B and non-GAAP EPS of $4.80 — if that figure is a clean pre-earnings snapshot, then Cisco's guidance is 5–6% above consensus on both revenue and EPS, with the deviation consistent on both lines and internally coherent. But that page has an obvious flaw (its "EPS Growth 2027 +25.9%" back-solves to a base of FY25's $3.81 rather than FY26's $4.33, and the FY26 FCF on the same page contradicts the sum of the quarters), so this brief does not treat "guidance above consensus" as established fact. If this is overturned (i.e. actual consensus is $5.10+), the attribution for the decline has to be rewritten from "valuation and quality" to "guidance below expectations".

⚖️ The counter-evidence that must be given (three items from the same 8-K pointing the other way): honesty requires listing what is inconvenient for this brief's argument —

1. Q1 FY27 non-GAAP operating margin guidance is 35.5%–36.5%, a midpoint of 36.0%, above the 35.9% just delivered. That is, in the very same table where it lowered gross margin guidance, the company raised operating margin guidance.
2. FY27 guidance implies non-GAAP EPS growth of +16.6%+18.0%, above revenue growth of +14.1%+16.0%. The company is telling the market that even with gross margin declining, EPS will grow faster than revenue. This is direct counter-evidence against "the gross-margin tax = an S-grade structural negative".
3. The 8-K forward-looking statements explicitly list tariffs as a risk variable for Q1/FY27 guidance. This brief attributes the gross-margin decline to hyperscale bargaining power, but the company itself does not break out the attribution — tariffs are another equally plausible explanation. So this section's attribution should be read as "mixed factors (customer mix + tariffs), not broken out by the company", rather than a single cause.

A further note on the quality of GAAP EPS: within Q4 GAAP EPS of $0.97 (+52%), pre-tax profit of $4,933M includes $869M of investment gains (17.6% of pre-tax; already excluded from non-GAAP). Stripping it out, GAAP EPS ≈ $0.80. The clean convention is non-GAAP EPS of $1.22 (+23%).

Theme and stage: AI networking demand is in the mid-to-late fermentation stage, but the narrative that "Cisco is an AI beneficiary" is itself entering a repricing stage — from "AI makes Cisco grow again" to "AI makes Cisco grow, but less profitably and with less cash generation".

Pre-market and technicals: 8/12 close $123.88 (+2.86%), this morning's pre-market $116.79 / −6.09% (snapshots 08:01–08:11 ET, pre-market volume 767,000 shares, ample liquidity — this decline is credible). The 52-week range is $65.75–$130.366 — yesterday's close was only about 5% below the 52-week high, so it received this report from a high perch, which amplifies the damage from the gross-margin negative. Worth noting: the initial after-hours drop was about −4.19%, widening to −6.09% by this morning's pre-market — the decline deepened with time rather than converging, which suggests this is not an emotional instantaneous reaction but institutions continuing to cut after digesting the call and the guidance.

Final verdict: avoid (short term). The reason is not "it fell", but the combination of three negatives: ① the gross-margin guidance comes from the company itself, covers the next quarter, points clearly downward, and follows four straight quarters of decline, with no information event that could falsify it near term; ② cash-flow quality is deteriorating (FCF −3.9%, inventory +80%), and almost nobody discussed this this morning; ③ valuation is at its highest percentile in five years, and three-quarters of the past year's gain came from multiple expansion rather than earnings. "Cheap" simply does not apply here — it is not cheap.

The signals to wait for are very specific: FQ1 non-GAAP product gross margin stabilizing sequentially (no lower than 64.8%), and days of inventory falling back from 84. Until those two, the decline itself is not a reason to buy.

This brief's biggest information gap (which must be stated honestly): the 8/12 conference call transcript was not obtained. The pivot of this section's conclusion is that "the 65–66% of Q1 guidance will carry through the full year", which is an extrapolation, not company disclosure — the 8-K contains no full-year gross-margin guidance, while the call usually does. If management gave a full-year range on the call, or said Q1 is the trough with a recovery after, the attribution above would need rewriting. Readers should check the call transcript themselves before making a judgment.

5.2 COHR (Coherent) — it looks the same as CSCO, but it is a completely different thing

Related news: 8/12 16:05 ET, company press release².

First-hand figures (from the SEC 8-K EX-99.1¹⁴ original):

• FQ4'26 (ended 6/30): revenue $2,045.5M (+13.3% q/q, +33.8% yoy); consensus $1.99 billion → beat by 2.8%
• GAAP gross margin 38.5%, GAAP EPS $1.19; non-GAAP gross margin 40.2%, non-GAAP EPS $1.74 (consensus $1.61 → beat by 8.1%)
• Against the company's own FQ4 guidance (revenue $1.91–2.05 billion, non-GAAP EPS $1.52–1.72, non-GAAP gross margin 39.0–41.0%): actual revenue of $2,045.5M came in slightly below the $2,050M top end (it only shows as $2.05 billion after rounding); EPS $1.74 broke through the top end; gross margin of 40.2% was only 20bp above the midpoint. ⚠️ Correction: the first draft said "revenue hit the top end", which is inaccurate — of the three, only EPS genuinely beat the top of guidance, and gross margin is the one dimension in this report that did not beat.
• FY26 full year: revenue $7.118 billion, a record (+22.5%); non-GAAP operating profit +40.5%; non-GAAP EPS +58.3% yoy
• FQ1'27 guidance: revenue $2.2–2.4 billion, non-GAAP gross margin 39.5%–41.5%, non-GAAP operating expenses $400–420 million, tax rate 18%–20%, non-GAAP EPS $1.85–2.05
• Segments: datacom and communications $1,615.0M (+58.6% yoy, 78.9% of revenue vs 66.6% a year earlier); industrial $430.5M (−15.8% yoy)
• Management commentary (media relay, to be verified): CEO Jim Anderson said FY27 growth will be higher than FY26 and that customer orders are booked into calendar 2028

Catalyst logic — clear the arithmetic first, then talk narrative

The most important thing about this stock today is that it did not really fall.

The report came after the close on 8/12, and the stock had already risen 8.24% that day ($328.57 → $355.64). Comparing this morning's pre-market $331.03 with the last clean, unaffected close before the report, $328.57 (8/11):

$331.03 / $328.57 − 1 = +0.7%

That is: taking "front-running + earnings + this morning's give-back" as one whole, the market's net pricing of this report is +0.7%, roughly zero. This morning's −6.68% is not the market vetoing the report; it precisely gives back the bar bought a day early. If you look only at the number "pre-market −6.68%" and conclude "the good news is out", you reach exactly the opposite conclusion.

Three errors in this brief's first draft that must be publicly corrected (found by the fundamentals check):

More importantly, the "crowded at the highs" characterization does not hold at all. COHR's 52-week high is $440.00 (2026-06-03); the 8/12 close of $355.64 is −19.2% from that high, and this morning's pre-market $331.03 is −24.8%; the 6/30 close was $394.47,8/12 is 9.8% below that quarter-end. The real shape is: from $440 down to $222 in July (−49.5%), then a 60% rebound over 10 trading days. The 60-day annualized volatility is 106.8%. On a stock with annualized volatility above 100%, a 7% pre-market move is less than one standard deviation, and treating it as "a vote against the earnings report" is statistical over-reading.

Equally must be withdrawn: the first draft said "the FQ1 gross-margin guidance midpoint of 40.5% versus this quarter's actual 40.2% is only 30bp of expansion, i.e. expansion is slowing" — that is a convention mismatch, comparing an "actual" against a "guidance midpoint", which systematically understates. The company's gross-margin guidance midpoints across four quarters were:

The company has raised its gross-margin guidance midpoint by exactly 50bp every quarter for four straight quarters (39.0→39.5→40.0→40.5), with no change in cadence. Compared on the same basis, "slowing" does not hold. A harder test is incremental gross margin (the marginal gross margin on new revenue): FQ4 actual 44.7%; the FQ1 guidance midpoint implies 42.9% — still well above the installed-base average of 40.2%, i.e. new business is still pulling the overall gross margin up; this is not a topping formation.

But a range must be given rather than a single point: the low end of guidance (revenue $2.2B / gross margin 39.5%) implies an incremental gross margin of only 30.2%, while the high end (revenue $2.4B / 41.5%) reaches 49.0%. A 200bp guidance width corresponding to a 30%→49% span in incremental gross margin itself says the company is not confident about yields and ramp timing on new capacity. Only if it lands at the low end would there be a genuine sequential contraction.

So what is COHR's real negative today? It is the cash flow page — new information disclosed in full for the first time after the close on 8/12, and this morning's discussion has barely mentioned it.

Non-GAAP EPS +58% yoy was achieved on a share count up 30% yoy — unit profitability really is improving, and the company deserves credit for that. But this year's accounting profit did not become cash, the company is still building capacity at an intensity equal to 27% of revenue, and it is filling the cash gap by issuing stock. Inventory +79.5% has two diametrically opposed readings: stocking up for FY27 shortages (positive), or demand pulled forward / a precursor to obsolescence (negative). The current data cannot distinguish them; that requires the inventory breakdown (raw materials / work in process / finished goods) in the FY26 10-K, which has not yet been filed.

Valuation (figures from the fundamentals check, on the 8/12 close): non-GAAP TTM P/E 63.4x; forward P/E (FY27 consensus EPS $8.37) 42.5x — the highest value across the past 20 quarter-ends is 37.45x, so this is the 100th percentile; P/S 9.77x (~95th percentile); P/B 6.38x (~95th percentile); FCF yield −1.47% (the worst on record). The only metric that is not expensive is PEG 0.86 (~39th percentile), but that rests on FY27 non-GAAP EPS actually reaching $8.37 (+49%). Bulls and bears are each holding a different row of this same table — which itself says the disagreement is about valuation, not results.

Theme and stage: the 800G→1.6T optical upgrade is in mid-fermentation and the fundamental cycle has not peaked; but growth is entirely staked on a single engine — the datacom share has risen from 66.6% to 78.9%, while the industrial segment is −15.8% yoy. This is a valuation for "AI optical interconnect purity", not for a "photonics platform".

Pre-market and technicals: 8/12 close $355.64 (+8.24%), this morning's pre-market $331.03 / −6.68% (pre-market volume 460,000 shares, credible). 52-week range $84.35–$440.00; −24.8% from the 52-week high.

Final verdict: watch closely. Results and guidance really are strong (incremental gross margin still rising, full operating leverage, order visibility into 2028), and this morning's decline is basically the flattening of front-running rather than a rejection of the results; but the new cash-flow negative is real, and valuation is at a five-year peak percentile with a negative FCF yield. The conclusion is not "wrongly sold off, buy it" but "it did not fall much and it did not get cheaper" — a 42.5x forward PE combined with negative free cash flow needs an FY27 cash-flow inflection, not another beat. Specific checkpoints: ① whether FQ1 actual gross margin continues the pattern of landing 0–20bp above the midpoint (i.e. ≥40.5%); ② FY27 capex guidance and the timing of FCF turning positive (on the call; not obtained by this brief); ③ the inventory breakdown in the 10-K.

Methodology note (worth remembering): the four errors in this section's first draft — overstated gains, the wrong window, the "crowded at the highs" characterization being backwards, and the gross-margin convention mismatch — share one cause: I had the narrative "sold off after a beat = crowding being flushed out" first, then went looking for numbers to support it, without first computing "the last clean close before the report". The correct order is always: the pricing impact of an after-hours earnings report must be measured against the price at the moment before release, never against the closing price of the release day.

5.3 CBRS (Cerebras) — it did not miss at all, but its reasons to fall are far worse than a miss

Related news: after-hours earnings 8/12 16:05 ET (StreetInsider¹⁸, CNBC¹⁹).

⚠️ Major correction: the premise on which this brief's first draft (and nearly all of this morning's media coverage) rests — a "double miss on revenue and EPS" — does not hold after first-hand verification against the 8-K and 10-Q.

After pulling the reconciliation table from SEC 8-K EX-99.1³, it is clear the sell side never modeled the GAAP convention:

The media subtracted the GAAP $180.11M from the sell side's $194M built on core revenue, and the GAAP −$2.98 from an adjusted-basis −$0.18 — subtracting two different things. On the two conventions the market actually models, CBRS beat on both. So the pre-market −15.3% is not a reaction to a miss — it has to be explained by something else. And the real explanation is uglier than a miss.

Catalyst logic — what is inside the "core revenue" convention

The 8-K reconciliation table verbatim (thousands of dollars, Q2'26):

GAAP revenue                              180,110
Less: Pass-through revenue                (14,503)   ← data-center pass-through revenue, excluded
Add:  Amortization of customer warrant     44,262   ← customer warrant amortization, added back
Core revenue                              209,869

Note the direction: "core revenue" is not a subset of total revenue; it is $29.8 million higher than GAAP total revenue. The mechanism is that Cerebras issues warrants to large customers in exchange for contracts, and under ASC 606 that warrant amortization reduces GAAP revenue, which the company then adds back in its non-GAAP convention. Three consequences follow that must be spelled out:

First, growth is inflated by 43 percentage points by the convention. Core revenue is +103.1% yoy, but the add-back in Q2'25 was zero (there were no customer warrants then). Excluding the add-back on a like-for-like basis, real growth is +60.3%. The $44.262M add-back equals 21.1% of core revenue.

Second, the direction of the hardware business is flipped by this add-back — this is the sharpest point:

The same business is shrinking 23% on GAAP and growing 17% on "core", with the entire difference coming from an accounting add-back (34.1% of core hardware revenue). And the GAAP hardware gross margin is only 1.8%.

Third, the sequential data is deteriorating across the board, and the year-over-year growth rate hides it:

• GAAP total revenue $193.406M (Q1) → $180.110M (Q2) = −6.87%, the first sequential decline
• Hardware revenue −51.1% q/q ($110.6M → $54.1M); the cloud business +52.1% q/q
• GAAP gross margin collapsed from 44.6% in Q1 to 14.2% in Q2
• Q3 guidance deteriorates further: core gross margin 38–40% (Q2 was 40.6%), core operating margin −25%~−23% (Q2 was −16%)

Fourth, the quality of the "raised full-year guidance" depends on a very steep shape. The full-year core revenue guidance midpoint was raised by $25 million (+2.9%). But H1 core revenue is already $401.217M and Q3 guidance is $214–216M, which back-solves to Q4 core revenue of $268.8M, i.e. +25.0% sequentially. A shape with Q3 flat and Q4 jumping 25% concentrates the delivery risk of the full-year guidance in the final quarter.

On the $377 million of stock compensation: it really is one-off, but the "non-cash" label is overrated. Q1'26 SBC was only $9.59 million and Q2 was 39.3x that — the classic catch-up recognition of IPO double-trigger RSUs (granted pre-IPO, conditioned on "service period + liquidity event", all recognized in the IPO period). Structurally it will not repeat. But cash really did go out: the half-year financing activities include a line for "Tax withholding from tender offer and IPO −$416.66 million", plus $22.8 million of employer payroll taxes. Around this RSU settlement the company actually paid out about $440 million in cash. Calling it a "purely non-cash expense" understates the economic cost. Not obtained: the normalized post-IPO SBC run rate (the 10-Q shows no "unrecognized stock-based compensation cost" balance); it will not return to $9.59 million per quarter, but where it will land cannot be judged.

Customer concentration: it has not gone away, it has changed hands — and the new structure is more troublesome. From the 10-Q notes (share of total revenue):

The top three customers were 76% of revenue in Q2. More concerning is the OpenAI line — the 10-Q states that its master agreement recognized $56.8 million of revenue in Q2 (31.5% of GAAP total revenue), while Customer B accounts for 57% of accounts receivable; at the same time OpenAI provided Cerebras with roughly $1 billion of secured promissory notes in January 2026 (the $918.2 million working-capital loan on the balance sheet), and Cerebras in turn issued Class N warrants to OpenAI. Revenue, receivables, creditor exposure and equity are all tied to the same counterparty — those four legs fail in the same direction; that is correlation risk, not diversification. Media claims such as "G42 was 85% in 2024" were not verified against primary documents by this brief and are tagged to be verified; the table above comes from the 10-Q.

RPO of $25.4 billion (6/30/26), which the 10-Q says is substantially attributable to the OpenAI master agreement. Recognition schedule: about 22% in the next 24 months (≈$5.6 billion), 43% in months 25–48, the rest later. Positive: 22% over 24 months ≈ $2.79 billion a year, which lines up with the FY27 consensus of $2.904 billion, showing that "revenue triples in 2027" has contractual support rather than being a slogan. Negative: 78% of RPO falls after mid-2028, and RPO is stated gross, including data-center pass-through costs (explicit in the 10-Q), including near-zero-margin sublet power and rent, which inflates its economic meaning.

⏰ Time-sensitive item — the one most easily missed this morning: the 10-Q states the lockup runs to the second trading day after the Q3'26 earnings report or 180 days from the prospectus date (≈2026-11-10), whichever is earlier; but the same passage discloses that up to about 171.1 million shares may be released during the lockup through early-release provisions, of which about 15 million shares are held by directors and officers. 171.1 million shares = 72.0% of the existing share count. The specific triggers for early release were not obtained by this brief (to be verified, highest priority) — if any tranche is tied to an earnings date, a supply shock could be imminent.

Valuation and share count (must use the true share count): the 10-Q cover shows the three classes of stock totaling 237,564,041 shares, while most free market-data sources (including the Yahoo Finance family of APIs) record 222.851M, understating it by 6.2% — these sources miss one of the three layers of the share structure, and valuations computed from the market cap they display are systematically about 6% too low; use the figures recomputed here on the true share count. On the true share count: 8/12 closing market cap $62.26 billion, pre-market $52.38 billion; pre-market P/S (TTM GAAP) is still 77.0x, or 59.2x on the FY26 core guidance midpoint. Net cash is $7.688 billion, so liquidity is not a near-term risk. The cash burn is accelerating: Q2 operating cash flow −$59.8 million, capex −$416.9 million, single-quarter free cash flow −$476.7 million, roughly 4x wider than Q1's −$119.6 million, driven by a self-built 600MW data center — the business model is shifting from selling systems (asset-light) to running an inference cloud (asset-heavy), which is a substantive change to the valuation framework.

Two background items also need correcting: ① the company did not list "around 2025"; its first trading day was 2026-05-14 at an IPO price of $185.00, only 62 trading days ago; the pre-market $222.08 is +19.2% versus the IPO price but already −29.1% versus the first-day close of $311.07. ② The FY2025 book "profit" came from a $363.3 million non-cash gain on the release of a forward contract liability in Q2'25, while operating profit in the same period was −$145.3 million — it was not operating profitability.

Pre-market and technicals: 8/12 close $262.06 (+11.63%, entering the report on a big gain), this morning's pre-market $222.08 / −15.26% (pre-market volume 628,000 shares, credible). 52-week range $160.81–$386.34.

Final verdict: avoid — but note that the reasons are completely different from the ones the media gives. It is not because of a "miss" (it actually beat), but because: GAAP revenue declined sequentially for the first time, hardware revenue halved sequentially, GAAP gross margin collapsed from 44.6% to 14.2%, 21% of the core-revenue increment comes from an accounting add-back that did not exist in the base period, both Q3 margin guidance lines were cut, the full-year guidance concentrates delivery pressure on a +25% sequential jump in Q4, a single customer is simultaneously 32% of revenue / 57% of receivables / a $1 billion creditor, and 72% of the share count faces early release during the lockup. At 77x TTM P/S, that combination leaves very little room for execution slippage.

This section is today's most valuable lesson: when the media tells you "it missed", first check which convention the sell side is actually modeling. This time the fact is — it beat, and its reasons to fall are far worse than a miss.

5.4 DELL / HPE — the only group in the field that "rose yesterday and is still rising today", and why they are dangerous

Related news: no first-hand disclosure of their own. The driver is SMCI's FY27 revenue guidance of $65–72 billion given after the close on 8/11 (FY26 was $39.1 billion) plus CoreWeave's earnings, forming an AI server sector read-across (24/7 Wall St.²⁰). Dell's own AI server backlog of $51.3 billion is media-sourced and was not verified against primary documents by this brief.

Catalyst logic: this group is today's outlier. Of the 24 names that rose more than 1% yesterday, 19 are down pre-market this morning, while DELL (+3.37%), HPE (+3.54%) and SMCI (+1.20%) are not only not falling but rising. Why is it precisely the group with the worst gross margins that gets the exemption?

Because what the market is punishing today is not "low gross margin" itself but "the volume story is finished and only the gross-margin problem is left". Cisco's volume is already fully reflected in guidance ($72.2–73.4 billion says it all), so the only thing left to reprice is gross margin — and it gets hit. For the AI server OEM group, the magnitude of the volume revision is still expanding — SMCI guided FY27 revenue from $39.1 billion to $65–72 billion, a +66%~+84% revision, large enough to swamp any gross-margin discussion. As long as volume is still being revised up, gross margin is not priced; the moment the upward revisions stop, gross margin becomes the entire story overnight — Cisco today is a demonstration of that script.

Pre-market and technicals (the key risk):

• DELL: 8/12 close $484.50 (+9.87%), this morning's pre-market $498.99 / +2.99% (08:33 ET, pre-market volume 431,000 shares, credible). 52-week high $485.70 (intraday basis, 2026-08-05). ⚠️ Two corrections: ① the first draft wrote "pre-market $502.00 / +3.37%" — those two numbers are internally inconsistent ($502.00 / $484.50 = +3.61%), and $502 was a high from an earlier point; by the time of writing it had eased back to $498.99; ② the first draft said it was "already above the 52-week high" and described it as "chasing a new high" — this needs to be softened: the 8/12 close of $484.50 was itself only 0.25% below $485.70, so the pre-market "crossing" is just threshold noise and should not be given structural meaning. The accurate statement is: DELL is running right up against its 52-week high and briefly edged slightly above it pre-market.
• HPE: 8/12 close $58.79 (+8.11%), this morning's pre-market $60.26 / +2.50% (08:33 ET, pre-market volume 713,000 shares). 52-week high $64.25, about 6% away. (The first draft's "+3.42%~+3.54%" was likewise inconsistent with the prices and has been corrected.)

Final verdict: watch closely, do not chase. The logic holds, but the location is the worst possible — DELL is already above its 52-week high, and the basis for the rise is a second-hand read rather than its own disclosure. If you want to participate, the right approach is to wait for Dell's own earnings to give the gross-margin number for AI servers (not the order number) — that is the real verification point of this story. To put the risk in one sentence: you are buying the "volume" narrative, and the verification material for that narrative is missing exactly the gross-margin page — the same page Cisco filled in last night.

5.5 HLIT (Harmonic) — today's highest-quality advance on fundamentals

Related news: after-hours earnings on 8/12. Adjusted EPS $0.24 vs consensus $0.12 (double the estimate); broadband revenue $133.5 million, above the top end of the company's own $115–125 million guidance range, +54% yoy. This is the first full quarterly report after the 6/16 completion of the video business sale (Investing.com⁸).

Catalyst logic: three things hold at once, which is rare today — (1) it beat its own guidance top end (not the sell-side estimate but the company's own promise, a harder bar); (2) EPS was double the estimate, huge elasticity; (3) the business was just simplified — after divesting video it is a pure broadband access name with a clearer valuation anchor. Most of today's advances are either M&A (ARX, already at the price), read-across spillover (DELL/HPE) or holdings news (NFLX); HLIT is one of the few driven by its own operating results.

Pre-market and technicals: 8/12 close $12.00 (+4.80%), this morning's pre-market $14.81~$14.90 / +23.42% (pre-market volume 243,000 shares, ample for this market cap). 52-week range $8.465–$17.68; the pre-market price is already at the 69th percentile of the 52-week range, about 19% below the high.

Risks: a small-cap name; customer concentration not verified by this brief; it has already gapped 23%, so gap-fill risk is high; a single quarter's beat needs a second quarter before it can be called a trend.

Final verdict: priority deep-dive. The highest signal-to-noise name on today's list. But to be explicit: do not chase a 23% gap at the opening bell — the right observation point is whether it holds the low end of the pre-market range in the first 30 minutes and whether volume follows through.

5.6 CLBT (Cellebrite) — today's cleanest avoid

Related news: 8/13 pre-market. The company cut its full-year ARR and revenue outlook while raising its adjusted EBITDA target, and announced a CEO change: Shiven Ramji replaces Thomas E. Hogan, effective immediately (PR Newswire⁶).

Catalyst logic: "cut revenue guidance + raise profit guidance + CEO departs immediately" is a very specific combination, and it says more than any single item: the company has given up on its growth target, is switching to defending profit, and is changing the person who executes. For a SaaS/ARR business, a cut to ARR guidance is the most damaging kind of cut — it is not a quarter's revenue deferral but a downward revision of the future revenue base itself. Offsetting it with an EBITDA raise usually means cost cutting, and profit bought with cost cuts is a negative valuation signal for a growth stock, not a positive one.

The phrase "effective immediately" is also worth noting — an orderly CEO handover usually has a transition period, and immediate effect is more common in involuntary changes. This brief did not obtain an explanation for the change and will not infer one, but the wording itself raises uncertainty.

Pre-market and technicals: 8/12 close $15.25 (−4.87%, already falling before the report), this morning's pre-market $10.97~$10.98 / −28.00% (pre-market volume 316,000 shares, credible). 52-week range $11.02–$19.98 — the pre-market price of $10.98 has already broken below the 52-week low of $11.02, a new 52-week low.

Final verdict: avoid · short watch. Three negatives (ARR cut, revenue cut, CEO change) at once, plus a new 52-week low. But the specific risk of shorting must be stated clearly: after a −28% gap, any oversold bounce can be violent, and a new CEO's early days often come with "restructuring expectation" speculation — a short should wait for a bounce to fail rather than chase the gap low.

5.7 STUB (StubHub) — how "record results" and "down 20%" can both be true

Related news: 8/12 16:05 ET, SEC 8-K EX-99.1⁷ (primary document).

First-hand figures:

• GMS $3.1 billion (+34%), a record; revenue $573.1 million (+33%), a record
• Net profit $14.6 million (versus a net loss of $53.8 million a year earlier), net margin 3%
• Adjusted EBITDA $105.7 million (+94%), margin 18% (nearly 600bp of yoy expansion)
• Operating cash flow $321.9 million, free cash flow $309.7 million (just $9.7 million a year earlier)
• Cash $1.7 billion; $1.1 billion of debt repaid over the past 12 months, net leverage down from 4.5x to 3.0x
• Full-year GMS guidance raised to $10.1–10.3 billion; adjusted EBITDA guidance maintained at $400–420 million (reaffirmed only, not raised)

Catalyst logic — why a report like this earns −20%:

Almost every operating metric improved, and by a lot (EBITDA +94%, FCF from $9.7 million to $310 million, leverage 4.5x→3.0x).

⚠️ But the explanation circulating this morning ("adjusted EPS $0.00 vs consensus $0.24") is a conventions trap — the same error exposed in the CBRS section — and this brief does not adopt it. Checking the 8-K original: the term "Adjusted EPS" never appears anywhere in it. What the company disclosed is GAAP diluted EPS of $(0.00), and the cause of that $(0.00) is not an operating loss — net profit was +$14.6 million, but the amount "attributable to common stockholders" was −$40,000, the difference coming from earnings allocated to preferred / participating securities. Comparing that against a consensus built on an adjusted basis is subtracting two different things. The real reason for the decline has to be found elsewhere — and it does exist (see below).

There is a structure here that must be spelled out: the company raised GMS guidance but only "reaffirmed" adjusted EBITDA guidance. Put together, those two things tell the market that the extra business does not bring extra profit. For a platform company, if GMS growth cannot be converted into EBITDA growth, the valuation multiple on growth itself has to come down. That is a more substantive problem than a single quarter's EPS miss.

Another piece of context worth flagging: Q2 included the World Cup, a non-recurring event (the press release explicitly mentions a "record-setting World Cup"), which means part of that +34% GMS growth is not sustainable — and the market may be concluding from exactly that that the second half will decelerate.

Three more convention issues that are in the 8-K but were barely mentioned this morning, and they are more substantive than the fake EPS gap:

1. The $321.9 million of operating cash flow is not freely disposable. The 8-K states explicitly that it is "inclusive of net inflows of buyer receipts and seller payments" — i.e. it includes float collected on behalf of others. So the "$309.7 million" of free cash flow substantially overstates true cash generation.
2. Against the $1.7 billion of cash sits "$1.2 billion payable to sellers", a net of about $500 million. Using $1.7 billion of cash to argue balance-sheet strength distorts the picture.
3. First-half stock compensation jumped from $7.5 million a year earlier to $100 million — which is exactly why the company could raise GMS guidance while only "reaffirming" adjusted EBITDA guidance.

Pre-market and technicals: 8/12 close $8.54 (+4.27%), this morning's pre-market $6.79~$6.82 / −20.49% (pre-market volume 324,000 shares). 52-week range $5.74–$27.89 — the stock is about 75.6% below its 52-week high.

⚠️ Correction: the first draft called this "the first quarterly report after listing". That is wrong — SEC filings show its 424B4 prospectus is dated 2025-09-17, with earnings 8-Ks since then on 2025-11-13, 2026-03-04, 2026-05-13 and 2026-08-12, making this the fourth, not the first. The related discussion has been deleted.

Final verdict: avoid. The reason is not the non-existent EPS gap but: GMS guidance raised while EBITDA guidance is only reaffirmed (incremental business brings no profit), operating cash flow includes pass-through float, net cash is only about $500 million, and stock compensation went from $7.5 million to $100 million in a year. Cheapness (−75% from the high) is not a reason to buy in this situation, because the anchor for the valuation reset has not yet been repaired.

5.8 TPR (Tapestry) — why "hitting the targets two years early" is a negative

Related news: 8/13 pre-market, SEC 8-K EX-99.1⁵ (primary document).

First-hand figures:

• FQ4'26 (ended 6/27): net sales $1.8766 billion (+9%, +8% in constant currency; pro forma +12%/cc +11%); the Coach brand +15% (cc +14%)
• FQ4 GAAP gross margin 83.3% (+700bp); non-GAAP gross margin 78.1% (+180bp). ⚠️ Note that GAAP gross margin (83.3%) is higher than non-GAAP (78.1%); this anomaly comes from one-off items and cannot be read as a trend; the 8-K breaks the non-GAAP +180bp into about +170bp from operating improvement, +60bp from the Stuart Weitzman divestiture, and −60bp from tariffs
• FQ4 non-GAAP operating profit $362 million (+25%), non-GAAP operating margin 19.3% (+250bp)
• FQ4 GAAP EPS $1.68, non-GAAP EPS $1.32
• FY26 full year: net sales $8.0042 billion (+14%, cc +13%; pro forma +18%/cc +17%); non-GAAP operating profit $1.8658 billion (+33%); GAAP EPS $7.27, non-GAAP EPS $7.05
• FY26 returned $1.7 billion to shareholders; the board approved a 16% dividend increase
• FY27 guidance: mid-single-digit revenue growth, further operating margin expansion, low-double-digit EPS growth; the media-sourced figures are revenue $8.4–8.5 billion and non-GAAP EPS $7.80–7.90

Catalyst logic: the first line of the company's own press release headline is "achieved the investor day revenue, operating margin and EPS targets two years ahead of schedule". That sounds like the highest possible praise, but for the share price it is precisely the problem — targets hit two years early means there is no new target left to beat, only a return to the normal growth rate of the "long-term commitment".

Laying out the numbers makes it clear:

There is also a divergence at the brand level: Coach +15% in FQ4 while Kate Spade was −7% (media-sourced) — growth is carried entirely by a single brand while the second brand is still contracting. FY27 guidance also embeds a tariff rate in the middle of the 20% range on U.S. imported inventory, with management saying the net year-over-year impact is neutral (media-sourced, to be verified).

Pre-market and technicals: 8/12 close $153.74 (−4.24%, already falling the day before the report), this morning's pre-market $141.00 / −8.29% (pre-market volume 348,000 shares, credible). 52-week range $92.62–$164.80.

Final verdict: avoid. This is a very clean "deceleration" case: the quarter's data was excellent, the dividend was raised 16%, gross margin expanded — but all of that looks backward; the line that looks forward reads 14% → 5–6%. The valuation anchor the market gives consumer stocks is growth, not a historical report card.

5.9 Independent fundamental verification: the check overturned four claims in this brief's first draft

Before finalizing, this brief ran an independent first-hand data check on CSCO and COHR (quarter-by-quarter 8-K originals from SEC EDGAR + stockanalysis financial pages). The check overturned four claims from the first draft, each corrected in the main text. They are listed here because each represents a class of error that is easy to repeat:

The check also added two facts the first draft lacked entirely, and they carry real weight: ① the cash flow of CSCO and COHR is deteriorating at the same time (CSCO FY26 FCF −3.9%, inventory +80%, turnover 53→84 days; COHR FY26 FCF −$1.023 billion, operating cash flow only $79.5 million vs net income of $805 million, inventory +79.5%, share count +30%) — this upgrades the "gross-margin tax" into "gross-margin tax + cash-flow tax"; ② CSCO's valuation is at a five-year high (EV/EBITDA 28.03x vs 11.1–14.0x in FY22–24), and of the past year's +75.29% gain about 54pp came from multiple expansion and only about 14pp from earnings.

Note on data conventions: the financial data for CSCO and COHR come from the SEC EDGAR 8-K EX-99.1 originals for each period (first-hand) and stockanalysis.com (second-hand, cross-checked against the 8-Ks wherever possible). The quarter-by-quarter non-GAAP gross margin series all come from the 8-K originals rather than second-hand aggregates, and the full-year figure back-solved by weighting quarterly revenue matches the company's disclosure (CSCO: back-solved 66.94% vs 66.9% disclosed).

Two explicit "not obtained" items — do not treat them as confirmed: ① neither the CSCO nor the COHR 8/12 conference call transcript was obtained — the full-year gross margin / capex guidance for both is usually disclosed only on the call and is not in the 8-K, and this is this brief's largest information gap; ② the FY27 sell-side consensus conventions for both are in doubt (CSCO's $4.80 comes from a page found to contain a base-year error), so this brief does not use "guidance vs full-year consensus" as a pass/fail line.

5.10 JD (京东, JD.com) — a return to profit cannot cover a revenue decline

Related news: 8/13 pre-market, Q2'26 results (company IR / media relay¹⁰).

Figures (media-sourced): net revenue RMB 346.4 billion (media conversion about $51.1 billion, an implied exchange rate of about 6.78, with no conversion date stated), −2.9% yoy (the company attributes this to a high base); operating profit RMB 4.5 billion (versus a RMB 900 million loss a year earlier), operating margin 1.3%; non-GAAP operating profit RMB 5.5 billion (RMB 900 million a year earlier); net profit attributable to parent RMB 7.1 billion, non-GAAP net profit RMB 8.9 billion; diluted EPS RMB 5.01, non-GAAP RMB 6.29.

A convention conflict that must be flagged: the media simultaneously reports that "JD's food-delivery losses keep narrowing" and that "the non-GAAP operating loss of new businesses reached RMB 14.8 billion this quarter". If new businesses lost RMB 14.8 billion in a single quarter while the company's overall non-GAAP operating profit was a positive RMB 5.5 billion, the retail core's profit base would have to be enormous to offset it — this brief could not verify the relationship between these two conventions against the primary report, so it is not used as a basis for judgment and is simply flagged as a conflict.

Final verdict: avoid. There is only one core reason and it is a hard one: revenue −2.9% yoy. The profit improvement comes from efficiency gains in the core business and narrower losses in new businesses — a cost-side story; but for an e-commerce platform, a decline on the revenue side is the more fundamental problem. Pre-market −3.86% (pre-market volume 992,000 shares, credible).

5.11 AMAT (Applied Materials) — after the close at 16:00 ET tonight, today's only first-order event that has not yet happened

This is the most important item on today's calendar, and it has not happened yet. The report comes at 16:00 ET with the call at 16:30 ET. This morning's pre-market is $547.44 / −0.13% (pre-market volume 106,000 shares); the 8/12 close was $548.15 (+4.29%).

Market expectations (sell-side consensus aggregated by Yahoo Finance, retrieved 2026-08-13):

Consensus sits slightly above the midpoint of company guidance (revenue +0.6%, EPS +1.0%), a routine setup.

Why "beating expectations" is not enough tonight: AMAT has beaten on adjusted EPS in all of the past 8 quarters, by +3.45% to +7.89%, with a median of about +4.9%. A beat of around 5% is already fully priced. What actually trades tonight is the FQ4 guidance, not this quarter's numbers. The market currently implies FY27 revenue of +30.0% ($43.561 billion) and EPS of +40.0% ($17.254) — that is what needs supporting. In the past 30 days 11 analysts raised FY27 and none cut; the FQ3 EPS estimate went from $2.891 to $3.394 over 90 days (+17.4%). Expectations themselves have already run fast, and that cuts both ways.

Valuation: this is the hardest data point this brief obtained —

Even looking only at the past 12 months (which already contains the entire re-rating), the current P/E is still in the 88.9th percentile. PEG on the FY27 growth rate is 0.79, which looks cheap — but for a semiconductor equipment company, using a 2-year growth rate as the PEG denominator is fragile, because the denominator is itself a cyclical quantity. PEG<1 here is not evidence of "cheap"; it only says current pricing already assumes the AI capex cycle does not peak before FY27.

⚠️ An earnings-quality issue that must be spelled out: FQ2'26's net margin of 35.5% looks excellent, but of the $3.225 billion of pre-tax profit, "interest and other income, net" accounted for $771 million (21.8%), which the 10-Q states is "primarily driven by increased unrealized gains on equity investments"; the effective tax rate in the same period was only 13.0% (8.0% a year earlier, 25.2% in H1'25). Stripping out all non-operating items and recomputing at the TTM effective tax rate, operating EPS is about $8.50, implying a P/E of 64.5x rather than 55.4x.

This also explains an anomaly: FQ2'26 GAAP EPS of $3.51 is actually higher than non-GAAP EPS of $2.86 — because non-GAAP strips out the unrealized equity investment gains. So valuation in this section is stated on a non-GAAP basis throughout. There is another hidden correlation buried here: if the AI complex pulls back, AMAT loses its operating narrative and its investment mark-to-market at the same time.

Other fundamental facts worth noting:

• Most of the China risk has already been released: China's share of revenue went from 37% in FY24 → 30% in FY25 → 27% in FQ2'26 (a three-year low). The freed-up share was picked up by Taiwan (→27%) and Korea (→20%), i.e. a shift from Chinese mature nodes to AI logic and memory/HBM. Judgment: the remaining 27% is mostly mature-node equipment, precisely the part facing the greatest domestic-substitution pressure, so its quality is lower than the headline share suggests; but by the same token, the marginal damage from a one-off export-control shock is already far smaller than in FY24.
• Gross margin is a slow variable: the past 5 quarters have drifted up within a narrow 48.0%–49.9% band, and FQ2'26's 49.90% is a 5-quarter high; annually it is about +0.7~1.0pct a year — steady but not explosive.
• The cash-flow trend is downward: TTM free cash flow of $5.343 billion is below FY25's $5.698 billion and FY24's $7.487 billion (capex has gone from $1.19 billion in FY24 to an annualized $2.56 billion in FY26). FQ2'26's single-quarter operating cash flow / net income conversion was just 30% (83% in FQ1); TTM at 94% is still healthy but deteriorating.
• Buybacks have hit the brakes: TTM buybacks were $2.644 billion, versus $4.895 billion for all of FY25 and $1.670 billion in FQ2'25 alone. The pace has more than halved versus a year ago — at the 98th valuation percentile, the company itself is no longer an aggressive buyer.
• Quality is still first rate: ROE 39.2%, ROA 22.8%, net debt of just $154 million.

The 52-week position and that July crash nobody has explained in detail: the 52-week high is $739.67 (2026-06-30) and the stock is −25.89% from it; but zooming out, it is +254.9% above the 52-week low of $154.47, +113.3% YTD, and 42.1% above its 200-day moving average. The key detail: 6/30 close $723.00 → 7/29 close $436.45, −39.6% in a month, then a rebound to $548.15. Single days of −9.97%, −8.40% and +14.97% appeared repeatedly in that stretch — for a $435 billion equipment company, that daily volatility is itself evidence of "deeply divided narratives".

And the most intriguing point: during July's −30% in the share price, the sell-side FY27 EPS estimate actually rose from $16.66 to $17.25. Price and earnings estimates diverged in July — that was a valuation compression, not an earnings cut. Which means if tonight's guidance holds up, the compressed valuation has room to repair; if it does not, the compression continues.

On the causes of the July crash, the circulating explanations include profit-taking after a nearly 200% gain within the year, SK Hynix slowing HBM expansion, Meta selling AI compute capacity, and Michael Burry shorting AMAT at $729.40. These are all media-sourced; this brief verified none of them against primary sources (13F / company IR), tags them to be verified, and does not use them as a basis for judgment.

Final verdict: watch closely (no position into the event). Its fundamental quality is the best on today's list, but valuation sits at the 97.9th percentile of five years, the FCF yield is 1.23%, 22% of pre-tax profit comes from unrealized investment gains, and the company itself has cut buybacks in half. Tonight's asymmetry is clear: a small beat is already priced (8 straight beats), while any hesitation in the FQ4 guidance will collide with a 98th-percentile valuation. Do not build a position ahead of the print, and especially do not express the view with long options (IV crush risk the next day is high, and this brief did not obtain its IV data). The real observation points are the FY27 capex customer commentary and advanced-packaging orders on the call, not this quarter's EPS.

---

6. Negative / Avoid List

---

7. Intra-Theme Ranking

7.1 AI hardware (networking / servers / optical) — ranked by "is the volume revision still under way"

7.2 Software rebound — ranked by "does it have its own catalyst"

The shared problem in this group: apart from CRM and IBM, the only reason the rest are up is "they fell yesterday". "It fell yesterday" is not a reason to buy, and no new evidence emerged today to falsify the structural narrative that "AI erodes the SaaS seat-subscription model".

---

8. Open-Bell Verification Signals

① Pre-market (now to 09:30 ET)

• The single most important observation — but use the right metric. When watching COHR, do not look at the pre-market decline; look at the net change versus the last clean pre-earnings close ($328.57, 8/11). The −6.68% pre-market is measured against the 8/12 close, and 8/12 had already risen 8.24% that day; reading them together guarantees a misjudgment. As long as the net change is still positive, the market has not rejected this report.
• For CSCO, watch whether the decline keeps deepening. One caveat: this brief did not obtain a primary source for the "after-hours −4.19%" figure, and the after-hours tape is extremely thin; with the press release at 16:05 and the call at 16:30, a decline that widened after the call is more likely due to the call's content than to "institutions continuing to cut". This brief therefore downgrades it to a description of the phenomenon and no longer infers a mechanism from it.
• Whether DELL can hold above the 52-week high of $485.70. Falling back below it = a failed chase driven by a second-hand read.
• Whether HLIT's +23% gap is gap-and-go or gap-fill: watch whether it holds the low end of the pre-market range (around $14.81) in the first 30 minutes.

【 08:29 ET update — this observation point is already starting to give an answer 】 In the half hour from 08:01 to 08:29, the two stocks moved in opposite directions:

	08:01–08:11 snapshot	08:29 snapshot	Direction
CSCO	−6.09%	−6.53% (pre-market volume 905,000 shares)	Still deepening
COHR	−6.68%	−6.51% (pre-market volume 569,000 shares)	Starting to converge

Table scrolls horizontally

This is exactly the shape this brief's core judgment expects: CSCO's decline went from after-hours −4.19% → pre-market −6.09% → −6.53%, deepening all the way, consistent with the repricing pattern of "institutions continuing to cut after digesting the call and guidance"; COHR is starting to converge, consistent with "stabilizing naturally after flattening the front-running". Other names in the same window: CBRS −16.24% (still deepening), TPR −9.91% (deepened from −8.29%), DELL +2.99% (cooling from +3.37%, but $499.01 is still above the 52-week high of $485.70), ARX +44.6% (pre-market volume has reached 9.79 million shares). Note this is only a half-hour sample; whether it continues after the open is the real verification.

② Intraday (09:30–10:00 ET)

• Sector ETF co-movement is today's chief referee: if IGV keeps outperforming SMH, "rotation" holds and yesterday's winners will keep giving back; if SMH recovers and overtakes after the open, this morning was just profit-taking rather than a turn in money flow, and yesterday's main line is intact.
• Use the 35-name sample table in section 1.4 to test this brief. Under strict conventions the pre-market readings are 9/24 for "those up >1% yesterday falling >1% this morning" and 1/6 for "those down >1% yesterday rising >1% this morning". If those two ratios do not rise after the open, "rotation" does not hold at all and this brief's downgrade in section 1.0 was correct; if they rise significantly, the rotation happened after the open rather than pre-market. The sample and the conventions are fully published and can be re-checked directly.
• Volume confirmation: without volume behind it, today's direction can easily reverse at midday.

③ Options sentiment

• This brief obtained no implied volatility data at all; the following is inference from price action only, so discount accordingly: CSCO, COHR and CBRS are all on the day after earnings, so IV crush risk is high — even with the direction right, buying options can lose money as IV collapses. Do not express views with long options the day after earnings.

④ Risks

• The 08:30 ET PPI is the only variable today that can rewrite every conclusion. If PPI comes in clearly above the expected 4.9% y/y, "rotation" becomes "everything falls", and every relative single-stock judgment in this brief yields to macro.
• Reversal after a gap: HLIT (+23%), OMER (+17%), BIRK (+11%), CLBT (−28%) and STUB (−20%) are all large gaps, and the first 30 minutes often run opposite to the gap direction.
• A lone advance with no sector follow-through: if DELL/HPE rise alone while AI hardware weakens across the board, that is a danger signal, not a sign of strength.
• Index futures diverging from Treasuries: this brief did not obtain index futures quotes and substitutes SPY/QQQ pre-market, so precision is degraded — please note.

---

9.0 [08:30 ET addendum] July PPI is out: the headline is cool, the core is accelerating

The release came later than the main text's snapshot; the following is an addendum written after it. Data taken from the BLS release²¹ itself, not media relays.

This data cannot be read from the headline alone. The m/m came in at 0.0% almost entirely because energy pushed it down: final demand energy −3.1%, with gasoline −5.7%, and food −0.9%. Meanwhile goods prices excluding food and energy were +0.1%, the services component excluding trade and transportation/warehousing was +0.6%, and the core convention (ex food, energy and trade services) jumped from +0.1% in June to +0.4%.

In other words: cheap gasoline is masking accelerating underlying costs. At the same time June's m/m was revised up from −0.3% to −0.1%, a hawkish revision.

⚠️ A conventions trap that must be flagged: the market expectation quoted in section 1.1 of the main text ("core +0.3% m/m, +4.1% y/y") refers to the ex-food-and-energy convention, whereas the core the BLS release leads with is ex food, energy and trade services (+0.4% / +4.7%) — these are two different metrics and cannot be subtracted from one another directly. This brief therefore draws no conclusion on "whether core beat expectations" and only states the fact that, on the BLS convention, core accelerated from +0.1% to +0.4%.

Market reaction (the observation window must be stated honestly): this brief took successive readings roughly 20–60 seconds after the release: 10-year Treasury 4.672% (essentially unchanged), 2-year 4.178%, VIX 14.53 (slightly lower), DXY 99.935, SPY pre-market +0.16%, QQQ 0.00%, SMH −0.14% (from +0.02% into negative), IGV +0.57% (holding). But readings within one minute are not enough to represent the market's pricing of this data — this brief explicitly declines to write "the market ignored PPI"; the true reaction has to be seen after the 09:30 open. The only thing that can be said is: within that very short window, the divergence between SMH and IGV (software stronger than semis) was not interrupted by PPI, consistent with theme 2 in the main text.

The relationship between this data and this brief's core argument is tighter than it looks: at the micro level today, what happened is that Cisco's and Coherent's gross margins and cash flows are being squeezed; at the macro level this PPI says production-side costs excluding energy are accelerating (+0.1% → +0.4%), only masked by cheap gasoline. Two scales of the same thing — companies' input costs are rising while they have no pricing power against hyperscale customers, so the difference lands on gross margin. That also makes falling energy prices a double-edged factor: it lowers headline inflation, but the core pressure it masks is the source of the gross-margin story.

Implications for the September FOMC: the headline y/y falling from 5.5% to 4.7% reduces the urgency, but the core m/m acceleration and the June upward revision point the other way. Pricing is still "roughly fifty-fifty between a hike and no change", and this brief does not think this data is enough to break that balance; the decision still rests with the coming employment and CPI data.

---

9. Final Conclusions

① The 5 names most worth watching today

② Today's 3 strongest themes

③ What to avoid today

1. Every software bounce that is "up today because it fell yesterday" (except CRM and IBM, which have their own catalysts) — no fundamental support; they give it back when the rotation ends.
2. Chasing a name pinned to its 52-week high that is driven by a second-hand read rather than its own disclosure (DELL pre-market $498.99 vs the 52-week high of $485.70; and its $51.3 billion backlog figure carries no as-of date and may be last quarter's number).
3. The combination of a guidance cut plus a management change (CLBT).
4. Platform stocks whose "operating metrics look great but do not convert into profit" (STUB: GMS raised, EBITDA only reaffirmed).
5. Companies that tell a growth story on a custom convention that runs opposite to GAAP (CBRS: GAAP hardware revenue −23%, "core" hardware revenue +17%, with the entire difference an accounting add-back).
6. Every "unusual move" on pre-market volume below roughly 50,000 shares (PANW, CIEN, FN, WWW, JACK, ENS) — this is the easiest way to lose money today.
7. Energy stocks: Brent's six-session streak is broken, and EIA inventories rose 17.4 million barrels in a single week, the largest since early 2023.

④ The final one-sentence judgment

**Today is not a day to buy, it is a day to distinguish — and half of what needs distinguishing was reported backwards this morning: CBRS did not "miss" (it beat on both conventions the sell side models, but its reasons to fall are far worse than a miss), and COHR was not "a perfect report that got sold" (it is actually still up 0.7% versus the pre-earnings close, and this morning only gave back the front-running bar). The one thing unambiguously deteriorating is CSCO — non-GAAP gross margin down four straight quarters, free cash flow turning negative, inventory +80%, and valuation at a five-year high. Put those three things together with the PPI just released (a 0.0% headline squeezed out entirely by gasoline −5.7%, while core accelerated from +0.1% to +0.4%), and today's real main line is a single sentence:costs are rising, pricing power sits with the customer, and accounting profit is looking less and less like cash — and that has only just begun to be priced.

⚠️ Data-retrieval and operations log for this run (not sent to clients)

Retrieval failures and workarounds:

1. CNBC index futures @ES.1 / @NQ.1 returned empty fields this time (the other cross-asset fields were normal). Substituted the stockanalysis pre-market prices of SPY/QQQ/IWM/DIA, with the loss of precision noted in section 8.④ and in the opening conventions.
2. WebFetch returned HTTP 403 on direct connections to cnbc.com, thestreet.com and sec.gov; SEC succeeded after switching to python3 + urllib with a compliant User-Agent (including an email address); direct connections to CNBC/TheStreet were abandoned in favor of search summaries plus cross-validation from other sources.
3. All three candidate endpoints of the stockanalysis premarket screener API returned 404 (/api/screener/s/f, /s/i, /s/d); switched to WebFetch on the /markets/premarket/gainers|losers/ pages, which worked. If this is to be hard-wired later, the API paths need to be re-probed.
4. WebFetch on the globenewswire Coherent press release timed out (60s), so COHR's first-hand figures were taken from multi-source cross-checking in the search results (Seeking Alpha + StockTitan + the GlobeNewswire summary); the full text of that release was not obtained, so COHR's off-statement data such as cash flow and debt was left blank in this brief.

Information traps (worth retaining): 5. The SEC URL returned by search was for the wrong quarter. Searching "Tapestry FY2026 Q4 8-K" returned ef20072115_ex99-1.htm, which is actually the Q3 press release from 2026-05-07, even though the title says "FY2026". Switched to data.sec.gov/submissions/CIK0001116132.json and located that day's 8-K by filingDate (0001140361-26-032624) to get the correct Q4 document. Lesson: any EDGAR link returned by search must be re-verified by date via the submissions API; you cannot judge the quarter from the FY string in the filename. This is the same failure family as stale-source-silent-lag in memory. 6. investing.com's pre-market roundup article mixed two trading days into one table. It listed "SMCI +19.02%" (actually the 8/12 regular session) alongside "CSCO −5.8%" (genuinely this morning's pre-market), and MQ's "+4%" also did not line up (actual pre-market +0.77% on 4,945 shares). This brief only found it after re-checking each ticker with the quote API, splitting chg%_prev_session and pm_chg% into two columns. The public correction is in section 1.3 of the main text.

A convention concern in the previous report (needs re-checking): 7. The "pre-market broad market and cross-asset" table in the 2026-08-12 brief may have used the previous close as the pre-market price. That table said "SPY $772.49 +0.25%", while today sa_q.py returns prev_close=772.49 and chg%_prev_session=0.25 for SPY — i.e. the SPY row in that 8/12 table matches "the 8/12 close + the 8/12 daily change" to the cent. The probability of a pre-market price equaling the same day's close to the cent is tiny; more likely the retrieval used the p/cp columns (previous close + prior-session change) rather than ep/ecp (pre-market price + pre-market change). If so, the entire ETF table in that 8/12 brief actually showed 8/11 closing data rather than 8/12 pre-market. This brief uses ep/ecp throughout and labels the two trading days in separate columns (see section 1.2). Recommend re-checking the same table in the 8/12 brief and a few earlier ones.

8. yfinance was rate-limited across the whole chain this time. Two fundamentals-analyst runs independently reported YFRateLimitError, failing at the crumb stage, with neither .info nor .history usable (yfinance 1.5.2). Switched to SEC EDGAR originals (first-hand) + stockanalysis.com (second-hand). Note this differs from both root causes recorded in memory under yfinance-rate-limit-root-cause — it was not a leftover process but a rejection at the crumb handshake, which looks like a pure IP-level block. Per the skill's rules, the main text only states "where the data came from" and does not mention rate limiting.

9. Two dirty data fields to hard-code as exclusions in future retrievals:

• yfinance.info['forwardEps'] returns 13.59 for COHR (implying a forward PE of 26.2x), while the sell-side consensus is $8.37 — a 62% gap, judged dirty data; do not cite it anywhere.
• The trailingPE / Revenue(ttm) fields for COHR on both stockanalysis and yfinance are still stuck on the March quarter; citing them directly pre-market lags by a full quarter. TTM must be recomputed by hand from the latest 8-K.
• The stockanalysis forecast page has a base-year error for CSCO: its "EPS Growth 2027 +25.9%" back-solves to FY25's $3.81 rather than FY26's $4.33; the same page shows FY26 FCF of 7.72B, contradicting the quarterly sum of 12.77B. The consensus figures from that page are tagged "to be verified" in the main text and are not used as a basis for judgment.
• The stockanalysis Debt/EBITDA row for COHR uses a single quarter's EBITDA (reporting 13.94x / 6.11x); the correct values are about 2.25x / 0.86x. Do not cite that row directly.

10. Sub-agent execution: fundamentals-analyst × 3 (CSCO / COHR / CBRS+AMAT) + risk-auditor × 1. The CSCO and COHR reports came back and overturned four first-draft claims (see section 5.9 of the main text, publicly corrected) — these two agents were extremely valuable this time, again validating the judgment in memory's us-report-fundamentals-agents-worth-waiting: do not skip them before finalizing. Where other agents did not return before publication, the corresponding sections are explicitly marked "not obtained" rather than left blank.

11. The first draft's most instructive error (methodology, worth retaining): I formed the narrative "sold off after a beat = crowding being flushed out" for COHR first, then went looking for numbers to support it, and ended up citing four figures that were wrong or on the wrong convention — and those four errors were mutually consistent and reinforcing, so the whole text read without a single visible flaw. What actually punctured it was not a logic check but one piece of elementary arithmetic: comparing the pre-market price against the last clean close before the report ($328.57) showed a net change of +0.7% rather than −6.68%. Lesson: the pricing impact of an after-hours earnings report must be based on the price at the moment before release, never the closing price of the release day — the closing price already contains the front-running. This belongs to the same family as memory's self-correction-becomes-exemption and summary-stat-vs-per-row-conflict.

---

⚠️ Risk warning: 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 after earnings there is IV crush and the possibility of guidance reversals; automatically generated content may contain stale information or factual errors. Company disclosures and SEC filings prevail; it must not be used directly as a basis for trading.