U.S. Market Recap | 2026-08-14 (ET) Friday

Coverage window: 2026-08-14 09:30–16:00 ET regular session, plus the 16:00–17:05 ET after-hours session.

Quote conventions:

• Open/high/low/close, volume, daily change, 52-week high/low and after-hours price for single stocks and ETFs come from the stockanalysis.com quote API, snapshot 17:00–17:05 ET, with the closing-price timestamp uniformly "Aug 14, 2026, 4:00 PM EDT" (regular-session close basis, excluding after-hours).
• Index levels (S&P 500 / Nasdaq 100 / Russell 2000 / VIX / VIX9D) come from CBOE delayed_quotes, with last_trade_time checked one by one to be 2026-08-14T16:09–16:15, confirming the data is not stale. The Nasdaq Composite comes from api.nasdaq.com. The Dow failed to pull from the local API, so a media-sourced figure is used and cross-checked by back-calculating from DIA (see the footnote in Section 1.1).
• Treasury yields come from the treasury.gov official daily yield-curve CSV, a primary source.
• The "vs. pre-market price" column: today's close compared with the pre-market price recorded in the 8/14 pre-market list (snapshot 08:00–08:11 ET). This column is the real result of "placing the orders pre-market off the list"; looking only at the daily change systematically overstates the hit rate — see Section 2.3.

Three verification actions in this report (each one changed a conclusion):

1. Of PAYC / VRSK / MET / GEN, which yesterday's list judged "data unavailable", three did in fact fall today. After checking each one: the falsification itself still stands (every value on that table has been proven wrong today), but the headline the list used in Section 6, "a non-existent negative", overreached. See Section 2.4 — this is the most important correction this report makes to yesterday.
2. The attribution for AMD's +6.50% has been replaced. The first attribution article found wrote it up as "$475 million…bond issuance + SMCI earnings EPS beat of 84% + CoreWeave guidance"; after checking: the SMCI and CoreWeave results are old news from around August 4, stitched into today's article. The genuinely new information today is Baird doubling its price target to a Street-high $1,250. See Section 3.3.
3. The single biggest gap in yesterday's list (DRAM/NAND prices) got a partial reading in this report, but it is still secondary + quarterly frequency + published on July 3, not a current spot price. It did not close the gap; it only drew the boundaries of the gap more clearly. See Section 3.2.

Data this report did not obtain: ① raw NYSE / Nasdaq advance-decline counts not obtained (the same gap as yesterday), with breadth proxied by "how many of the 11 SPDR sectors closed higher" + RSP/IWM vs SPY; that proxy cannot produce true advance-decline counts; ② options implied volatility (IV) not obtained, as VIX/VIX9D are index values, not single-stock IV; ③ for the declines in AVGO and ORCL this report could not obtain the original primary sell-side reports, so the related figures are media relays and are labeled "media-sourced" at every point of use; ④ current DRAM/NAND spot prices still not obtained (requires a paid source); ⑤ no single new catalyst specific to 8/14 could be located for ORCL, so this report honestly writes it up as "a continuation of the repricing of existing information" rather than a same-day event.

---

0. One-Line Recap

The one line most worth taking away today is: the index fell only 0.17%, but its insides had a blood transfusion — 8 of the 11 sectors closed higher, and the 3 that fell happen to hold every narrative the market has run for the past year.

Four things happened at once, all pointing the same way:

• Macro data weakened across the board: July retail sales came in at −0.6% month-over-month against an expected +0.1%, the first decline in nine months and the largest drop in over a year, with the control group posting its first negative reading since September 2025; the August University of Michigan consumer sentiment preliminary print was 51.0 against an expected 54.5 and a prior 55.2.
• Yet rates did not fall — they gave back all of yesterday's move: the 10Y rose from 4.63% back to 4.68% (+5bp), drawing a perfect circle back to exactly the 4.68% of 8/12. The cause is not growth but inflation expectations and oil: Michigan 1-year inflation expectations rose from 4.2% to 4.3%, while the U.S. threatened "economic isolation" of Iran and said a blockade of Iranian ports could last "indefinitely"; Brent +1.7% to $88.52 and WTI +1.4% to $82.40, both up more than 5% on the week.
• So capital did something very clean: sell duration, buy physical. Energy XLE +1.39% was the day's best (only 2.4% from its 52-week high), with utilities +0.61%, materials +0.44% and industrials +0.39% following; while software IGV −2.07% was dead last, tech XLK −0.40%, healthcare XLV −0.60%.
• At the same time, a theme entirely absent from yesterday's list surfaced: the market has begun to price "how AI expansion is financed." AVGO −5.94% (media-sourced: Bank of America estimates its chip-financing vehicle could carry senior debt of $370 billion by mid-2029; compounded by reports that the VMware vCenter vulnerability is being actively exploited), ORCL −3.65% (already −56.5% from its 52-week high). And on the same day, AMD rose +6.50% because it successfully sold $4.75 billion of bonds.

Put these four together and you get today's real conclusion: yesterday's list framed things as "the book profits of AI hardware look less and less like cash" (the inventories and receivables of CSCO/COHR/AMAT); today the market pushed that one notch further — "AI expansion increasingly runs on borrowed money." And it has already started discriminating: AMD got the money, and at a spread 25bp tighter, +6.5%; AVGO was calculated to need $370 billion of future borrowing, −5.94%; ORCL has already borrowed its way to the last rung of investment grade and sits −56.5% from its high. What separates them is not "borrow or not," it is "does the market believe you can pay it back."

The reconciliation of the pre-market list: 17/21 correct on direction (81.0%), but only 11/21 (52.4%) when executed at the pre-market price. That 29-percentage-point gap is not luck, it is structural — the biggest gainers on the list had already finished gaining before 08:11 ET. See Section 2.3.

Of the list's three checkable judgments, two were right and one asked the right question but got the bad answer:

1. ✅ AMAT's decline is multiple compression rather than fundamental deterioration, verification point $531.79 — the high for the day was only $523.00, never approaching that level, closing at $507.18 (−5.12%). Called correctly, and the decline is not over.
2. ⚠️ AMAT's guidance did not spill over to equipment peers, verification point being whether LRCX/KLAC follow higher — all weakened: LRCX −1.38%, KLAC −2.70%, ASML −0.21%, TSM −0.96%. The list itself wrote "if they weaken together, that is a far more serious problem than one stock called AMAT." The right question, with the bad answer.
3. ✅ All the elasticity in the memory chain is anchored to ASP — today all four memory names rose (SNDK +7.39%, STX +5.65%, WDC +4.41%, MU +2.30%), while the only price reading obtainable on the day says price increases are decelerating (TrendForce: 3Q26 DRAM contract prices +13–18% QoQ, NAND +10–15%, versus more than +100% in 1H26). That divergence is the single most worth-watching item for next week.

Tone into the next session: VIX 14.25 (−2.60%), but VIX9D only 10.61 (−6.68%) — front-end volatility is 3.6 points below the 30-day, meaning the market has priced literally no event premium into next week. And next week holds the FOMC minutes (Wednesday), the RDDT index inclusion taking effect (Tuesday), and four retail earnings reports — HD/LOW/TGT/WMT — colliding with a retail sales print that just turned negative. In that combination, what is cheap is not the direction, it is volatility itself.

---

1. Market Overview

1.1 Indices

⚠️ Note on the Dow's basis: CBOE's _DJI endpoint returned a last_trade_time of 2026-07-31 today (two-week-stale dead data, though the price fields were filled in to look normal), and this report refuses to use it. The Dow figures in the table are media-sourced and have been cross-checked by back-calculating from DIA (close 536.80, −0.21%): 536.80 × 100 ≈ 53,680, consistent with 53,732; and the 8/13 close of 53,839.99 − 107.58 = 53,732.41, internally consistent.

⚠️ The S&P's 7,800 level, a second day without holding it: today it opened at 7,806.60, above 7,800, made an intraday high of 7,810.01, and closed back down at 7,785.76. Together with 8/13 (intraday high 7,816.70, close 7,798.99), this is the second consecutive day of trading above 7,800 without closing above it.

This week: SPY went from the 8/07 close of 773.26 → the 8/14 close of 776.34, +0.40% on the week, which media report as a third consecutive up week (this report did not independently verify the earlier weekly sequence; that "third week" is media-sourced).

1.2 Market Breadth — Today Was a Classic "Index Down, Stocks Up"

⚠️ Raw NYSE / Nasdaq advance-decline counts were not obtained by this report (the same gap as yesterday). The three proxy readings below are highly consistent in direction, but they are still not advance-decline counts — please use them as proxies.

How to read it: yesterday's list used "5bp is within the noise range" in Section 1.3 to reject the breadth inference from RSP vs SPY, and the same yardstick must be applied today — today's 22bp on RSP vs SPY is still small, and on its own it is likewise insufficient to characterize the day. But IWM's 72bp and "8 of 11 sectors higher" are of a different order of magnitude, and all three point the same way, so this report's characterization is: today's index decline came almost entirely from the mega-cap weights, not from a broad decline. This was a structural, not a broad-based, pullback. For exactly which weights, see Section 5.1.

1.3 Cross-Asset

1.4 Treasuries — Yesterday's Downshift Was Repaid Today to the Basis Point

(treasury.gov official daily yield-curve CSV, primary source)

This table is the cleanest signal of the day: the 10Y went 4.68% → 4.63% → 4.68%, a complete circle in two days. Yesterday's entire curve downshift driven by a softer PPI (−3 to −6bp) was fully erased today in the face of a retail sales print far weaker than that PPI.

Why do weak data come with rising yields? This report's judgment rests on three parallel reasons, the third of which is directly supported by this report's own data:

1. Inflation expectations are turning up: Michigan 1-year inflation expectations 4.2% → 4.3%, with the 5-year holding at 3.3%.
2. Oil is rising: Brent +1.7% on the day and more than +5% on the week, and it feeds simultaneously into the energy component of CPI and into the formation of inflation expectations.
3. Yesterday's downshift was itself excessive: the 5Y fell 6bp in a single day yesterday, bought with a PPI print of "0.0% month-over-month." Today amounts to pushing that pricing back into place, not adding a new round of tightening pricing. Supporting this is the fact that the 2Y rose only 2bp — if the market were truly repricing rate hikes, the front end would not be the least-moved part of the whole curve.

⚠️ One disagreement that must be disclosed: intraday today there were media headlines reading "Treasuries Rise as Weak Retail Sales Dampen Fed Rate-Hike Expectations" (bond prices up = yields down). That is the opposite direction from the official closing curve (up across the board). This report's handling: treasury.gov's closing primary data governs, and those headlines are taken to describe the immediate reaction right after the data release, a reaction that was reversed intraday. Yesterday's list wrote in Section 8.1 that "the price reaction in the first 60 seconds after a data release is not sufficient to represent pricing" — today that sentence was validated in full in the bond market.

1.5 Actual Macro Prints (Not Yet Released When Yesterday's List Went Out)

Components: motor vehicles and parts dealers −1.8% (the largest decline), non-store retail (including online) −2.2%; apparel +1.9%, health and personal care +0.7%, food services +0.5%. ⚠️ Note the structure: what fell was autos and online (big-ticket discretionary + price-sensitive), what rose was apparel, pharmacy and dining (small-ticket staples). This is not "consumption collapsing," it is "trading down" — and the two have completely different implications for rates and for retail stocks. Next week's four retail earnings reports will test this directly (see Section 6.2).

The Michigan survey's attribution (per the report's own wording): consumers turned worse because of war, higher bond yields and geopolitical uncertainty; short-run business expectations fell 11% and long-run expectations fell 17%; only 8% of consumers expect income growth to outpace prices; declines were especially large among older, lower-income and non-college groups.

⚠️ Yesterday's list's scenario map got both legs wrong today — that is worth logging separately. Section 8.1 of the list read: "control group ≤ 0% (weak): lower rates favor long-duration assets, but it raises a second-order worry about weakening consumption, which is usually bad rather than good for small caps (IWM)." The actual result: the control group did turn negative, but rates rose (10Y +5bp), long-duration assets fell hard (IGV −2.07%), and small caps rose (IWM +0.52%). All three inverted.

Where it went wrong: that scenario map assumed "growth data is the only independent variable," while today's real independent variable was oil and inflation expectations (i.e., the supply side). When the shock comes from the supply side, the transmission chain "weak growth → lower rates" is itself broken. This is the most important lesson this report leaves for the next scenario map: before writing "weak data → lower rates," take a look at where oil and inflation expectations are.

---

2. Pre-Market List Reconciliation

Note: "pre-market price / pre-market %" are taken from the 8/14 pre-market list (snapshot 08:00–08:11 ET). "vs. pre-market %" = today's close ÷ that pre-market price − 1, the real result of placing the orders pre-market off the list. The difference between the two columns is the core of this section — see Section 2.3.

2.1 Long / Watch Direction (15 names)

2.2 Short / Avoid Direction (6 names)

2.3 ⚠️ Hit Rate — Two Numbers, and the 29 Percentage Points Between Them

That 29-percentage-point gap is not luck; it has a specific, identifiable cause: for the biggest gainers on the list, the gain had already finished happening before 08:11 ET.

Two of the six (CAPR, MU) had pre-market contributions above 100%, i.e. they were net losses after the open.

The conclusion, and it holds for every pre-market list: a pre-market list's "daily-change hit rate" is inherently overstated, because it books to its own account a stretch of price movement you could not participate in at all. The one judgment in yesterday's list that truly created excess return was not any of its long picks, but its discipline that "avoid everything gapping more than +10% this morning" — that discipline avoided 19.6% on CAPR alone, more than the sum of all 15 long names' returns against the pre-market price for the day.

A hard requirement this report sets for the next list: the "vs. pre-market price" column must be permanent, and the hit rate must always be reported as two numbers. Reporting only the directional hit rate amounts to handing yourself a certificate of merit.

2.4 ⚠️ Correction to Yesterday's Section 1.4: The Falsification Stands, but the Headline Overreached

Yesterday's list spent an entire section (1.4) falsifying investing.com's pre-market decliners table, and in the avoid table in Section 6 lumped PAYC / VRSK / MET / GEN into one row under the headline "a non-existent negative." Today's actual results:

Two conclusions, and please note that their strength is completely different:

1. [On falsification: fully valid, needs no correction] Not one of that table's four values was proven right today. MET's "−2.92%" corresponds to an actual +0.19%; GEN's "−2.89%" actually fell 3.69% (right direction but still the wrong value); PAYC's "−5.44%" was actually −2.46%. The judgment that "that table's prices cannot be reproduced on any known basis and are unusable" was confirmed today by all four samples. And the mechanism that saw through it (pre-market volumes of 356/123/9/126 shares) remains the most valuable technical move of the week.

2. [But the headline overreached, and this is an error that must be owned] The phrase "a non-existent negative" quietly swapped "this data is wrong" for "this thing will not happen." Three of the four did fall on the day.

So why did they fall? This report's finding: the reason has nothing to do with that table — they were dragged down by today's selloff in software/services. The evidence is the cross-section: today IGV −2.07%, PAYC (payroll SaaS) −2.46%, VRSK (data analytics services) −2.30%, GEN (consumer software) −3.69% — all three are software/information services, with declines of the same order as IGV; while the only one that did not fall, MET, is an insurer, not on that chain, closing +0.19%. Sector membership explains 4 of 4 perfectly, while "there was a pre-market negative" cannot explain MET.

The lesson, and it is more general than yesterday's: falsifying a data source only gets you "this data is unusable," it does not also get you "this direction will not happen." The former is a judgment about information, the latter a judgment about the world, and each needs its own independent evidence. Yesterday's list put it very precisely in the body of Section 1.4 ("that table's readings are unusable"), yet in the table headline of Section 6 wrote it as "a non-existent negative" — when two statements inside one report differ in strength, readers take the stronger one as the conclusion.

2.5 ⚠️ The 50,000-Share Volume Threshold: Today It Both Earned Its Keep and Failed Twice

The methodological innovation yesterday's list was proudest of was "for a stock with pre-market volume < 50,000 shares, its pre-market change is not treated as a signal," together with the requirement to "treat both sides, up and down, alike." Today this rule's report card is split:

✅ What it got right: for the software sector (MDB/MNDY/DDOG/SNOW/NET/CRM all had pre-market volume below 50,000 shares), the roughly −0.5% readings at the time genuinely carried no information — but today's actual declines were MNDY −7.18%, NET −4.55%, CRM −2.56%, SNOW −2.51%, MDB −2.53%, 5–9× the magnitude of the pre-market readings. The list said "software this morning neither gave back nor extended, it simply has not started trading yet," and as a description of the information content at that moment, that sentence is precisely correct.

❌ The two times it got it wrong, and both wrong in the same place:

In both cases the low-volume pre-market reading understated the real move rather than overstating it. This matters, because yesterday's list built the case for this rule on "low-volume readings manufacture false signals (such as that decliners table)" — but today's two samples show the other failure mode of a low-volume reading is "the real signal has not been traded out yet," and the list applied the same treatment (downgrade to 'watch only') to both cases.

The correct dividing line should be the news itself, not the volume:

• ETON had its own verifiable primary catalyst: Q2 revenue $37.59 million (+99% YoY, versus an expected $27.66 million), EPS $0.35 (versus an expected $0.12), full-year revenue guidance raised to >$145 million, adjusted EBITDA $16.20 million vs $3.10 million a year earlier (43% margin vs 16%). The list itself wrote in Section 3.1 that "Q2 EPS beat by $0.28, full-year revenue guidance raised," yet downgraded it because of 13,000 shares of pre-market volume — this is letting volume veto news.
• STX is the opposite: this report could not locate a primary company announcement dated 8/14; the attribution articles seen cite "FY26 revenue $12.2 billion (+34%), FQ1'27 guidance of $4.0–4.2 billion revenue / $7.10–7.50 EPS," which is content already disclosed in late July, stitched into today's article (the same category as the AMD case in Section 3.3 of this report). STX's rise today can only be attributed by this report to spillover along the memory/HDD chain + sell-side price target hikes (media-sourced: several raised into the $900–1,400 range).

A revision suggested for the next list (not a repeal, but adding a precondition): the volume threshold should only be used to veto "pre-market moves with no news behind them," not to veto "pre-market moves with a primary catalyst." The order of judgment is news first, volume second — not volume first. On ETON, the list already held primary, 8-K-grade earnings figures in hand, and let 13,000 shares of volume override them.

---

3. Theme Verification

3.1 Theme Scorecard

3.2 Memory: The Theme Was Called Right, but the Day's Only Price Reading Is Going the Other Way

Actual performance today (all on a regular-session close basis):

The list ranked memory #1, and that is the single most correct judgment of the day — all four rose, and they were near the top of the day's gainers. But three details must be recorded:

1. The intraday shape in every case was "kill the gap first, then take it back," not gap-and-go. Section 8.3 of the list set the criterion "if more than half the gap is filled in the first 30 minutes, yesterday's investor-day move is already being distributed." In fact: SNDK's low filled 65% of the gap (pre-market +6.91% → low +2.41%), and WDC filled 101% (+4.10% → −0.04%, i.e. the gap was completely filled and it briefly traded below the prior close). By the list's own criterion, both triggered the "distribution under way" signal; they simply took it back into the close. This report's judgment: this was a two-way day with both absorption and distribution, not a straightforward continuation of strength.
2. The leadership changed hands. Yesterday's and this morning's leader was SNDK (which started the investor day), but against the pre-market price, today's strongest was STX (+3.01%), with SNDK only +0.45%. The initiator underperforming the followers is a classic marker of a theme rotating from "event-driven" into "sector rotation" — and usually the first signal that the event premium is starting to decay.
3. MU is the only one of the four to close below its pre-market price, entirely consistent with the list's judgment of "weakest of the four memory names, pure sector beta, no catalyst of its own." The list called this one very well.

⚠️ Yesterday's Biggest Gap: This Report Got a Partial Reading, but It Did Not Close the Gap

Yesterday's list declared in four places (the header, Section 2, Section 7.1 and Section 9.3) that "current DRAM/NAND spot and contract prices were not obtained; this is the largest gap in the report, and until a reading is in hand neither bulls nor bears have evidence." The readings this report obtained are:

TrendForce gives two reasons for the deceleration (per its own wording): ① "record contract prices mean customers in consumer markets such as PCs and smartphones are hitting the ceiling of what they can absorb"; ② a high base effect. It also calls 3Q26's increase "a noticeably slower pace than in previous quarters."

This report must handle this reading with great restraint, for three reasons:

• It is secondary (TrendForce is a research house, not an exchange or a manufacturer);
• It is quarterly frequency, while today's stock prices are daily;
• It was published on July 3, more than six weeks ago — and in a market where prices rose more than 100% in a quarter, six weeks is a long time.

This report's conclusion is therefore a conditional sentence rather than a judgment: today all four memory names rose, while the only price information obtainable on the day says price increases are decelerating. There are two possibilities for that divergence — ① the market knows something not contained in that July 3 forecast (for example, more recent contract negotiation results); ② the market is continuing to pay up for a variable that is already decelerating. This report cannot distinguish between them, because distinguishing them requires exactly the current spot price that has still not been obtained.

But one thing can be said with certainty today: yesterday's judgment that "all of memory's elasticity is anchored to ASP" was neither falsified nor confirmed today — another day simply passed without obtaining the independent variable. That is now two days running, and over those two days the memory group has risen by double digits.

3.3 🆕 The Most Important Surprise Theme of the Day: The Market Has Begun to Price "What Finances AI"

None of the 7 themes in yesterday's list covered this, and today virtually all of the decline in the mega-cap weights came from it.

⚠️ An attribution correction that must be disclosed (this report's second verification action): the AMD attribution article this report saw first wrote the rise up as "the bond offering + SMCI earnings EPS beating by 84% + CoreWeave sales guidance beating expectations." After checking: those two earnings reports from SMCI and CoreWeave date from around August 4 and have nothing to do with 8/14 — they were stitched into today's article. And a single $4.75 billion bond offering (about 0.6% of AMD's market cap) can hardly explain a +6.50% move either. The genuinely new information today is Baird's report doubling the price target. This is the same kind of operation as yesterday's list's attribution falsification on MU (treating a $250 million venture fund = 0.023% of market cap as the cause of a +4.23% move).

Only by putting all three together can you see what actually happened today:

The market priced AMD's debt and AVGO's debt in exactly opposite ways. AMD borrowed $4.75 billion at a spread 25bp tighter, and the stock rose +6.50%; AVGO had a sell-side report calculate that it "must borrow $370 billion over the next four years," and the stock fell −5.94%; ORCL has already taken this road to the edge of investment grade in the ratings, sitting −56.5% from its 52-week high.

What distinguishes them is not "borrow or not," it is three variables: how much (relative to its own cash flow), what for (whether there is contracted revenue to match), and on whose balance sheet the borrowing risk sits. The reason that $370 billion figure for AVGO is so damaging is not its size, but that it points to an off-balance-sheet financing vehicle — that is, the location of the risk is itself the problem, not just the scale.

This theme connects to yesterday's framework, and is the next notch along it. Yesterday's list distilled from three samples — CSCO (inventory +80%), COHR (inventory +79.5%), AMAT (receivables +33.2%, DSO 71.9→76.8 days) — the idea that "the book profits of AI hardware look less and less like cash." Today the market pushed the same thing a step further: if profits cannot become cash, and capex still has to double, then the difference can only come from debt. So the market shifted from "auditing profit quality" to "auditing financing structure." Yesterday's list was already very close to this layer — in its AMAT section it wrote "9M buybacks −71% YoY, shareholder returns only 65% of FCF, below the company's stated 80–100% target" — but it left that observation at the single-stock level and did not elevate it into a theme. That is the omission most worth recording today.

3.4 🆕 Energy and Geopolitics: The List Explicitly Said "Not Macro," and Today's Strongest Sector Came from Macro

The exact words in Section 0 of yesterday's list were: "the driver type is 'earnings + guidance + index events,' not macro — though there is a first-tier event today at 08:30 ET (retail sales)." It foresaw retail sales, but did not put geopolitics/energy in the field of view at all.

Trigger events: the U.S. said a naval blockade of Iranian ports could last "indefinitely," reigniting concerns about energy transit through the Strait of Hormuz; Treasury Secretary Bessent warned in an interview of measures "never seen before" aimed at the "economic isolation" of Iran. At the same time, some institutions estimate the global crude market could face a supply shortfall of 1.8 million barrels per day this quarter, more than double the prior forecast.

Why this matters more for next week than for today: oil is the explanatory variable behind today's "weak data + higher rates" paradox (see Section 1.4), and it simultaneously appears in the Michigan survey's attribution for worsening consumers ("war" is listed among the leading factors). In other words, energy is no longer just a sector; it is at once an input to rates, an input to inflation expectations, and an input to consumer confidence. A variable that can drive three chains at once should not be absent from a pre-market list.

An input for the next list: even when the day's driver type is judged to be "earnings/event-driven," crude oil and geopolitics must be listed as a standing check item — because it does not work through sectors, it works through rates and inflation expectations.

3.5 Software: The List's Process Judgment Was Right, Its Directional Conclusion Was Wrong

Note the volume markers in the last three rows: IGV, PLTR and NOW all cleared the pre-market volume bar (≥50,000 shares), with readings of −0.08%, −0.40% and −0.05% respectively, versus actual closes of −2.07%, −2.78% and −2.55%. In other words, even a pre-market reading that clears the volume bar understated today by 7–51×.

This deserves to be spelled out separately, because it corrects an implicit assumption in yesterday's list's methodology: clearing the volume bar only means "this reading is not noise," it does not mean "this reading is a preview of the day." Yesterday's list's wording in Section 1.5 was actually accurate — "software this morning neither gave back nor extended, it simply has not started trading yet" — and as a description of information content that is completely correct. But what it wrote immediately after, "any claim of a 'software one-day wonder' had no evidentiary support this morning," had evidence by the close.

The distance between "no evidence pre-market" and "this thing will not happen" showed up twice today as the same error — in software and in PAYC/VRSK/GEN (Section 2.4). Taken together, those two constitute this report's most substantive criticism of yesterday's list.

The reason software faded is clear, and entirely consistent with the rates story in Section 1.4: yesterday's IGV +3.10% was driven by the long-duration asset re-rating that a softer PPI brought (10Y −5bp); today the 10Y handed back those 5bp exactly, and IGV handed back 2.07%. The round trip in long-duration assets against rates was symmetric and explicable, and requires no fundamental change to explain it.

3.6 Semiconductor Equipment: The List's Most Important Verification Point Landed on the Bad Branch

Section 8.3 of yesterday's list called this "the single observation this report most recommends," and pre-wrote what each branch would mean:

"Whether LRCX / KLAC / ASML strengthen after the open in line with AMAT's raised guidance. If they strengthen → AMAT's decline is purely a single-stock valuation issue and the industry logic is intact; if they weaken together → the market is de-rating the entire WFE group, which is far more serious than a problem in one stock called AMAT."

The answer: all weakened.

Three conclusions:

1. AMAT's verification point of $531.79 was never touched. The high for the day was $523.00, 1.65% below that level; the opening price of $499.40 was already 6.1% below the 8/13 intraday low, and it traded below that line from start to finish. By the list's own criterion, the branch "the post-earnings repricing is still under way" is clearly activated. Cumulatively from the 8/12 close of $548.15, AMAT is −7.47%; it is −31.4% from its 52-week high of $739.67; and it is still about 31% above its 200-day moving average of $387.30. The list's line that "strong fundamentals are not a reason for the decline to be over" is right today.

2. KLAC was the hardest hit of the group (−2.70%), and it was +0.02% pre-market this morning. This stock had no news of its own, so its decline can only be explained by sector multiple compression. That is precisely the bad branch the list laid out.

3. But this conclusion needs a qualifier, otherwise it gets written too heavily: SMH fell only 0.22% today, and SOXX only 0.06% — semiconductors as a whole were flat, while the four memory names were up sharply. So the accurate statement is not "semiconductors are de-rating," it is "the WFE (front-end equipment) sub-sector is de-rating while memory rises." Today's divergence within semis matters far more than the sector's own direction: the equipment sellers fell and the chip sellers rose — consistent with the logic yesterday's list corrected (equipment revenue is a leading indicator of the supply side, not confirmation of the demand side). If the market really is pricing "future supply will increase," then selling equipment first and buying memory second is exactly the right trading order for that logic. This report labels this an observational hypothesis, not a conclusion — one day of data is not enough to confirm a sequence.

---

4. After-Hours Earnings Moves

4.1 ⚠️ After-Hours Was Empty Today — That Is Itself a Piece of Information

Check result: after 16:00 ET on 8/14, no S&P 500 constituent or major tech name reported earnings. This is the Friday at the tail end of earnings season and is normal. After-hours moves in the four broad benchmarks were all within noise: SPY −0.05%, QQQ +0.02%, DIA −0.01%, IWM −0.13%.

The after-hours gainers/losers leaderboard (as of 17:05 ET) is entirely occupied by micro caps, none with a verifiable primary catalyst, and this report does not recommend it for any purpose:

⚠️ Every name in this table is priced between $0.34 and $7.25, mostly micro caps/penny stocks. At that market cap and price scale, after-hours percentages typically have no relationship to "news." They are listed for completeness only, and this report offers no judgment on any of them.

4.2 The Only Meaningful After-Hours Move: IMXI

This is the only non-micro-cap name with an after-hours move above 1% and substantive background. Section 5.6 of yesterday's list judged that "the market has priced only half of this news" (NYDFS approval is a positive, but California's DFPI suspended the approval extension on 8/13, a hidden negative), and cut the expectation-gap score from 8 pts to 0 pts on the grounds that "that discount is precisely the market's pricing of the California risk."

Today's price action gives a partial answer to that judgment: the discount narrowed all the way from 11.2% pre-market to 5.8% after hours, indicating the market is marking down its estimate of the California risk. But this report obtained no new information about any change in the California DFPI's position, and therefore cannot judge whether this is convergence based on new information or simply arbitrage flows. This is the first thing worth verifying before Monday's open (see Section 6.3).

After-hours moves in the rest of the pre-market list's names were all within ±0.5% and carry no information: AMAT −0.07%, SNDK +0.25%, MU −0.08%, WDC −0.06%, STX +0.11%, RDDT −0.05%, CAPR −0.90%, GLOB −0.65%, INTC +0.50%.

---

5. Flows and Sentiment

5.1 Sector Rotation — Today Was a Very Clean "Sell Duration, Buy Physical"

How to read it (three points, in descending order of evidentiary strength):

1. [Strongest] Of the 8 sectors that closed higher, 6 are physical-asset / short-duration / defensive (energy, utilities, materials, industrials, real estate, staples); the 3 that fell are the three longest-duration groups in the whole market (software, tech, healthcare). That stratification maps exactly onto 10Y +5bp, and requires no other explanation.

2. [Moderate] Of yesterday's two lead themes, one reversed today and one held. Yesterday IGV +3.10% → today −2.07% (reversed, giving back two-thirds); yesterday XLC +2.07% → today +0.36% (held, but momentum decayed sharply, with NFLX going from +5.43% to zero at −0.10%). The SMH vs IGV divergence (the second verification point in Section 8.3 of yesterday's list) went from +2.37pct yesterday to −1.85pct today, a complete reversal. The rotation is real, and its direction is from software into semiconductors.

3. [Weakest, observation only] XBI +0.35% is a bounce after yesterday's −1.59%, too small to characterize; XLF −0.17% is in line with the index and carries no information.

5.2 Volatility — The Cheapest Thing Today Was Not Stocks, It Was Volatility

✅ VIX9D, which yesterday's list marked as "data pull failed, cannot assess," was obtained today. Yesterday CBOE's _VIX9D endpoint silently returned 0.0, and the list honestly wrote "cannot assess whether front-end volatility still carries no event premium; yesterday's value will not be carried forward" — and that handling has been proven right today: today's reading gives an 8/13 prior close of 11.37, whereas the VIX9D recorded in the 8/12 list was 11.09. They are not the same, and carrying the old value forward would have produced a wrong conclusion.

Today's readings say something important: VIX9D at 10.61 is 3.64 points below VIX at 14.25, and that inversion is still widening (−3.26 points yesterday). Front-end volatility has been pressed down to 10.6, meaning the options market has priced almost no event risk into the next 9 trading days.

And the next 9 trading days contain:

• 8/18 (Tuesday): RDDT's S&P 500 inclusion takes effect, forcing index-fund trading
• 8/19 (Wednesday): the FOMC meeting minutes
• 8/18–8/20: four retail earnings reports — HD, LOW, TGT, WMT — colliding with a retail sales print that just turned negative
• 8/28: Jackson Hole (outside the 9-day window, but it will enter pricing in advance)

This report offers no directional recommendation on volatility, but points out a factual asymmetry: in front of a VIX9D of 10.6, if any of the above events produces a surprise, the upside in front-end volatility is far greater than the downside — because there is almost no downside left. This is an observation about the shape of the distribution, not a judgment about direction.

5.3 Risk-On / Risk-Off Characterization

This report's judgment: today was neither risk-on nor risk-off, but a "duration clear-out." The reason is that these four signals contradict each other and cannot be explained on a single risk-appetite axis:

What unifies them is not risk appetite, it is the discount rate: what rose today was short duration and physical assets, what fell was long duration, and the 10Y rose 5bp. That VIX fell instead of rising is exactly the point — the market did not treat this adjustment as a risk event. It was an orderly repricing sorted by duration, not a panic.

⚠️ But this characterization has one clear soft spot that must be written down: it rests on the magnitude "10Y +5bp," and 5bp is a very small number. Using a 5bp rate move to explain a sector move of IGV −2.07% implies an unnaturally high elasticity. This report retains the explanation because the stratification itself (8 short-duration sectors up, 3 long-duration sectors down) carries far more information than the magnitude; but if rates fall next week and software does not bounce, this explanation should be overturned, and the alternative hypothesis "software is pricing its own fundamentals" would then need to be considered (next week's ADI/TJX and other earnings will provide samples). This is a falsifiable point this report leaves for the next one.

---

6. Next-Day Outlook (Next Trading Day = Monday 8/17)

6.1 Theme Continuation Ratings

6.2 Next Week's Calendar (8/17–8/21)

⚠️ Monday (8/17) is itself empty — no important earnings, and only second-tier macro data. The real density is Tuesday through Thursday.

Three linkages this report thinks are most worth thinking through in advance:

1. [Retail earnings × retail sales] Today's retail sales component structure is "autos −1.8% and online −2.2% down; apparel +1.9% and dining +0.5% up" — i.e. big-ticket discretionary weakening while small-ticket staples hold up. The implication for next week's four reports is split: WMT (mainly staples) and TJX/ROST (off-price retail) should in theory benefit from trading down, while HD/LOW (big-ticket discretionary + rate-sensitive housing-related) should in theory suffer. Verification point: if HD confirms weakness in big-ticket discretionary on Tuesday and WMT confirms resilience in staples on Thursday, then today's "trading down" reading is confirmed, and the XLP vs XLY divergence becomes one of next week's lead themes.
2. [FOMC minutes × inflation expectations] Today Michigan 1-year inflation expectations rose to 4.3% and oil is up more than 5% on the week, while current pricing for the September FOMC is "about 71% hold / about 29% a 25bp hike" (that pricing is from yesterday's list; this report did not re-pull it, so please treat it as pending verification). If Wednesday's minutes show the committee more focused on inflation than the market expects, then today's 5bp rate rebound is not a give-back, it is a beginning.
3. [VIX9D 10.61 × all of the above] See Section 5.2. Front-end volatility prices these three days at close to zero.

6.3 Focus List (Ticker + Verification Point)

6.4 What to Avoid

1. "Yesterday's best-performing sector" — this pattern has been punished repeatedly this week: 8/13 software +3.10% → 8/14 −2.07%; 8/13 XLC +2.07% → 8/14 +0.36% (momentum decay). This report does not elevate it to a rule (n=2, and there is no control group), recording it only as a pattern to watch.
2. Anything that has already filled more than half of today's gap — WDC (101% filled, completely closed), SNDK (65% filled). They took it back into the close today, but distribution already showed up intraday.
3. CAPR — it fell from +96.44% pre-market to +57.96% at the close today (−19.59% against the pre-market price). The premium on a binary event is decaying fast, and an extended PDUFA date means the next catalyst has been pushed back. The position is worse than yesterday, not better.
4. Shorting an already sharply lower name at the pre-market price — GLOB taught the lesson today: shorting at the pre-market −12.10% would have lost 3.75% by the close. ZNTL is the same (+1.71%). On the day a negative materializes, the pre-market print is often near the low.
5. ORCL / highly leveraged AI infrastructure — being −56.5% from the 52-week high looks like a deep decline, but what is driving it is the credit rating (S&P BBB−, the last rung of investment grade), and rating downgrades come in steps, not continuously. Until the ratings path is clear, "it has fallen a lot" is not a reason to buy.
6. Any position built on the chain "weak data → rate cuts → buy growth" — that chain broke in live trading today (see Section 1.5). Do not restart this mapping until oil and inflation expectations come down.
7. The after-hours gainers/losers leaderboard for micro caps (see Section 4.1) — the same category of noise as yesterday's pre-market decliners table.

6.5 Inputs for the Next Pre-Market List

1. ⭐ The "vs. pre-market price" column must be permanent, and the hit rate must always be reported as two numbers. Today directional hits were 81.0% versus 52.4% executed as listed, a gap of 29 percentage points (see Section 2.3). Reporting only the former amounts to handing yourself a certificate of merit.
2. ⭐ The order in which the volume threshold is applied must change: news first, volume second. Volume should only veto "a pre-market move with no news behind it," not "a pre-market move with a primary catalyst." On ETON today, the 8-K figures in hand and the 13,000 shares of volume pointed in opposite directions; the list chose volume and missed +16.44% (see Section 2.5).
3. ⭐ "Data unavailable" and "the event will not happen" must be written separately, and the strength of the wording must be consistent throughout. That error appeared twice today (the "non-existent negative" on PAYC/VRSK/GEN; the "no evidentiary support" on software). It was written correctly in the body and overwritten in the headline — and what readers remember is the headline.
4. Crude oil and geopolitics must become a standing check item, even when the day's driver is judged to be "earnings/event-driven." It does not work through sectors, it works through rates and inflation expectations (see Section 3.4).
5. Elevate "AI financing structure" to a standing theme. Yesterday's list had already collected three samples of "profits not becoming cash" (CSCO/COHR/AMAT), yet left them at the single-stock level. Today the market elevated it into a theme by itself (AVGO/ORCL/AMD). Recommend the next report build a standing table: net debt, capex/operating cash flow ratios, off-balance-sheet financing vehicles, and rating changes for AI-related companies.
6. The DRAM/NAND price channel is still not established, now for a second straight day. The TrendForce data this report obtained is secondary + quarterly + published six weeks ago, insufficient to support any conclusion. Until a stable channel exists, every piece on the memory theme can only write conditional conclusions — and over these two days memory rose by double digits. The cost of this gap is rising.
7. Four explicit, checkable judgments were made today; the next report should verify them one by one: ① AMAT's new key level is $497.10 (the old $531.79 is invalidated); ② whether KLAC keeps falling with no news, which decides whether "WFE sector de-rating" holds; ③ if the 10Y falls back and IGV does not bounce, the rates explanation in Section 5.3 is overturned; ④ the reason for IMXI's narrowing discount is unknown, and the California DFPI must be checked on Monday.

---

Data-Pull Failures and Implementation Notes (Internal)

Channel status:

1. yfinance was "half usable" today, unlike the full rate-limiting of recent days. The first single-ticker call (SPY) succeeded, then:
   • Batch yf.download() (34 symbols, threads=True) → immediate YFRateLimitError, all 34 failed, only ^GSPC came back by luck.
   • Switching to sequential single-ticker + 1.2s intervals → 6 of 12 succeeded (^RUT/^VIX/^TNX/^FVX/^TYX/SPY), 6 failed.
   • Switching again to sequential + 3.5s intervals + 4 retries (6s each) → a background job over 80 symbols was launched, and the very first, ^IXIC, failed all 4 retries, so that path was judged unusable and actively killed in favor of stockanalysis.
   • Conclusion: yfinance's rate limiting triggers on request rate, and the batch endpoint triggers it fastest. If it must be used next time, it has to be sequential + ≥3s intervals, and it cannot be the primary channel.
   • ⚠️ One more trap: the Volume field returned for ^RUT is the S&P's volume (4,574,170,000), not the Russell's own. The volume field is unusable for index symbols.
2. This run's primary channels (recommend making them permanent):
   • stockanalysis.com /api/quotes/s/<sym> — 77/77 all succeeded, zero failures, with 0.4s intervals being sufficient. The key fields for this after-hours recap are ep/ecp/es/eu (after-hours price/change/session marker/timestamp), used in Section 4. Note that epv (pre-market volume) returns null during the after-hours session; it only has a value during the pre-market session — the mirror image of the noted "CNBC extended-hours fields are unavailable pre-market": stockanalysis's pre-market volume field is unavailable after hours. Each source has half the capability in each of the two sessions.
   • CBOE cdn.cboe.com/api/global/delayed_quotes/quotes/_<SYM>.json — _SPX/_NDX/_RUT/_VIX usable with fresh timestamps (16:09–16:15). _VIX9D returned to normal today (returning 10.61) after returning 0.0 yesterday — this endpoint fails silently and intermittently, so the return value must be checked against 0 every single time.
   • api.nasdaq.com /api/quote/<SYM>/info?assetclass=index — COMP/NDX usable, giving a Nasdaq Composite of 26,729.16. /historical for pulling the weekly series to check the weekly change is usable.
   • treasury.gov daily yield-curve CSV — official primary source, the most stable, succeeded on the first try.
3. ⚠️ Hit CBOE's symbol-level stale data again (already noted, and hit again today):
   • _DJI returned last_trade_time = 2026-07-31T16:14:59 (two weeks stale), while the close/open/high/low fields were all filled in to look normal, and price_change_percent even gave +0.5276%. Looking only at the price fields would have put 52,485 into the report as today's Dow (actual about 53,732, a difference of 1,247 points).
   • _COMP is even worse: last_trade_time = 2021-03-23T14:33:09, five years stale, with a value of 13,344 (actual 26,729, off by a factor of two).
   • Handling: both refused; the Dow switched to a media-sourced figure with a DIA back-calculation cross-check, and the Nasdaq Composite switched to api.nasdaq.com. The Dow's sourcing has been disclosed to clients in the footnote to Section 1.1.
4. Failed channels:
   • WebFetch returned 403 for cnbc.com (twice: the Treasury yields article and the next-week outlook article) — so the primary narrative for the 10Y rise and next week's calendar had to come from search summaries.
   • WebFetch 403 for benzinga.com (BofA's $370 billion estimate for AVGO) and 403 for tipranks.com (AVGO attribution) — this is why all three drivers for AVGO in Section 3.3 are labeled "media-sourced"; the primary sell-side report was not obtained.
   • Raw NYSE/Nasdaq advance-decline counts: search came up empty, the same gap as yesterday. Honestly labeled in Section 1.2 and replaced with three proxy readings (with an explicit statement that the proxies are not advance-decline counts).
   • TrendForce current spot prices: still requires a paid source, only the quarterly forecast published 7/3 was obtained.

⚠️ Three traps caught this time:

1. ⭐ The AMD attribution article stitched in old earnings (the second hit on the noted "AI limit-up attribution articles stitch in old announcements"). The first attribution article found wrote AMD's +6.50% up as "bond offering + SMCI EPS beat of 84% + CoreWeave guidance beat." Those two SMCI and CoreWeave reports date from around August 4 and have nothing to do with 8/14. And it collapses on a cross-sectional check: if the driver were the AI server narrative of SMCI/CoreWeave, then AVGO — on the same chain — should not have fallen 5.94%, and SMH should not have fallen 0.22%. Chasing it down led to Baird's price target doubling to $1,250, the day's genuinely new information. Written into the body of Section 3.3.
2. The STX attribution article likewise stitched in an old quarterly report. "FY26 revenue $12.2 billion (+34%), FQ1'27 guidance EPS $7.10–7.50" is content already disclosed in late July (the same category as the noted "cumulative-period order figures are old news"). Handling: Section 2.5 honestly writes "this report could not locate a primary company announcement dated 8/14," and the attribution stops at "chain spillover + sell-side price target hikes (media-sourced)," without using the old quarterly report to fill the gap.
3. Bond-market headlines pointed the opposite way from official closing data. Intraday there were headlines reading "Treasuries Rise as Weak Retail Sales Dampen Rate-Hike Bets" (yields down), while the treasury.gov closing curve was up across the board. Handling: the official CSV governs, and the contradiction itself was written into the note in Section 1.4 as a live-market validation of "the 60 seconds after a data release do not represent pricing" — which is exactly the discipline written down in Section 8.1 of yesterday's list.

Two substantive criticisms this report makes of yesterday's list (written into the body, not just internal notes):

• Section 2.4: the table headline "a non-existent negative" overreached. The body of Section 1.4 said "that table's readings are unusable" (accurate), while the Section 6 headline said "a non-existent negative" (overreach). Three of the four did fall on the day, just for the reason of the software selloff rather than that table. When two statements in one report differ in strength, the reader remembers the strong one.
• Section 2.5: the 50,000-share volume threshold let "volume" veto "news." ETON held a primary earnings report with a 192% EPS beat, +99% revenue and raised full-year guidance, yet was downgraded to "watch only" because of 13,000 shares of pre-market volume, missing +16.44%. Recommend changing it to: volume is only used to veto moves with no news behind them.

Items this report itself left blank/unfinished:

• Raw NYSE advance-decline counts (second straight day) — the breadth characterization in Section 1.2 rests on the directional consistency of three proxy readings, not on advance-decline counts. This gap should be closed as soon as possible, because "the index fell while most stocks rose" is the first sentence of Section 0.
• The original BofA report on AVGO ($370 billion / $150 billion / 20GW / the XPV structure) — all media relays, labeled at every point of use in Section 3.3, with Section 6.3 also noting "that figure is pending verification."
• ORCL's single-day catalyst for 8/14 — not located, honestly written up as "a continuation of the repricing of existing information."
• The California DFPI's latest position on IMXI — not obtained, listed as the first action before Monday's open in Section 6.3.
• The latest rate pricing for the September FOMC (71%/29% is from yesterday's list) — this report did not re-pull it, marked "pending verification" in Section 6.2.
• Current DRAM/NAND spot prices — not obtained for a second straight day, with Section 3.2 stating explicitly that it did not close the gap, only drew the boundaries more clearly.

Channel recommendations for the next report:

1. The combination of stockanalysis + CBOE + api.nasdaq + treasury.gov independently completed all data pulls today with yfinance half dead; recommend fixing it as the default primary, with yfinance demoted to an optional supplement.
2. CBOE must be validated symbol by symbol on last_trade_time, not merely on whether the endpoint returns 200. Today both _DJI and _COMP were of the "200 + normal-looking price fields + stale timestamp" silent-failure form.
3. _VIX9D/_VIX1D must be checked for a return value ≠ 0.0 every time (0.0 yesterday, 10.61 today — the same endpoint in two states on two days).
4. Use stockanalysis's epv for the pre-market slot and stockanalysis's ep/ecp/es for the after-hours slot; each field returns null in the other's session, so do not expect one call to get everything.
5. The local data channel for the Dow needs fixing: CBOE _DJI returned 7/31 dead data today (price fields normal, only last_trade_time gives it away), so timestamps must be checked at symbol granularity, not just at endpoint granularity. Today it was saved by the DIA back-calculation; next time there may not be such a convenient cross-check.

---

⚠️ Risk disclaimer: this recap is a post-close review of information and observations only and does not constitute investment advice. Data may differ in timeliness or basis; please refer to company disclosures/SEC filings, and do not use this directly as a basis for trading.