US Market Recap | Tuesday, 2026-09-15 (ET)

Coverage window: 2026-09-15 09:30–16:00 ET regular session + 16:00–17:05 ET after-hours. Methodology notes:

• Open, high, low, close, volume and after-hours price for indices/single names/ETFs come from the CNBC quote API (read at 09-15 16:55–17:05 ET), each one cross-checked by reverse-computing close − change = previous close.
• Treasury yields use the US Treasury's official closing curve for the day (daily_treasury_yield_curve, 15:00 ET fixing), not the quote vendor's change/percent fields (reason in the internal notes at the end).
• Every single name in this piece carries three numbers at once: ① daily change (vs previous close) ② open→close ③ close vs this morning's pre-market price. Column ③ is the primary basis for reconciliation — the pre-market list went out at 08:15 ET, the earliest price a reader could actually get is the open, and the gap segment was already complete and unexecutable by the time the list was published. Reconciling on daily change alone systematically overstates the hit rate.
• Pre-market prices come from the pre-market list's own saved file work_uspre0915/premarket_scan_20260915.csv (570 names); VERA/PLAY/WAY/MPC are not in that CSV and have been reconstructed separately from the numbers in the list's body text and flagged as such.
• Sector/theme baskets are computed by me, name by name, equal-weighted, not third-party sector indices; every basket lists its constituents and count.

---

0. The Recap in One Line

1. Today was risk-off, but the real axis wasn't AI — it was crude oil. All three major indices turned from green to red: S&P −0.45% (7,585.73), Dow −0.63% (52,093.11), Nasdaq −0.78% (25,981.57), Russell 2000 −0.76%. But the index declines masked a 3.92 percentage point sector spread: Energy XLE +2.17%, the only one of the 11 SPDR sectors to close up more than 0.5%; Consumer Discretionary XLY −1.75% at the bottom. WTI crude +4.12% to $105.57, heating oil +5.92% — Saudi Arabia's East-West pipeline entered its fifth day offline since the 9/11 shutdown, and the market today began pricing how long rather than merely that it is down.
2. The pre-market list's most serious misjudgment happened on the very structure it was most confident about — and it was struck down by the falsification condition the list itself had written. The core of the list was "Monday's moves mean-revert" (full sample of 562 names, corr −0.55). Recomputing today on the same set of names:
   • corr(Monday change, today's gap) = −0.62 (n=540) — the mean reversion is real;
   • corr(Monday change, today's open→close) = +0.20 (n=540) — after the open the direction flips: Monday's move continues intraday rather than reversing;
   • corr(this morning's pre-market change, today's open→close) = +0.008 (n=547) — that pre-market table's predictive power over the regular session is exactly zero. Conclusion: the −0.55 structure the list described is entirely real, but it had already been 100% realized before the 09:30 opening bell. The list wrote up a completed overnight repricing as a trading plan for the day.
3. Two legs moved in opposite directions intraday — a clean 19-to-0 flip. Semiconductor equipment, 8 names, 8/8 negative open→close (avg −2.28%); pure-play cybersecurity, 11 names, 11/11 positive open→close (avg +4.31%). The list's theme 2 ("oversold AI hardware bounce") and theme 3 ("cybersecurity sentiment fading") both ran in reverse inside the tradable window. The list's §8 wrote its own criterion: "if the SMH-vs-IGV spread re-widens after the open, then theme 2 of this piece is wrong" — today's open→close was IGV +0.31% vs SMH −0.82%, the spread re-widened by 1.13pp, in the same direction as Monday. By its own rule: wrong.
4. Hit rate (on the pre-market-price basis): longs 0/17, short watch 0/2, Top 5 list 0/5; while the avoid list went 11/12 with a mean of +3.19%. (On the closing-price basis the figures are 10/17, 0/2, 0/5 and 9/12 — the difference is entirely the gap segment.) Almost all of this piece's value was in what not to touch; not one of the "touch this" calls worked.
5. No heavyweight earnings after the close — nothing leaves a single-name overnight catalyst today, so every chip is on 2 p.m. tomorrow. I scanned after-hours quotes for 620 large caps: the largest move was NetEase (NTES) −4.18% (notional $1.4 million), everything else within ±2% and on extremely thin volume; stockanalysis's after-hours top-ten gainers/losers were all micro caps under $100 million market cap. Tomorrow: 08:30 ET August retail sales, 14:00 ET FOMC decision (roughly 91% priced for a 25bp hike, the first since July 2023), Lennar (LEN) earnings after the close.

---

1. Market Overview

Indices and Volume

All four indices closed in the bottom third of their intraday range, and all four were negative open→close — this was not a gap-down-then-stabilize day, it was a slow grind lower all session. SPY's volume ratio of 1.25x was modestly elevated, QQQ only 0.87x: the selling was concentrated at the index level rather than in panic turnover of the mega-cap tech weights.

Market Breadth (self-computed, basis stated)

The sample is the 547 names, out of the same 570 large caps saved down by the pre-market list, that have all three prices today (same yardstick, cross-checkable against the list):

Breadth was weaker than the indices: the S&P fell 0.45%, but the median of these 547 fell 0.63%, and decliners outnumbered advancers 2.17 to 1. The weights were holding the tape up; the median stock was bleeding.

The Base Rate for Gaps (the single most important table in this piece)

This table is the whole-market answer to the question "can a gap-up hold", and any single name's gap-fill has to be netted against it first:

Today was an extremely asymmetric fade day: 3/4 of the gap-ups were sold back, and 6/10 of the gap-downs kept going down. In other words, direction lived in the gap and resistance lived in the session — anything you bought after the open faced a headwind. That explains why all 17 of the pre-market list's long ideas closed below this morning's pre-market price, without a single exception.

Rates, FX, Volatility

The 10-year touched 5.041% intraday (CNBC quotes, the highest since 2007); the Treasury's official close fixed at 5.00%. The curve made a near-parallel small shift higher (+2~3bp across the whole curve) — neither bear steepening nor bear flattening — today was not a rates-driven day; rates were only the backdrop.

Sentiment read: risk-off, but not flight-to-safety — reflation. Three mutually independent pieces of evidence point to the same conclusion:

1. Crude +4.12% while gold −0.42% and the dollar +0.25%. If the market were reading the Saudi pipeline as a geopolitical risk event, gold should have rallied and the dollar should have rallied more. Gold falling says money read it as "supply shock → inflation → higher for longer," not "war → safe haven."
2. VIX rose just 0.10 point to close at 17.20. With the index down 0.45% and the first rate hike since 2023 one day away, the volatility market still refuses to price panic. Even the intraday high of 18.03 was set near the open.
3. High-yield HYG −0.19%, investment-grade credit unchanged, KRE −0.08%. Credit was completely calm: today's selling did not transmit to the credit side; it was a sector reallocation inside equities, not systemic de-risking.

---

2. Pre-Market List Reconciliation

The primary basis for reconciliation is "close vs this morning's pre-market price" (the third value column). The reason is in the methodology notes at the top: the list went out at 08:15 ET and the gap segment was not executable. Both bases are shown so readers can choose.

① Long/Constructive Bucket, 17 names — 0/17 on the pre-market-price basis

Summary: of the 17, 10 closed up on the day (58.8%), but 0 were positive against the pre-market price, with a mean of −2.00%.

② Short Watch, 2 names — 0/2, and the most expensive error in the whole list

Shorting these two lost an average of 3.41% executed at the pre-market price, and 2.94% executed at the open. The evidence chain the list built for this trade was the thickest in the piece (EV/S at the 100th percentile of three years, the breakdown of PANW's NGS ARR definition, three mutually corroborating points on the CyberArk acquisition) — every piece of evidence held up, and the trade lost money on every leg. See the methodology review in §3.

③ Avoid / Bearish List — 11/12 on the pre-market-price basis; this is the only genuinely profitable part of the piece

Avoid list, 12 names: 11/12 realized on the pre-market-price basis (mean +3.19%, i.e. these names fell an average of 3.19%); 9/12 on the closing-price basis.

④ Energy/Refining, 6 names (the list gave them "no recommendation slot") — 0/6, the biggest error in the piece

6/6 rose, averaging +3.43%, and 6/6 were positive open→close (avg +3.10%) — on a day when 76.6% of gap-ups got sold back, energy walked uphill against the base rate. This was not a free ride; it was the only active buying of the day.

⑤ Top 5 List — 0/5

Top 5 hit rate 0/5. Three verification points landed precisely (VERA broke the pre-market price, LRCX gapped up and faded, the CIBR spread converged), yet not one of them changed a position conclusion.

⑥ Aggregate Hit Rate and Self-Critique

The total row excludes the Top 5 — those 5 are a subset of the four groups above (VERA/AMAT/LRCX/LLY are in the long bucket, CRWD is in short watch), and are listed separately only to show the outcome of "the 5 slots the list itself cared about most". They are not summed, to avoid double counting. The total of 37 = 17 + 2 + 12 + 6.

The self-critique in one line: this piece failed not on facts but on the clock.

Every fact in the list stands up to re-checking — the mean-reversion correlation is real (recomputed today on gaps it is −0.62, even stronger than the −0.55 the list itself computed), the cybersecurity valuation percentiles are real, refiners' P/B at the 99th percentile of five years is real, AMAT's three multiples above their five-year range are real. What was wrong was placing those facts on the wrong time axis: for a list published at 08:15 ET, the first price a reader can execute is the 09:30 open, and the variable the list ranked on (how much a name fell on Monday) was fully priced before 09:30 (corr drops from −0.62 in the gap to +0.20 intraday). It wrote an after-the-fact explanation of an overnight move and shipped it as an intraday trading plan.

Second layer of self-critique: the list wrote down three precise falsification conditions, all three triggered, and none was acted on. "SMH/IGV spread re-widens = theme 2 is wrong" — it widened 1.13pp; "LRCX gaps up and fades" — the opening price was the high of the day; "VERA breaks $39.10" — it opened 5% below. The quality of the criteria far exceeded the quality of the conclusions; the problem is that the criteria were only written down, never given veto power.

---

3. Theme Verification

Sector Panorama (11 SPDRs, sorted by daily change descending)

The XLE-to-XLY spread of 3.92pp was the main axis of today's cross-section. And XLY's volume ratio of 2.23x was the highest of any sector — selling consumer discretionary was the only "high-volume" sector trade of the day. At the single-name level: CMG −5.94%, CVNA −5.60%, CASY −5.55%, DLTR −5.35%, TTWO −4.93%, PLAY −19.01%.

This transmission chain deserves to be spelled out on its own, because it is the only internally consistent macro narrative of the day: Saudi Arabia's East-West pipeline still offline on day 5 → WTI +4.12%, heating oil +5.92% → energy is the beneficiary, consumer discretionary is the cost-bearer → and on the same day the Empire State manufacturing survey showed the prices-paid index up 5 points to 63.1 and the prices-received index up 5 points to 28.1 (while the headline index plunged 13 points from 20.6 to 7.6, far below the 14.75 expected). A stagflationary "growth weakening + prices accelerating" combination, released the day before a rate decision — which neatly explains why gold didn't rally, why VIX didn't rally, and why energy and consumer discretionary split in opposite directions. ⚠️ Must be noted: the pre-market list, in its §8 risk section, explicitly anticipated that "the prices-paid component within it may do more damage than the headline index under the current inflation narrative" — that call was completely right; the data was simply released at 08:30 ET, after the list's read time, so the list could not fold it into its conclusions.

Two Missed Themes: Not a News-Detection Miss, a Coverage Miss

① Life science tools (8 names +4.46%, open→close +4.11%)

Six names moving together, volume ratios broadly 1.7–2.1x, and all of the gain coming after the open (every gap within ±0.4%) — this was real money flowing in during the regular session, not the pricing of overnight news. ⚠️ I could not find a same-day first-hand catalyst that explains this collective move. The adjacent facts I did find: the Rally for Medical Research event on Capitol Hill on 9/16–9/17 (related to NIH funding priorities), and a third-party life-science-tools industry outlook report published 9/15. Neither is enough to support a sector move with 6 names up together on 2x volume, and I will not write them up as the cause. Attribution: unconfirmed. On the same day XBI (biotech) was −2.27%, which says this was not "biopharma broadly strengthening" but something occurring precisely at the "tools and instruments" layer — that structure is itself a clue, but I did not obtain a first-hand source explaining it.

② RF chip M&A endgame (SWKS +13.55%, QRVO +9.34%, QCOM +4.25%)

SWKS was the top gainer among the 620 large caps I pulled data on today (QRVO second, RVTY third, ILMN seventh), open→close +11.75%, volume ratio 1.49x; QRVO open→close +8.91%, volume ratio 1.58x. Both had tiny gaps (SWKS +1.61%, QRVO +0.39%), with nearly the entire move occurring inside the regular session. The backdrop is the $22 billion merger of SWKS and QRVO: US antitrust review cleared in early August, China's SAMR is in Phase III (the final stage), and the SWKS CEO recently told an investor conference that only two jurisdictions remain, with the goal of closing this year. But I found no new announcement or regulatory approval dated 9/15 to correspond to this 13.55%. The most recent dated, specific development I found was the two rallies on 9/10 and 9/11 (roughly +10% and +9.8% respectively); today is the third jump on the same logic, but I could not confirm today's specific trigger, so it is flagged as attribution unconfirmed.

What these two misses have in common: it was not that "the news wasn't found," it was that "these two industries weren't on the list's coverage roster." The pre-market list's 759-name scan sample did cover them (SWKS, QRVO, ILMN and TMO are all in the ≥$15 billion market cap pool), but the list's narrative frame had only two legs (AI hardware / cybersecurity), and stocks belonging to neither leg were never read, even though they appeared in the scan data. This is the classic form of ranking-as-filtering: the scan is whole-market, the attention is not.

Methodology Review: Why "All the Evidence Right, All the Trade Wrong" Happened on the Cybersecurity Leg

Not one of the three pieces of evidence behind the CRWD/PANW short watch was falsified today:

• EV/S at the 100th percentile of the past three years — after today's rally it is only higher, so it still holds;
• PANW's +63% NGS ARR comes mainly from the CyberArk acquisition, while its own FY27 guidance is only +22~23% — holds;
• Monday's 12~16% had no order/ARR/guidance support behind it — holds.

But all three answer "is it expensive" and "is the growth real"; none answers "will it fall today." The verification point the list attached was "does the gap between CIBR and the pure plays converge" — and today that criterion passed (Monday was CIBR roughly +4% vs pure plays +12.04%, an 8pp gap; today CIBR +0.95% vs pure plays +2.18%, a 1.2pp gap, 85% converged), while the short trade lost 3.41%.

The criterion measured "will the breadth of buying normalize"; the variable that loses money is "direction." The two are orthogonal. This matches exactly a same-family lesson recorded on this machine before: a verification point must test the variable that will cost you money; a criterion written starting from "how will I lose money" should have been — "of the top 5 pure-play cybersecurity gainers from Monday, did a single one close below its opening price today?" Today's answer is 11/11 closed above the open, and that criterion would have pulled the short watch within an hour of the open.

---

4. Earnings and Corporate Events

After the close today: no heavyweight earnings

I scanned after-hours quotes for 620 large caps one by one (read 16:55–17:05 ET). The largest move was NetEase (NTES) −4.18%, on just $1.4 million of after-hours notional; the only name with after-hours notional above $10 million was CRCL −1.25% ($30.9 million), a natural extension of the intraday decline rather than new information. The third-party after-hours top-ten movers were all micro caps under $100 million market cap (WAFU +66.9%, TPST +40.3%, etc.), unrelated to the broad market.

Conclusion: no single-name overnight catalyst tonight. The counterparty to every position is that statement at 14:00 ET tomorrow.

The real corporate event of the day: AXON's $1.0 billion 0% convertible (confirmed first-hand)

This is the one item this piece needs to backfill specifically — because yesterday's pre-market list explicitly wrote "I could not confirm this claim from a first-hand source, so it is not a basis for trading."

Today Axon (AXON) −9.81%, among the worst performers in the S&P 500, gapping down −4.59% at the open, falling another 5.47% after the open, closing at $442.08 in the day's low zone, volume ratio 3.18x.

First-hand verification result: the rumor is true. Axon filed an 8-K on 2026-09-15 (Items 1.01 / 2.03 / 7.01 / 9.01) plus a 424B5 prospectus supplement; exhibit EX-99.1 is the company press release, which reads:

"Axon Announces Proposed Offering of $1.0 Billion of 0% Convertible Senior Notes… due 2031… with an additional $150 million over-allotment option granted to the underwriters… underwriters are Goldman Sachs, Morgan Stanley, J.P. Morgan, RBC and Citi."

At the same time, the company raised its revolving credit facility from $300 million to $500 million (with a further $150 million accordion), extending maturity from 2030-03-11 to roughly 2031-09-18, with that amendment conditioned on completion of the convertible offering. Use of proceeds: paying the cost of the capped call plus general corporate purposes (including acquisitions).

The methodological point (more important than the conclusion): the pre-market list's 08:15 ET statement that it was "not confirmed by a first-hand source" was accurate at the time — the company's press release came out that same day, 9/15. But the list used that to tag AXON as "attribution questionable," while the market was already pricing that "unconfirmed" news pre-market (−3.00% pre-market). "I can't find a first-hand source" is a statement about my search, not a statement about the world. When thin search meets thick price action (AXON already −3% pre-market, −4.6% at the open), the thing to doubt is the search, not the price. The correct handling today would have been: keep the avoid tag (which proved entirely right), but write the reason not as "the rumor may be false" but as "the price is already pricing news I cannot confirm, therefore I do not participate."

Other first-hand facts to backfill

• Empire State manufacturing survey (released 9/15 08:30 ET, New York Fed): headline index 7.6 (prior 20.6, down 13 points), consensus 14.75, a large miss; new orders 2.0; shipments −3.2 (turning negative); prices paid 63.1 (+5 points) and prices received 28.1 (+5 points), both accelerating further from already elevated levels. Firms remain optimistic about the outlook. This is a textbook stagflationary reading, published the day before a rate decision.
• Saudi East-West crude pipeline: shut down on 9/11 after drone attacks launched from inside Iraq on 2026-09-10/11; 1,200 km long with capacity of roughly 4–5 million barrels per day (some reports say 7 million bpd at full capacity), it is Saudi Arabia's only alternative export route bypassing the Strait of Hormuz; Hormuz has been largely closed since the US-Israel war with Iran began in February this year. Today's new information is that the combination of "repairs may take weeks" and "global inventory buffer is only 5–7 days" was accepted by the market — i.e. exactly the "unpriced variable: duration" the list itself had written about.

---

5. Flows and Sentiment

Direction of Sector Rotation

Inflows: energy (XLE +2.17%, XOP +3.22%, OIH +2.35%), materials (XLB +0.48%, within it LYB +4.27%, DOW +3.29%, CF +3.28%, all oil-price chain), cybersecurity (CIBR +0.95%, open→close +1.79%), life science tools. Outflows: consumer discretionary (XLY −1.75%, volume ratio 2.23x, the only sector on elevated volume), utilities (XLU −1.20%), the whole crypto chain (IBIT −3.64%), biotech (XBI −2.27%), memory/storage (−2.17%).

Three Mutually Independent Sentiment Readings

1. This is not flight-to-safety, it is inflation repricing. Crude +4.12% while gold −0.42% and the dollar +0.25%. Gold does not fall in a geopolitical safe-haven rally. Money read the Saudi pipeline as a cost shock, not a tail risk — which points to the same thing as the Empire State prices components accelerating.
2. The volatility market refuses to price tomorrow's event. VIX closed at 17.20, up just 0.10 point, with the intraday high of 18.03 coming near the open. On the eve of the first rate hike since July 2023, VIX parked at 17 means either the market is convinced 25bp is fully priced, or too little protection is being bought. This is consistent with yesterday's list, and it still holds today.
3. Credit did not participate at all. HYG −0.19%, KRE −0.08%, KBE −0.28%, XLF −0.32%. The tail risk yesterday's list flagged — "10Y above 5% re-amplifies unrealized securities losses at banks" — was not activated today (10Y moved only 3bp). Bank stocks actually stabilized: JPM +0.67%, BAC +0.08%, WFC +1.14%; Monday's broad bank selloff did not continue.

One Volume-and-Price Fact Worth Recording Separately

Of the 111 names that gapped up today, 76.6% closed below the open, while all 11 pure-play cybersecurity names gapped down and 11/11 closed above the open. The significance of that contrast: on a day when the whole market was drifting lower, one sector going 11/11 against the tape is not noise — someone is persistently absorbing supply. Yesterday's list argued that Monday's cybersecurity surge had "narrow buying, concentrated in the few purest-labeled names" — today's evidence runs the other way: the buying did not narrow, it spread from CRWD/PANW out to second-tier names like TENB (+8.58%), QLYS (+7.21%) and S (+3.85%). Spreading is a signal of continuation, not of fading.

---

6. Next-Day Outlook (2026-09-16, Wednesday)

① Theme Continuity

② Tomorrow's Calendar (first-hand confirmed)

⚠️ Calendar risk warning: VIX closed at 17.20 today, having built no risk premium at all for this decision. If the statement or the press conference departs from the "25bp + neutral language" combination, the catch-up in volatility could arrive very fast. Every position based on "today's trend" could be erased in one stroke after 14:00 tomorrow — this is the same judgment as yesterday's list, and now there are fewer than 24 hours left.

③ Focus List (Ticker + verification point)

This time every verification point is rewritten to "test the variable that loses money," and each is given veto power: if the criterion fails, the conclusion is withdrawn, not merely marked down.

Variable-diversity statement: the 5 lines above are really only 3 independent variables — physical oil (XLE/VLO/XLY take 3 slots, two sides of the same shock, together they should not exceed 1 net position), rates/housing (LEN), and cybersecurity (now neutral, no position). Strictly speaking, there is only one tradable independent idea tomorrow, and its counterparty shows up at 14:00.

④ What to Avoid

1. The whole crypto chain (COIN / CRCL / MSTR / BMNR / IBIT): all 6 names fell 7.71% today, and a rate hike directly raises leverage costs — this is currently the most internally consistent criterion in the piece.
2. Consumer discretionary (especially restaurants / discount retail / used cars): CMG −5.94, CVNA −5.60, DLTR −5.35, CASY −5.55; the cost pass-through from $105 oil is only beginning.
3. Any setup that depends on "the gap keeps running": of the 111 names that gapped up more than 0.5% today, 76.6% closed below the open. On an event day like tomorrow, that ratio will only be higher.
4. A "second bounce" in semiconductor equipment: today proved the bounce happens only inside the gap. Unless first-hand data appears at the current-order / book-to-bill level, stop going long on "it fell a lot."
5. Strong names with unconfirmed attribution (SWKS / QRVO / ILMN / RVTY): this is not a bearish call, it is non-participation. I do not have a first-hand source for today's catalyst, and building a position on an unverified attribution after a 9~13% move does not work on risk/reward.

⑤ Inputs for Tomorrow's Pre-Market List

1. The first sentence of tomorrow's list must be about the clock, not a correlation coefficient. Today's lesson: for any structure that ranks on "yesterday's change," compute both correlation coefficients first — one for the gap segment and one for the session segment — and use only the session-segment number for trading conclusions. Today those two numbers were −0.62 (gap) and +0.20 (session), opposite in sign.
2. Fix the reconciliation basis: every recommendation must state both a "vs previous close" and a "vs pre-market price" expectation, with the latter as primary. Today the overall hit rate differed by 20.8 percentage points between the two bases (51.4% → 30.6%); looking only at the long bucket, the difference is 58.8 percentage points (58.8% → 0%) — the choice of basis determined the conclusion in this piece more than any single-name judgment did.
3. Tomorrow is an event day; the list should explicitly state "the conclusions in this piece are valid until 14:00 ET", and give each recommendation a handling plan under both a hike and a no-hike scenario, rather than only one direction.
4. A "scan outside the narrative" step must be added: today's missed life science tools and RF M&A were both in the scan sample, filtered out by the narrative frame. I propose codifying a hard rule — every day, go through the top 20 gainers in the full sample one by one and check whether any of them belongs to none of the piece's themes; if one does, it must appear in the body at least once, even if the conclusion is "attribution unconfirmed, not participating."
5. Gaps to backfill: ① the same-day first-hand catalyst for the 8 life science tools names rising together; ② the specific trigger for today's jump in SWKS/QRVO (was there a SAMR approval or closing announcement); ③ the official restart timetable for the Saudi East-West pipeline; ④ the items left over from yesterday's list — current orders and book-to-bill for AMAT/LRCX/KLAC, ASML's most recent quarterly bookings, MU's HBM contracts and DRAM spot prices — the bull case for these four still lacks current demand-side evidence; today's decline did not change that gap, it merely made it less urgent.

---

Data Collection and Failure Log (internal)

1. The primary data channel this run was the CNBC quote API, available throughout: indices, single names, ETFs, futures, Treasuries and commodities all retrieved, including OHLC, volume, 10-day average volume ratio, after-hours price and after-hours volume. All 570 names of the full sample were retrieved in one pass, MISSING=0.
2. yfinance remains unusable: the query2 chart endpoint returned Edge: Too Many Requests this time (yesterday's list recorded it as "the only Yahoo endpoint not yet blocked"; today that endpoint is blocked too, so that note needs updating). stockanalysis's /api/quotes/e/ is blocked by Cloudflare (returns the "Just a moment..." challenge page), but its HTML pages are readable via WebFetch (the after-hours movers list came from that path).
3. 🔴 Silent failure in CNBC batch requests — a check must be codified. In a batched symbols=A|B|C request, XLK / XLY / XLI / TLT / CYBR return records where the symbol field is correct but every price field is None — my missing check was based on "does the key exist," so it reported MISSING=[] while 5 names were in fact entirely empty. After switching to one-at-a-time requests, everything was retrieved except XLY/CYBR, which needed one more retry. The criterion must change from "is the key present" to "does the last field have a value." This is the same-family form, on CNBC, of the "batch quotes silently drop samples" lesson.
4. 🔴 Nasdaq's ETF endpoint /api/quote/{sym}/info?assetclass=etf gave wrong change figures and has been dropped. Measured: for XLK it returned +0.21 / +0.11%, while the true values after cross-checking CNBC against Nasdaq's own summaryData.PreviousClose are −0.54 / −0.29%; for TLT it returned +0.03 / +0.04%, while PreviousClose 80.93 and last 80.74 reverse-compute to −0.19 / −0.27% (matching CNBC). lastSalePrice − netChange ≠ PreviousClose — the price triple fails to close on the spot. All ETF data in this piece uses CNBC. This is the first time the "the price triple must be reverse-computed" lesson has hit on Nasdaq's ETF path.
5. 🔴 CNBC's Treasury change_pct field is single-field broken again; worked around per the existing lesson. This time US2Y returned change: +0.031 but change_pct: -0.0586%, opposite signs within the same record. All Treasury yields in this piece use the Treasury's official daily_treasury_yield_curve CSV (09-15: 2Y 4.67 / 5Y 4.83 / 10Y 5.00 / 20Y 5.40 / 30Y 5.36; 09-14: 4.65 / 4.80 / 4.97 / 5.37 / 5.34); CNBC was used only for the 5.041% intraday high.
6. Sources returning WebFetch 403: cnbc.com article pages (including the 9/15 market live blog), same as yesterday. newyorkfed.org and sec.gov read normally.
7. First-hand verification list (everything written into the body is here):
   • AXON $1.0 billion 0% convertible → SEC 8-K 0001193125-26-391320 (2026-09-15, Items 1.01/2.03/7.01/9.01) + the EX-99.1 company press release text + the same-day 424B5. Amount, coupon, maturity, the $150 million over-allotment, the underwriting syndicate, and the revolver going from $300 million to $500 million all come from first-hand filings.
   • Empire State, September → New York Fed website: 7.6 (prior 20.6) / new orders 2.0 / shipments −3.2 / prices paid 63.1 / prices received 28.1, release date 2026-09-15.
   • Treasury yields → the US Treasury's official daily yield curve CSV.
   • Index closes → CNBC quote API, and digit-for-digit consistent with the three numbers in CNBC's own write-up (Dow −328.09 to 52,093.11; S&P −0.45% to 7,585.73; Nasdaq −0.78% to 25,981.57).
8. ⚠️ Two attributions could not be confirmed and are explicitly labeled "unconfirmed" in the body rather than filled in with invention:
   • 6 life science tools names rising together (RVTY +9.11%, ILMN +6.76%, GH +4.65%, TMO +4.61%, DHR +3.01%, A +2.36%): the only adjacent facts found were ① the 9/16–17 Rally for Medical Research event on Capitol Hill ② a third-party industry outlook report on 9/15 ③ TMO's $0.47 dividend ex-date falling on the same day (an ex-date should push the price down, not up — wrong sign, ruled out). None of the three can explain a sector-level move on 2x volume, so the body says "attribution could not be confirmed."
   • SWKS +13.55% / QRVO +9.34%: the merger developments (US antitrust cleared in early August, SAMR Phase III, the CEO saying only two jurisdictions remain) are all old information from 9/10–9/14, and no specific trigger dated 9/15 was found. Note: SWKS already rose roughly 10% on 9/10 and 9.8% on 9/11, and today is the third jump on the same logic — which itself hints that a new approval may have landed, but I do not have the document, so I do not write it.
9. Three items actively screened out and kept out of the body:
   • Media attributing the SWKS rally to "optimism about iPhone demand" — no same-day first-hand source, and it cannot explain QRVO's simultaneous +9.34% (QRVO's iPhone exposure is structured differently); ruled out.
   • "The 10Y touching 5% caused today's selloff" — the 10Y moved only 3bp today and the whole curve shifted up 2~3bp in parallel; using a 3bp move to explain a 0.45% index decline and a 3.92pp sector spread does not work on magnitude, so the body downgrades rates to a backdrop and states why.
   • The third-party after-hours movers list (WAFU +66.9%, etc.) — all micro caps under $100 million market cap, unrelated to the broad market; the body uses it only to corroborate "no heavyweight earnings tonight."
10. Statistics are saved and recomputable:
    • work_usrecap0915/recon.json — per-name daily change / gap / open→close / close vs pre-market price / Monday change / volume ratio for 547 names
    • work_usrecap0915/allrows.json — all 600+ names including the follow-up pulls
    • work_usrecap0915/ust.csv — the Treasury's official yield curve
    • Key statistics: corr(Monday, gap) = −0.6240 @n=540; corr(Monday, open→close) = +0.1985 @n=540; corr(pre-market %, open→close) = +0.0079 @n=547; breadth 172 up / 373 down / 2 unchanged @n=547; of those gapping up >0.5%, 85/111 (76.6%) closed below the open.
    • Sample basis: 570 names (the full sample saved down by yesterday's pre-market scan) → 547 with all three of today's open/close/previous close → 540 that also have a Monday change (i.e. the correlation sample). VERA/PLAY/WAY/MPC are not in the 570-name CSV; their pre-market prices were reconstructed from the numbers in yesterday's list body, and this is flagged at each occurrence in the body.
11. One self-imposed constraint on this reconciliation basis: the "realized?" column in §2 is judged uniformly on close vs pre-market price, not close vs previous close. If the latter were used, the long bucket's hit rate would go from 0/17 to 10/17 (0% → 58.8%, a 58.8 percentage point difference) — and those 10 names are all cases of "closed up on the day but below the pre-market price" (ASML is the clearest: +1.04% on the day, −2.20% against the pre-market price). The basis difference across all four groups combined is 20.8 percentage points (30.6% → 51.4%). That difference is the gap segment, and it was already unexecutable when the list was published.
12. Gaps left for the next piece: ① the same-day first-hand catalyst for life science tools and for SWKS/QRVO (highest priority — both were among the day's top gainers); ② the official restart timetable for the Saudi East-West pipeline (it determines whether the energy theme lives or dies); ③ current demand-side evidence for the four semis left over from yesterday's list, still not filled in; ④ the item in yesterday's internal log about "whether there is a 20-year reopening at 13:00 ET today" was not pursued further in this piece, and the problem of the Treasury upcoming_auctions endpoint returning stale data is still unresolved; I suggest switching to TreasuryDirect's auction announcement XML.

---

⚠️ Risk warning: this recap is a post-close review of information and observations only and does not constitute investment advice. Data may be subject to timing or definitional differences; please refer to company disclosures and SEC filings. It must not be used directly as a basis for trading.