Daily Report — 07/31/26 · “Three Attacks on Scarcity, and a Hedge Dated 2027”

Three attacks on scarcity, in one week

For most of this year the best trade in the market was scarcity. Memory, semiconductor equipment and the AI-infrastructure complex were priced on the assumption that supply was constrained and would stay constrained. In the last full week of July that assumption was attacked from three directions inside five sessions, and the entire month’s damage follows from it.

The first attack was a listing. China’s largest memory manufacturer went public in Shanghai on the twenty-seventh in Asia’s biggest offering of the year, and opened up more than four hundred and sixty percent. For a few hours it was the most valuable mainland-listed company in China, ahead of the country’s largest bank. It is a company with roughly eight percent of the global market, sitting fourth. The point was never the fundamentals — it was that the capital markets had just handed a competitor an enormous cheque to build capacity into the exact shortage the entire Western complex was being valued on.

THE LISTING Raised ¥57.92B (~$8.56B) — Asia’s largest IPO of 2026 · 6.688B shares at ¥8.66 · opened ¥49.50, +466% · market cap ¥3.31T (~$489B), briefly China’s most valuable mainland-listed company · ~8% global share, fourth worldwide · cost-per-bit more than 30% behind the three incumbents because it multi-patterns with older lithography rather than using extreme-ultraviolet tools · targets high-bandwidth volume production in 2027, roughly three years behind the leader, which holds ~56% of that segment. Sell-side fair values span ¥14.90 to ¥116 — a 678% disagreement.

The second attack was a manufacturing headline — reports the same week that China had begun mass-producing its own deep-ultraviolet lithography tools. That is the equipment layer, and it is why the largest European equipment maker fell almost six percent on the day specifically. If the tools can be made domestically, the export-control moat that underwrites the entire equipment complex has a date on it.

The third attack was an idea, and it is the one with the longest fuse. A leaked four-hour investor call from the founder of China’s leading open-weight AI lab laid out a deliberate strategy of capping margins: price the interface so the hardware pays for itself in ten months, publish the model weights, and let anyone build on top. His stated view of the incumbent chip designer was that it is digging its own grave — the software moat is code nobody wanted to rewrite, and machines write code now. His own lab has already shipped kernels in a second language and got day-one support on domestic silicon. His timeline for the domestic software problem being solved: within a year.

The efficiency point underneath it matters more for memory than the moat point. His flagship runs a small fraction of the active parameters of the largest Western models. If that architecture wins, the industry needs materially less memory bandwidth per unit of capability — which attacks the demand side of the same scarcity premium the listing attacked from the supply side.

Put together, that is a repricing, not a panic. The most-crowded factor in global equities then lost more than a third of its value in a month, the worst on record. And that is what broke the levered fund — not the other way round. The blowup was the leverage failing on a repricing that was already under way, and treating it as the cause is what produces the wrong conclusion about what happens next.

The honest counterweight, because it changes the clock. The new entrant is more than thirty percent behind on cost, three years behind on the high-bandwidth product that AI actually consumes, and barred from US federal procurement on a schedule that runs into late 2027. And the price war the thesis requires has not started: Chinese modules are reaching the market at prices that still track the incumbents, and on server memory the new entrant has actually been priced above one of them. Most of that capacity is being absorbed domestically rather than flooding the world.

So the thesis is real and the fuse is long. Which is exactly what the professional money did about it — see the multi-year collar below, and the term-structure split running through the whole complex. They agree with the direction and they dated it to 2027.

The month that ended exactly where it started

The S&P finished July down under a hundredth of a percent. Not roughly flat — flat. Over the same thirty-one days the index traded a three-hundred-and-fifty-point range, the Nasdaq lost more than six percent, small caps lost three, and the momentum factor — the single most crowded trade in global equities coming into the month — had the worst month in the history of the measurement, worse than the 2000 unwind and worse than 2008.

Everything was negative. Through Thursday the S&P was down, the total US market was down, international was down, the aggregate bond index was down and even municipal bonds were down. Then the last session dragged the headline back to unchanged.

That is the month in one sentence: an average of a collapse and a rotation that happened to net to zero — and the collapse was the scarcity trade being repriced. The candle it leaves behind is an inside month with a long lower shadow — a compression, not a resolution. The next monthly bar decides whether it was a pause or the middle of a reversal, and the market has spent the last two sessions arguing both sides at once.

It also matters how the mark got there. The index tagged its own daily band at the close on the last session of the month, and the last hour was spent manufacturing a green print on a tape whose institutional evidence ran one way all day. That is not an accusation, it is a description of month-end, and it is the reason the closing numbers need to be taken apart before any of them are believed.

The biggest positioning print of the month arrived on the quietest tape of the week

Friday produced the largest market-wide delta reading of July — tied with a single mid-month session for the highest of the last ninety-two — on a session where the index went essentially sideways. The session before, when the index rose more than one and a half percent in a straight line, the same measure printed less than a quarter as much. That inversion is the most important thing on Friday’s dashboard and it has a clean explanation.

Delta is not premium. The day’s options tape, once matched multi-leg structures and deep-in-the-money stock substitutes are stripped out, netted to almost nothing — the only negative session of the week — and the expiring book itself decomposed to a residue indistinguishable from zero on two-thirds of a billion dollars of gross. Nobody was buying exposure on Friday.

TAPE Whole file 42,331 rows, $16,510.09M gross, structure-clean residue −$13.4M (index −$88.9M, equity +$75.4M) after removing duplicate signatures, delta-one legs and same-second matched structures · expiring-book residue −$1.8M on $675M gross, 8.4% of the file.

The delta came from the book expiring, not from anyone positioning. Wednesday and Thursday’s rally pushed a month’s worth of accumulated call exposure deep into the money; on Friday those deltas converged to one as the clock ran out. The dealer ledger shows the expiring date carrying more than eleven billion dollars of call delta against two and a half billion of put delta — a net figure roughly five times larger than any other date on the board.

Two other panels say the same thing from different angles. Index gamma at the strike ten points above the close was the most negative single-strike reading of the cycle, sitting directly overhead all afternoon, with positive gamma stacked several points below — the classic pin geometry. And the cumulative flow line ran with put premium above call premium from the opening bell to the close, while price climbed all day. Price up, premium tilted to puts, delta exploding: those three facts only reconcile as an expiring book converging around a strike it could not get through.

Friday’s reading is a receipt, not a forecast. The exposure that produced it stopped existing at four o’clock.

The same thing happened at the end of May, and it is worth being precise about which half repeats

The pattern the operator flagged is real. Extracting the same panel back across the spring: the last session of May printed a high reading, the following Monday printed the single highest bar of the entire sample, and within four sessions the measure had flipped hard negative. Eight sessions after the peak it printed the deepest negative bar on the chart, and the index had fallen about four and a half percent from its high into that low.

The archive from that week is unambiguous about what it looked like at the time. The month-end file listed rising delta exposure as a bullish input — a mechanical dealer bid — while simultaneously flagging thirteen mega-caps whose fifteen-day accumulation pattern contradicted that Friday’s price tape, and an index gamma wall sitting at spot. Four sessions later the same file recorded the measure flipping to its most negative reading since December. Two sessions after that: the dealer bid did not return. Two after that: the index closed on its session low and sentiment printed its first fear reading of the cycle.

The first half of that sequence is arithmetic and it will repeat. A month-end expiration maximises the standing book’s delta by construction. The next session it is gone. Expecting the reading to fall from here is not a forecast.

The second half is where the signal actually lives, and it is conditional. Delta decaying to zero is mechanical. Delta going deeply negative means customers are buying protection faster than dealers can lay it off — and in June that required a trigger with a date on it. The trigger was an employment report that printed hot, gapped yields higher, and liquidated the most rate-sensitive trade on the board. The positioning reading did not cause the decline; it marked the moment the market was carrying the least protection, and then an event arrived.

Where the two set-ups differ is fuel. Late May entered its peak from a greed extreme with every index trend intact. This one enters from a liquidation that has already happened, with the Nasdaq’s own trend line still overhead, trend ranges roughly a third cooler across the index complex, and the accumulation-versus-distribution census net negative on a green day. There is materially less crowded length left to unwind than there was in June, so the honest version of the analog is: directionally similar, smaller in amplitude, and entirely dependent on an event that has a date.

That date is the first Friday of August.

Two dates in August carry the whole book

Before the conclusion, a correction to how the expiration panel is being read, because it matters and it recurs.

The flow map is a stacked bar, not a grouped one. The put segment begins where the call segment ends, so reading where the red bar terminates gives the combined total, not the put figure. That is the entire source of the discrepancy the operator spotted — the eight-twenty-eight row reads about ten million of puts by segment length and about nineteen by endpoint, and the fifty-five-million figure that seemed to belong to the August monthly actually belongs to the first-Friday row, which sits between two labels and is easy to misassign. The panel also labels only every second row, so a third of the expirations carry no label at all; the correct way to map them is against the timeline panel’s legend, which lists the real dates.

With that fixed, the operator’s two other readings are exact, and the picture is clean.

EXPIRATION MAP · segment lengths, mapped off the timeline legend 07/31 calls +$29.6M / puts +$10.3M · 08/03 +$8.4M / +$4.8M · 08/05 +$0.6M / +$1.4M · 08/07 +$24.5M / +$31.4M (puts 1.28× calls) · 08/10 +$2.3M / +$1.3M · 08/12 −$0.3M / −$0.7M · 08/14 +$2.9M / +$2.2M · 08/21 +$28.8M / +$18.0M · 08/28 +$8.3M / +$10.2M.

The single-day map is one photograph. The cumulative timeline is the film, and it is where the asymmetry sits: the August monthly has attracted roughly a quarter of a billion dollars of net negative premium over its life, having recovered from a deeper trough in mid-July and then rolled back over into month-end. The first Friday has attracted about half of that and is still making new lows — the only near-dated line still trending down. Everything else in August sits flat on the zero line.

The dealer ledger completes it. The first Friday of August shows negative call delta and negative put delta — both sides, which is unusual — and the August monthly carries the largest gross book on the board and the most negative net.

DEALER LEDGER · total deltas by expiration 07/31 +$11.36B calls / −$2.52B puts (the expiring book) · 08/07 −$0.23B calls / −$2.44B puts · 08/21 +$2.95B calls / −$4.58B puts — largest gross, most negative net. Cumulative net premium: 08/07 ≈ −$133M and falling · 08/21 ≈ −$265M · 08/14 ≈ −$47M · 08/28 ≈ −$35M.

A dealer community that is long a great deal of expiring delta today, short gamma into the first Friday, and carrying its largest and most negative book into the third Friday will amplify moves rather than damp them — on precisely the two dates the calendar has marked. The July employment report lands the morning of the first Friday. Consumer prices land the following Wednesday, producer prices the Thursday, retail sales the Friday after that. The August monthly expiration is the third Friday. There is no policy meeting in August at all; the next one is in September, and the curve now prices better than a seven-in-ten chance of a hike at it, with two hikes priced by December — from a decision that was a hold.

And Friday’s own closing tape pulled the book into exactly that window. In the last ten minutes of the month, two structures of four thousand contracts each rolled both sides of the index book — the upside and the downside — out of September and November and into the August monthly, for a combined credit of more than a hundred million dollars.

TAPE 15:50:09, 4,054 lots, four legs, $85.53M — Sep-18 7610C sold $42.81M (12.57 vol, delta 0.37) + Sep-18 7695C sold $27.54M (11.88 vol, delta 0.27) against Aug-21 7645C bought $11.41M (10.26 vol, delta 0.21) + Aug-21 7735C bought $3.77M (9.59 vol, delta 0.08); equal size on all four legs, so the sold legs carry ~2.5× the exposure of the bought legs — net short about $1.09B of delta for a $55.17M credit · 15:58:32, 4,000 lots — Nov-20 7300P sold $62.77M (16.89 vol) against Aug-21 7250P bought $10.70M (16.10 vol), plus Sep-18 7350P sold 2,500 for $22.60M at the same stamp (net credit $52.07M on the two-leg pair) · the Aug-21 7250 put was accumulated in nine tranches from 10:23 to 15:58: 10,550 bought against 1,072 sold — 9,478 net contracts, about $36.2M net. Combined across both structures the book is roughly $957M short delta and $7.7M per vol point short vega.

The read has two parts, and the second one matters more. On the time axis the book is concentrated inside the next three weeks and unwilling to pay for the two months after them. On the direction axis it is net short: every leg is the same size, so the closer-to-money, longer-dated calls they sold carry about two and a half times the exposure of the further-out, shorter-dated calls they bought, and the same inversion runs through the puts — near-dated protection bought, far-dated protection sold. This is not a desk positioning for a rally. It is a desk buying the right to be involved if something happens in the next three weeks, while saying it does not believe the index is meaningfully higher by the September expiration — and being paid to hold that combination. The August put strike they accumulated sits between Monday’s zone floor and the August one-standard-deviation floor, a little over three percent below Friday’s close.

The closing auction is not the tape

Nearly two hundred billion dollars printed in the dark on Friday with a net of minus twenty-four billion. Almost none of that headline is usable.

Two-thirds of the tape printed in names that traded six times or fewer, and a full third printed in names with no darkpool prints at all — blocks, which are transfers, not campaigns. The directional content lives in the eleven names that cleared fifteen prints, and those eleven net minus sixteen and a half billion.

TAPE · print-count buckets zero prints 2,631 names / $68.35B / 34.2% of tape / net −$6.22B · thin (1–6) 596 names / $67.14B / 33.6% / −$6.29B · mid (7–14) 39 names / $36.38B / 18.2% / +$5.10B · thick (15+) 11 names / $27.70B / 13.9% / −$16.48B.

But the mega-cap headlines are the month-end rebalance cross, and they must not be read as flow. Of the twenty largest block trades in the file, eighteen carry a four-o’clock timestamp. The side tag on a closing cross reflects which side of the auction the print was attributed to, not intent. Strip that one print out of each name and the biggest apparent buying in the file disappears.

TAPE · the 16:00 cross NVDA $4.46B at ask · AAPL $4.44B at bid · MSFT $3.80B at ask · AMZN $2.77B at bid · GOOGL $2.31B at bid · MU $2.30B at bid · AVGO $1.75B · SPY $1.67B · GOOG $1.22B · META $1.21B · AMD $0.98B. Microsoft screens +$5.53B on 14 prints with volume +71.0%; one $3.80B cross is most of it. None of the six largest mega-cap darkpool nets survive the auction adjustment, and none is cited here as directional.

What does survive, because it printed during the session rather than at the bell, is the index complex itself. The S&P tracker was distributed on ninety-two prints with volume expanding — there is no artifact reading of that. The Nasdaq tracker was distributed on sixty-six prints with volume contracting. And the AI-financing names that led Thursday’s squeeze were sold on thick tape one session later.

TAPE SPY −$10.24B on 92 prints, volume +17.1% → distribution on expanding volume · QQQ −$2.74B on 66 prints, volume −14.6% · AMZN −$2.45B on 19 prints, volume +36.9%, on a +15.32% day · ORCL −$0.82B on 18 prints · IREN −$255.89M on 14 prints · NBIS −$313.38M · APLD −$31.16M on 14 prints. Sector nets: the index-tracker bucket −$14.5B, technology +$4.0B (the cross), everything else negative.

The absorption read is the correct one, and it is now a mechanical verdict rather than an impression: on the last session of the month the index trackers were sold in size on rising volume while the closing auction produced a set of mega-cap prints that look like buying and are not.

The cleanest proof on the tape that the closing cross means nothing

One company, one session, two share classes — and the dark tape reports them in opposite directions. The voting class shows the second-largest institutional sale in the entire file. The non-voting class shows a large purchase. Same economic entity, same day, same closing auction. They cannot both be true.

TAPE GOOGL −$2.48B on 3 darkpool prints / 6 blocks, volume +90.55%, $2.73B at bid against $249.69M at ask · GOOG +$1.58B on 1 print / 3 blocks, volume +52.2%, $1.71B at ask against $132.43M at bid. Both are dominated by a single four-o’clock block — $2.31B and $1.22B respectively — and both sit inside the eighteen-of-twenty rebalance list above.

That is the whole argument in one company. The side tag on a month-end auction print reflects which side of the cross the trade was attributed to, not what anybody intended, and when the same issuer prints on both sides simultaneously the tag has demonstrated its own meaninglessness. Anyone reading either number as institutional conviction on Alphabet has read a bookkeeping artifact.

The readable Alphabet signal is the options tape, which is thin enough to have printed during the session and which agrees with itself across both classes.

What actually got bought and sold

Once every number is taken down to what survives structure-stripping, the day has one genuine bull structure and one genuine bear structure, and both are dated far out.

The bull structure is Amazon, and it is the only place on Friday’s tape where somebody paid up for dated upside. Two-thirds of the calls that were sold in the name struck in the money — that is closing and rolling, not shorting — and the overwhelming majority of the surviving position sits in November.

TAPE AMZN structure-clean +$130.06M; calls bought $338.0M against $213.3M sold, 67% of sold calls struck in the money; by expiration Nov-20 +$107.8M. Stock +15.32%, closed above its zone ceiling with the hottest range reading on the entire single-name board.

The third real bid is Alphabet, and it is the only name on the board with a streak. It has to be read as one company across two tickers, because the classes trade separately, carry separate data flags, and concentrate in different expirations — sum them or you are looking at half the book. Both were independently bid for a third consecutive session, which nothing else in the file can claim, and the money went long-dated rather than into the earnings pop: a two-and-a-half-year call bought outright at the ask in one class, and a pair of mid-2027 calls in the other, all lifted inside the first ten minutes of the session. Four fifths of the calls sold against that struck at or near the money — harvesting and overwriting, with only a token share genuinely short upside.

TAPE Combined structure-clean +$36.58M (+$18.39M and +$18.19M), third consecutive independent session across both share classes; week to date +$80.1M. Stock +6.73% / +6.88%. Sold-call composition 42% in the money, 39% within fifteen percent — only 3% opening and genuinely short upside. By expiration Jun-2027 +$18.2M and Jan-2028 +$13.6M. Prints: 09:50:59 Jan-2028 330C bought at the ask, 1,703 lots, $13.08M, 37.15 vol, OI 3,887 · 09:57:52–09:58:33 Jun-2027 320C + 325C bought at the ask, 1,150 lots, $10.04M, 35.05–35.17 vol · 13:27:27 Nov-20 410C bought to the ask, 5,000 lots, $5.26M, 34.61 vol, OI 2,028, opening. Week by class: GOOGL +12.2 / +11.4 / +2.9 / −2.6 / +18.4 and GOOG +20.8 / −5.2 / +4.9 / −0.9 / +18.2 — same direction on four of five sessions. Zone rows for both: GOOGL 310.42–340.23–370.04, trend 361.76 overhead, +3.91% of room, RANGE 86; GOOG 310.87–340.16–369.45, trend 359.68 overhead, +3.59% of room, RANGE 88.

Two caveats keep it a tier below Amazon rather than beside it. Roughly a third of Alphabet’s gross tape is zero-delta straddles — a 13,900-lot September structure and two 3,500-lot August ones, all opening, all unsided, carrying no direction whatsoever. That is vol buying, and it belongs with the dispersion book below rather than in the bull column. The non-voting class carries an unsided flag outright. And the darkpool is the four-o’clock cross — spectacularly so, in a way that deserves its own note below. Net of all that the dated call buying is real, it is the third session of it, and it is the cleanest mega-cap bid after Amazon — but the trend line still sits above price, with roughly four percent of room to the top of the band against thirteen percent below.

The bear structure is the semiconductor tracker, and it is the largest clean negative in the file — genuinely bearish by composition, with most of the sold calls well out of the money and opening against size larger than existing interest. But it is not a call on August. Essentially all of it sits in 2027 and 2028.

TAPE SMH structure-clean −$64.18M; puts bought $77.7M against $33.6M sold; 58% of sold calls >15% out of the money, opening, size above open interest; by expiration Jun-2027 −$48.7M and Jun-2028 −$19.0M. Stock +0.30%.

Everything else needs a caveat attached. The chip-equipment name that screens as the third-largest bid is one print — an in-the-money put sold for income, with almost no call buying behind it. The semiconductor index fund’s positive number is put-selling. Two of the biggest apparent bids of the week closed the month sharply lower.

TAPE LRCX +$32.77M is a single 14:24:07 Sep-18 340P sold at the bid for $33.15M; calls bought only $8.4M; stock −1.58% · SOXX +$13.21M is $16.9M of puts sold against $2.9M bought · MU −$34.18M with the front three expirations sold (Aug-07 −$10.8M, Aug-03 −$7.7M, Aug-28 −$7.3M), stock −5.90%, darkpool −$2.21B on 9 prints with volume −44.9% · MSFT −$48.82M with 63% of sold calls in the money — rolling up, not shorting · AAPL’s −$191.43M headline is 90% structure; the surviving figure is −$19.50M.

There is a term-structure split running through the whole semiconductor complex and it is the most interesting thing in the single names. Near-dated puts are being sold across the sector while multi-year puts are being bought in size. Nobody is hedging August in chips; they are hedging 2027. That is complacency in the front and conviction in the back, and the two are not in conflict — read against the opening section they are the same statement twice: the near-term risk in this complex is macro and positioning, and the structural risk is Chinese supply arriving on a schedule that is already published. The professionals put their money on the published schedule.

Korea: the long comes off after one session

Worth naming why this cohort is the one that broke hardest: the country’s index is roughly half two memory manufacturers, so an attack on the scarcity premium is an attack on the Korean index directly. The leverage that amplified it — a retail margin cascade running to well over a million accounts — was the transmission, not the cause. Same shape as the fund blowup, one country over.

Thursday’s report reopened a tactical long in Korea on a genuinely strong signature — eighteen darkpool prints, volume nearly doubling, half a billion of net buying, and an options book structured long into 2027. It required a second confirmation on thick tape to hold.

What arrived instead was the mirror image. Eighteen prints again — so the tape is still thick — but volume contracting by a quarter, the net inverting to a three-hundred-million-dollar negative, price down two and a half percent, and fresh put buying concentrated in the two August dates that matter. Two thick sessions, opposite directions, confirmation failed.

TAPE EWY darkpool −$304.96M on 18 prints with volume −24.1% → distribution; options structure-clean −$25.34M with puts bought $20.9M against $12.1M sold; by expiration Sep-18 −$22.7M, Aug-14 −$7.5M, Aug-07 −$6.3M; 15:54:40 Sep-18 180C sold to the bid, 17,000 contracts, $14.81M, opening. Stock −2.55%.

The multi-year structure may well still be in place; the tactical long is not. The tag comes off. This is the correct outcome of a discipline that demands a second confirmation before a promotion sticks — the cost of being wrong here was one session, not one drawdown.

Unusual structures

A one-year insurance structure printed on the month-end mark

At five minutes past the close, three legs went up together against zero existing interest, every strike set off the official closing price to the penny, dated one year out: an at-the-money put, a put fifteen percent lower, and a call fourteen percent higher. This is a structured-note or insurance hedge being struck on the month-end mark — the largest new index-tracker structure of the day.

TAPE 16:05:13, 11,504 contracts each, $85.88M total, all opening — Jul-30-2027 747.03P $46.36M (18.28 vol) + 634.98P $19.89M (23.62 vol) + 850.18C $19.63M (14.15 vol). The at-the-money leg priced at roughly 1.14× the index’s own volatility gauge — a year of downside bought close to fair value on the last tick of the month.

A million crash puts, sold

Two clips in the mid-afternoon sold more than a hundred and twenty million dollars of downside across September and October, in over a million contracts. The two are not the same trade — September keeps its far wing, October sells it and leaves the seller uncovered below the lowest strike.

Priced properly this is the most aggressive volatility sale of the cycle, and the pricing is the whole point. The bodies went off at roughly one and two-thirds times the index’s at-the-money volatility; the deepest wing went off at more than three times it. The tail is the expensive part of this surface by a wide margin, and somebody sold it anyway — in exactly the window this desk has been flagging for a week as the deliberately unprotected one.

TAPE 14:22:50 Sep-30 — 635P sold 150,000 / $30.30M / 26.42 vol + 535P sold 300,000 / $18.60M / 39.02 vol + 435P bought 150,000 / $3.15M / 52.43 vol · 14:48:24 Oct-16 — 625P sold 150,000 / $38.55M / 26.94 vol + 525P sold 300,000 / $26.40M / 38.59 vol + 425P sold 150,000 / $4.80M / 51.07 vol. $121.80M received, 1,050,000 contracts, all opening. Against a 15.99 volatility gauge that is 1.68× / 2.41× / 3.19× at-the-money.

A three-hundred-million-dollar bespoke volatility book

Nine structures, most with fractional strikes, all against little or no existing interest, all opening, every one a straddle or four-legged strangle with no directional exposure at all. That is the fingerprint of a single desk printing a dispersion book across a basket on the last day of the month.

The context makes it legible: index volatility closed at sixteen while the average constituent’s volatility closed in the mid-forties — nearly a three-to-one ratio, still elevated even after the July unwind partially closed it. This is the dispersion trade being re-established in size, and it is the cleanest available statement that the professional community expects the next leg to be idiosyncratic rather than directional.

TAPE GOOG 13:45:46 Sep-04 360 straddle, 13,900 each, $37.40M, opening · GOOGL 14:24:42 and 15:13:47 Aug-21 356 straddles, 3,500 each, $7.74M and $7.63M, OI 0, opening · NVDA 15:21:11 Oct-27 200.03 straddle, 16,383 each, $53.24M · INTC 15:18:04 Oct-27 93.03 straddle, 17,622 each, $50.22M (three minutes later, same bespoke date) · MRNA 12:10:38 Sep-18 75.01/55.01 four-leg, 19,450 each, $70.21M (86–87 vol) · SMCI 11:40:13 Sep-18 25.01/28.01 four-leg, 37,036 each, $56.36M (91–97 vol) · TSLA 11:30:43 Sep-18 310.01 straddle, 10,488 each, $45.20M · QQQ 14:32:53 Aug-28 690.53C, 8,554, $15.30M. $343.3M in total, all zero-delta, all opening.

A multi-year chip collar — the trade that dates the whole thesis

This is the single most important structure in the file, because it is the supply story from the top of this report expressed by somebody with size, and it tells you when they think it lands. Long a 2027 in-the-money put, short a 2028 upside call, on a sector tracker that closed marginally green. The put was bought at parity to the at-the-money surface — no premium paid for the strike, which is unusual for a hedge that far out and means the far volatility surface is flat rather than skewed. This is the structural expression behind the sector’s clean negative residue, and it is dated in years, not weeks — and the date is not arbitrary. It brackets the year the Chinese entrant targets volume production of the high-bandwidth product that AI actually consumes, the year the indirect US procurement ban takes effect, and the founder’s own “within a year” estimate for the software moat. Somebody read the same three attacks described at the top of this report and bought the 2027 put rather than the August one.

TAPE 13:45:50 SMH Jun-17-2027 625P bought at the ask, 3,900 contracts, $51.67M, 46.15 vol, open interest 49, opening · 10:38:28 SMH Jun-2028 730C sold at the bid, 2,000 contracts, $18.96M, 46.07 vol, open interest 3, opening. Both legs at roughly the same volatility as the sector’s own August band implies — the strike is priced at parity to the at-the-money surface, a flat far structure rather than a skewed one. Structure-clean residue for the name −$64.18M.

The index pillar book was rolled again, and it cost money

Two structures a hundred and three seconds apart, identical size, same strikes, adjacent expirations: a synthetic long established in September and a synthetic short established in December. Read one leg at a time and this file produces a violently directional index number in either direction. Read as a single trade it is a calendar roll with essentially no net exposure and a sixty-one-million-dollar debit — the desk is paying to pull its position from December back into September.

That roll, repeated three times during the session, is why the index’s day-level number survives structure-stripping at only a tenth of its raw value and carries no directional content whatsoever. The readable content is the roll itself, and its direction is shorter-dated, not shorter.

TAPE 13:27:27 Sep-18 — 7000C bought above ask $111.21M + 7000P sold below bid $8.79M + 8000C sold $1.53M, 2,000 lots (synthetic long) · 13:29:10 Dec-18 — 7000C sold to bid $140.49M + 8000P bought above ask $100.85M, 2,000 lots (synthetic short). Net $61.25M debit, near-zero net exposure. Repeated at 11:04:55 ($90.77M, 750 lots) and 14:55:19 ($71.63M, four legs). Index residue −$90.25M against a −$880.00M raw number — 10%, uncitable as direction.

Amazon’s position moved out three months and up in exposure

The largest opening single-name call purchase in the file went up mid-morning in a November strike. Ninety minutes later the August strike was sold in closing size and a December strike was opened. August out, November and December in, on a fifteen-percent earnings day.

TAPE 11:30:13 Nov-20 240C bought at the ask, 25,000 contracts, $98.25M, 37.1 vol, open interest 2,429, opening · 12:05:52 Aug-21 250C sold to the bid, 16,000 contracts, $35.78M, open interest 71,889 (closing size) · 12:05:52 and 15:31:28 Dec-18 295C bought, 8,750 and 8,650 contracts, $13.34M and $13.18M, opening. Read as one structure this is a roll out and up, not three separate bets; the name’s structure-clean residue after removing matched legs and delta-one prints is +$130.06M, of which Nov-20 is +$107.8M.

The week in one number

Run the same decomposition across all five sessions and the week has a spine.

The index was sold every single session. Monday, Tuesday, Wednesday, Thursday and Friday — through a three-percent liquidation day and a one-and-a-half-percent squeeze day alike — for a cumulative six hundred and eighteen million dollars. No exception, no reversal. Whatever else rotated underneath, the index itself was distributed continuously.

And Friday broke the other pattern. Four consecutive sessions of a quarter to nearly half a billion dollars of net equity buying, then a session that produced seventy-five million — an eighty-three-percent collapse — and the week’s only negative whole-file residue. The bid did not reverse on Friday. It stopped showing up.

WEEK 07/27–07/31 · structure-clean net Mon +$252.6M · Tue +$386.2M · Wed +$281.7M · Thu +$428.9M · Fri −$13.4M. Index leg: −$86M / −$151M / −$151M / −$140M / −$90M = −$618.0M. Equity leg Thursday +$456.5M → Friday +$75.4M. Week’s largest single-name bids: MU +$298.1M, SNDK +$237.2M, MSFT +$194.5M, INTC +$193.2M, LRCX +$142.9M, AMZN +$130.6M.

Four of the week’s five biggest bids closed the month red, and three of them closed red on Friday specifically. The exception is Amazon, whose entire weekly figure is Friday’s earnings reaction. The money that came in during the week went into memory and chip equipment, and memory and chip equipment are where the month ended lower. That is the most useful sentence in the weekly data.

What the Friday commentary is watching

The most useful outside read of the week landed Friday afternoon and its through-line is that nobody knows which way to feel — which, given the evidence, is an honest position rather than a hedge.

On the monthly picture: the index is above a rising monthly average and printed an inside candle — a strong month, then the bears, then a consolidation. The next monthly bar decides: a break up is an expansion, a break down completes a three-bar reversal. The Nasdaq tracker, down more than six percent, closed exactly on its own monthly average — the third bar of that same pattern and the most precarious position on the board. Small caps are down three percent and disconnected above their monthly average, a gap that typically closes within a month or two.

On regime: his conditions matrix flipped from bear-strong to bull-weak. The index reclaimed the gamma flip line by roughly forty points — which he explicitly calls “pretty close” and “neutral territory, path-dependent” — while the Nasdaq composite’s ten-day average is still declining and the twenty-day sits above it, also declining. Above the flip line with a declining short trend is the weakest of the four bullish states.

On volatility: the term-structure ratio inverted intraday on the liquidation day — a classic bounce tell — and has already normalised to the level he associates with consolidation or a pullback rather than continuation. The volatility-of-volatility gauge touched his “haywire” marker and came back. Constituent volatility fell hard but remains near three times the index gauge. Everything that was stretched has already snapped back, which removes the mechanical case for further upside from here.

On rates — his most emphatic point, and the one our own level board agrees with hardest: the dollar came off hard and broke a round number, which is helpful. But the ten-year closed at the top of its range out of a clean flag, and the thirty-year broke to a multi-year high he dates back to 2007. His words: this is probably the most important thing to watch, and if yields keep going and bonds keep cratering it could spell devastation. On Monday’s board the ten-year carries a range reading of 109 — the hottest of any instrument, macro or equity — with its trend line below price and barely a third of a percent of room to the top of its band. Volatility, the dollar and long bonds all carry dead ranges. When every volatility, currency and duration range decays to nothing and the yield range goes to the hottest reading on the board, the market is telling you which variable it intends to move on.

He pairs it with a housing divergence — the housing complex rolling over against a rising index, a configuration he last saw in 2005 — with the explicit caveat that the structure of that market is different now.

On levels: the futures channel is roughly 7,350 to 7,650 and the market is ping-ponging inside it. The cash index shows a series of lower highs against a higher low — a compression triangle. The Nasdaq future broke a key level, ran, and is snapping back like a released rubber band; a further retracement toward that level would be normal. Next week’s implied range is smaller than each of the last two weeks, which he finds striking given the index moved a hundred points in a single session, and he notes the market has not tagged its weekly band in two weeks and is likely to do so in the next three, back to back.

His single most actionable observation is that the index rarely reaches the lower band of his zone document, and that when it did on the twenty-ninth it produced a monster bounce to the upper bands. Monday’s document has the index sitting in the upper half of that same band — the opposite of the condition that produced the bounce.

What actually happens after a levered fund is liquidated

The mechanism was established Thursday: a negotiated block transfer does not print as selling, so the marginal seller did not arrive, it disappeared. But the framing at the top of this report matters here — the fund did not cause the repricing, the repricing broke the fund. Its named holdings were the memory, neocloud and AI-power names sitting directly in the path of the supply story. Treat the blowup as the cause and you conclude the problem left with it. It did not. The question is what history says about the sequel, and the record is more discriminating than the folk version.

Five precedents are close enough to be useful: a commodity fund whose book went to two rivals in September 2006; a credit fund bought by the same acquirer as this one in July 2007; the consortium rescue of a rates fund in September 1998; the meme-squeeze rescue of a short-seller in January 2021; and the family office whose positions were dumped in blocks in March 2021.

The first rule that falls out: the transfer usually happens at or near the low. In four of the five, the index low came within a fortnight of the transfer, and in three of those within days. The forced seller is the marginal seller; removing it removes the supply. That is the bull case, it is the case the headlines are making, and this desk already conceded it on Thursday — the twenty-ninth was a low.

The second rule discriminates, and it is the one that matters: the character of what follows is set by the policy backdrop, not by the blowup. The 2006 commodity case, and both 2021 cases, occurred with policy easy or on hold and liquidity expanding — straight-line recoveries, new highs within weeks. The 1998 case occurred with the central bank actively cutting — a V and new highs. The July 2007 case occurred with credit tightening: a violent chop that took the index roughly seven percent lower over the following fortnight, then a rally to a marginal new all-time high about ten weeks later, less than one percent above the pre-blowup high — and then the top.

This backdrop is the 2007 backdrop, not the 2006 or 2021 backdrop. The curve prices better than seven-in-ten odds of a hike in September and two hikes by December. The thirty-year just broke to its highest level since that same period. Credit is not confirming — the high-yield tracker closed with its trend line above price. And the acquirer in July 2007 was the same firm, on the same calendar day, nineteen years earlier. The coincidence proves nothing; the structural parallel is why it is worth writing down.

A third rule, from the 2021 block case: the liquidated names do not lead the recovery. Those names never came back. That maps directly onto Thursday’s finding that the biggest squeeze winners carried strong distribution signatures on the day they ripped, and onto Friday’s tape, where the same sleeve was sold on thick prints while the index held.

The counterweight, stated fairly: the sell-side’s own historical work notes that when the momentum basket falls more than twenty percent in a month, forward returns have tended to be positive — and a thirty-seven-percent month is far outside that sample. The mechanical case for a bounce is real and it has already partly paid. The distinction this desk has held since Thursday remains the right one: the liquidation phase ended, the repricing phase has not started, and none of the five precedents contradicts that.

The fourteen names, against their own bands

These are the single names the monthly move sheet carries, and every one of them gets read against its own levels each cycle rather than only when it does something interesting. Four observations do most of the work.

The distribution of implied moves has never been this uneven. Netflix is priced for less than nine percent this month and the largest crypto-treasury name for over forty. Between them sit two clear tiers: the megacaps at seven to twelve percent, and the memory-and-silicon complex at nineteen to twenty-eight. The market is not pricing one market in August. It is pricing a quiet index with four or five names inside it carrying event distributions.

AUGUST 1-SIGMA, AS A PERCENTAGE OF SPOT MSTR 41.1 · SPCX 27.7 · MU 23.0 · INTC 20.6 · AMD 19.6 · AVGO 12.9 · TSLA 11.8 · NVDA 11.7 · META 9.5 · NFLX 8.6 · GOOGL 8.2 · AMZN 8.0 · MSFT 7.6 · AAPL 6.7.

Only four of the fourteen have their trend line beneath price. Microsoft, Amazon, AMD and — on the index side — nothing else. The other ten are trading below the line that defines their own uptrend, including every one of Apple, Alphabet, Tesla, Nvidia, Broadcom, Intel, Meta and Netflix. That is the single-name version of the point made earlier about the Nasdaq: the broad index has reclaimed its trend and the names that constitute it mostly have not.

TREND vs PRICE (0803 zone document) Beneath price, uptrend intact: MSFT 408.02 · AMZN 252.06 · AMD 461.72. Overhead, not reclaimed: AAPL 309.16 · GOOGL 361.76 · TSLA 375.39 · NVDA 213.17 · AVGO 425.72 · INTC 112.07 · META 605.29 · NFLX 79.89. Declared gap — no zone row exists: MU, SPCX, MSTR; no band-relative claim is made about those three.

Three names are pinned at band extremes and they are the ones to watch first. Apple closed beneath its zone floor with a dead range — twelve percent of theoretical upside against half a percent of downside, which is the arithmetic of a name that has already moved. Microsoft and Amazon closed above their zone ceilings carrying the two hottest range readings on the entire board, with roughly twenty and nineteen percent of downside room beneath them. Extension is not a sell signal, but it tells you where the asymmetry has gone.

And the flow splits the roster cleanly in two. Amazon and Alphabet were bought and dated out to November, 2027 and 2028. Nvidia, Intel, Broadcom and Meta produced small positive residues that are mostly artifact or unsided. Everything in the memory-and-silicon tier was sold, and Microsoft's negative number is the roll-up described earlier rather than a short. Nothing on this list was bought for August specifically.

STRUCTURE-CLEAN RESIDUE · NEXT WEEK 1-SIGMA BAND AMZN +$130.1M (259.73–283.43) · GOOGL +$18.4M (342.37–369.89), with GOOG a further +$18.2M · NVDA +$13.7M, artifact-flagged and unsided (191.42–210.08) · INTC +$13.6M, unsided (80.87–99.53) · SPCX +$12.6M (88.83–127.91) · META +$11.9M, unsided (530.88–582.54) · AVGO +$0.5M (366.73–411.83) · NFLX −$1.5M, artifact (68.90–74.52) · MSTR −$8.3M, unsided (84.80–101.76) · AMD −$16.0M (419.92–532.38) · AAPL −$19.5M, artifact (297.38–320.44) · MU −$34.2M, clean, front months sold (723.20–922.86) · MSFT −$48.8M (447.14–482.30) · TSLA −$52.0M, unsided, almost entirely a 2028 structure (293.95–328.45).

Two of those bands deserve naming on their own. Micron is priced for a twenty-three percent month with its front three expirations net sold and its darkpool contracting — the widest distribution in the complex sitting under the most negative near-dated flow. And Alphabet is the only name on the roster with three consecutive sessions of independent buying in both share classes, yet its trend line still sits above price with under four percent of room to the top of its band. Those two are the roster's cleanest bear and bull tests for August, and both resolve inside the levels above.

The level board for August

All three timeframes — the day, the week and the month — are now committed to the permanent record and every subsequent report will be written against them, rolling each Friday evening and each month-end.

The August band on the index runs roughly two hundred and seventy points either side of the close, with the two-standard-deviation floor near the round number a full seven percent below. The week ahead carries a band of only a hundred and five points — the tightest in three weeks, into the biggest data day of the month. Monday’s daily band is about forty-six points.

The single cleanest structural statement survived month-end unchanged: the broad index has reclaimed its trend line and the Nasdaq has not. The Nasdaq’s sits about two and a half percent above Friday’s close. Add the technology sector fund, the communications sector fund, the largest chip designer and the largest foundry — every one of them trading beneath the line that defines its own uptrend — and you have the level-based definition of what “bull weak” means. The sectors whose trends are intact are financials, industrials, energy, healthcare and regional banks. That is a defensive and cyclical leadership set, not a growth one.

Three names sit at band extremes worth naming. The largest consumer-electronics name closed below its zone floor with a dead range — twelve percent of theoretical upside against half a percent of downside, which is the arithmetic of a name that has already gone. The largest software name and the largest online retailer both closed above their zone ceilings with the two hottest range readings on the board and roughly nineteen and twenty percent of downside room. Extension is not a sell signal; it is a statement about where the asymmetry has moved.

The August map, and what changes it

Base case, roughly half the weight: a compression that resolves on the first Friday. The index sits above its own trend and below the Nasdaq’s, inside a three-hundred-point futures channel, with the tightest weekly band in three weeks, into the biggest data release of the month, with dealers short gamma at that exact expiration. A compressed band plus short dealer gamma plus a dated catalyst is the configuration that produces a gap rather than a drift. Direction is genuinely two-sided: the positioning is short exposure, which means a soft employment print with yields backing off is the squeeze scenario, and a hot one is the June replay.

Bear path, roughly thirty percent: the June sequence, smaller. Delta exposure drains from Friday’s peak toward and through zero over five to eight sessions, a hot payroll or a hot inflation print gaps yields higher, the ten-year clears the top of its band and the thirty-year extends its breakout, and the index works down through its zone mid toward the floor and the strike where six and a half thousand puts were bought on Friday. That floor and the August one-standard-deviation floor are the same neighbourhood, a little over three percent below Friday’s close. The amplitude should be smaller than June’s because the crowded length has already been liquidated — the thirty-seven percent factor drawdown already happened.

Bull path, roughly twenty percent: the 2007 shape — which is not the same as being wrong. The liquidation low holds, the marginal seller stays gone, the post-drawdown statistic asserts itself, the Nasdaq reclaims its trend, and the index grinds to a marginal new high over six to ten weeks. Note what that path looked like in the closest analog: one marginal new high, then the top. A bull path in a tightening backdrop is not a bull market.

And one test that sits outside all three paths, because it decides the bigger story rather than the month. The supply thesis needs a price war, and the price war has not started — Chinese memory is reaching the market at prices that still track the incumbents. The thing to watch through August is not the index; it is whether contract pricing in that complex starts to break versus spot, and whether the equipment names begin to discount a domestic tool supply. Neither is an August event. Both are what the multi-year collar is paying for, and if either turns, the 2027 hedge stops being a slow burn and starts being the trade.

What changes the read, in order of importance. Upward: the Nasdaq closing above its trend line with the technology cumulative flow line turning back up, plus the ten-year backing off the top of its band. Downward: the late-August index protection block being closed into strength rather than added to — on Friday it was added to, which is why the current read holds — or the high-yield tracker losing its zone floor with its trend still overhead.

On sizing. One at-the-money session for the index is roughly six-tenths of a percent, so a one-and-a-half-times invalidation is about sixty-eight points. For the largest memory name it is nearly six percent per session — a two-percent stop on a ninety-volatility name is noise, not a stop. Set August invalidations off the daily band, not off chart levels. The names that broke in July broke through every chart level on the way down.

Bottom line

July was the month the scarcity trade got attacked from three directions — a competitor's listing, a domestic tool-manufacturing headline, and an efficiency argument that needs less of the scarce thing — and the most crowded factor in global equities lost a third of its value as a result. The levered fund that failed afterwards was the leverage giving way on a repricing already in motion, which is why its disappearance solved nothing. The index ended the month flat because a collapse and a rotation happened to cancel, and the last session was spent marking that flat close on a tape that was being sold. The largest positioning reading of the month arrived on the quietest session of the week because a month’s worth of expiring exposure converged to its terminal value — a receipt for what already happened, not a signal about what comes next. The same thing happened at the end of May, and eight sessions later the index was four and a half percent lower — but it took a dated macro event to get there, and this time that event is on the calendar for the first Friday of August, on the exact expiration where dealers are shortest.

Underneath, the evidence is one-directional and it has been all week: the index itself was sold in every single session, the trackers were distributed on rising volume at month-end, the mega-cap prints that look like buying are the rebalance auction, and the bid that showed up Monday through Thursday simply did not appear on Friday. Against that, exactly two things were genuinely bought: three months of dated upside in one retailer, and three hundred million dollars of single-name volatility with no direction attached at all.

The liquidation ended. The repricing has not started. And the entire options book — both sides of it — was pulled forward into the next three weeks in the last ten minutes of the month, net short. Whatever August is going to be, it is going to happen before the third Friday.

One footnote worth its own line, because it is the closest thing to a control experiment this desk has. The identical structure printed at the identical minute on the last trading day of May — same four-leg shape, same size class, front monthly bought against next monthly sold, the sold legs marked opening. The difference is the strike placement, and it inverts the sign completely: in May the front legs were struck below spot, the position carried roughly $2.8B of long delta, and it was a $131M debit. On Friday the front legs are struck above spot, the position carries about $1.1B of short delta, and it is a $55M credit. The May version was the wrong side of what followed. Whoever runs this program has reversed its sign at the next month-end — which is a more honest signal than anything anyone says out loud.

INDEX · HEDGE The August index put strike accumulated Friday in nine tranches — 9,478 net contracts, ~$36.2M net — sits between Monday’s zone floor and the August 1-sigma floor, ~3.2% below the close. Dealers carry −$2.44B of put delta into the 08/07 employment report at that expiry, and the same book is ~$957M short delta and short $7.7M of vega per point. Own the date, and note that the professional structure is short upside, not long it. Invalidate on a close above the zone high with the Nasdaq trend reclaimed.

AMZN · LONG (dated) Nov-20 240C, 25,000 bought at the ask for $98.25M at 37.1 vol, opening — the only dated upside anyone paid for on Friday, with 67% of the sold calls struck in the money (rolling, not shorting). First confirmation only. Invalidate on a close back inside the zone at 267.21 (≈1.5 daily bands).

GOOGL / GOOG · WATCH, 3rd confirmation Jan-2028 330C bought at the ask $13.08M at 37.15 vol, plus Jun-2027 320/325C $10.04M at ~35 vol — dated, lifted at the offer, third consecutive independent session across both share classes. Discount the ~$52.8M of zero-delta straddles in the same tape; that is vol, not direction. Trend still overhead — needs a close above it to promote.

SMH · STRUCTURAL SHORT (multi-year) Jun-2027 625P bought $51.67M at 46.15 vol against Jun-2028 730C sold $18.96M at 46.07 vol — a collar struck at parity to the at-the-money surface, no premium paid for the strike. This is a 2027 view, not an August view; do not size it as a trade.

DISPERSION · WATCH $343.3M of bespoke zero-delta single-name straddles (GOOG, GOOGL, NVDA, INTC, TSLA, MRNA, SMCI) dated Sep-18 and Oct-27, printed at month-end against a ~2.78× constituent-to-index volatility ratio. If the same fractional-strike signature prints again next week it is a campaign, not a month-end adjustment.

EWY · LONG CLOSED Thursday’s tactical Korea long is off after one session: 18 thick prints, volume −24.1%, net −$304.96M, price −2.55%, and puts bought in the 08/07 and 08/14 expirations. Second confirmation failed and inverted. Structure-clean options residue −$25.34M with puts bought $20.9M against $12.1M sold. No short; absence.

MU · DO NOT SHORT, DO NOT OWN Front-month implied volatility 91–93 against roughly 80 at the money — elevated and flat, which is squeeze-shaped, not top-shaped, so shorting remains wrong. But the front three expirations are net sold and the darkpool net came on volume −44.9%. The constructive case needs the front months to stop being sold first.

Scorecard — grading Thursday’s calls

Holding: the “a low, not the bottom” read (index up, but breadth negative, small caps down, the squeeze sleeve distributed on thick prints); the avoid on the AI-financing complex; the late-August protection block as the regime test, which fired correctly when the block was added to; and the hedge calendar, which is now not merely confirmed but explained — the crash premium sold Friday landed in precisely the window the calendar flagged as deliberately unprotected.

Wrong: the Korea long, which inverted in a single session and is now closed. And the removal of Alphabet from the watch list on a third consecutive failure was premature — both share classes took an independent bid for a third straight session, the money went into 2027 and 2028 strikes at the offer, and the name closed up nearly seven percent. It returns to watch, and the lesson is that a third-failure removal needs a flow check, not just a price check.

Mixed: the software reduce was right in structure and wrong in session — the name rose three percent, but the sold-call composition still reads as rolling up rather than shorting, and its apparent darkpool accumulation is the closing cross. The memory do-not-short held; the basket did not lead.

Sources

Friday’s full options tape (42,331 rows, $16.51B gross) and the prior four sessions, run through the same five-stage decomposition: duplicate signatures removed, delta-one stock substitutes removed, same-second matched structures removed, then split by expiration — only the surviving figure is quoted anywhere above. Friday’s darkpool file (3,277 names, $199.58B) split by print count with the day-over-day volume overlay applied to every cited name. Both dashboards read panel by panel: 22 options panels and 15 darkpool panels, with the expiration map, timeline and dealer ledger bar geometry extracted pixel-by-pixel against calibrated axes rather than eyeballed. The 618-name end-of-day census supplied every price and percentage move. Forward levels from Monday’s expected-move sheet, the 142-row zone document, the week-ahead sheet and the August monthly sheet — the last two now committed to the permanent record; several two-standard-deviation cells on the weekly sheet do not equal close plus or minus twice the move and were recomputed rather than copied. The Friday afternoon commentary and its 38-slide deck read in full. Six comparable reports from the late-May and early-June window pulled from the archive for the delta-exposure analog. Outside sources: the Shanghai listing's raise, first-day gain, market capitalisation, cost position, high-bandwidth timeline and US procurement-ban schedule; reporting that Chinese modules are pricing in line with the incumbents rather than below them; the 2026 wafer-capacity comparison; a leaked four-hour investor call from the founder of China's leading open-weight AI lab, read in full; the sell-side momentum-basket drawdown figures and the post-drawdown forward-return note; reporting on the liquidated fund’s size, leverage, returns and the acquirer; the 2007 precedent’s size, losses and transaction; the record-close chronology; and the August US data calendar. Gaps declared and not filled: no sentiment gauge dated Friday exists on disk, so that input is carried one session stale and is not counted as a fresh vote; and thirteen names cited in the single-name work carry no row in the zone document, so no band-relative claim is made about any of them. No file was moved, renamed or reorganized.