Daily Report — 08/03/26 · “The Bid Was Real. The Room Was Not.”

The bid was real. The room was not.

Start with what is not in dispute. SPY closed 757.67, up 1.42%. QQQ closed 700.07, up 1.76%. MSFT rose 4.93%, META 6.02%, GOOGL 4.88%, AMZN 4.58%. Dealer delta across the market printed its highest reading of 2026, roughly three and a half times its own five-day average. Every index and every mega-cap returned an accumulation verdict on the price-and-volume read. This was a genuine institutional bid, and anyone who spent the session arguing it was a bear-market rally was arguing with the tape.

Now the part that matters more for Tuesday. The move ended 15 cents above the top of the week’s expected range — on the first of five sessions. It ended with SPX 0.12% from the top of its zone and 3.79% from the bottom, which is better than thirty to one against on room alone. And it ended sitting on the single strike with the deepest negative dealer gamma on the board. The rally did not run out of buyers. It ran out of band.

TAPE Whole file 43,775 raw rows → 42,383 after removing duplicate signatures, $20.46B gross. Matched call-and-put legs $3.47B (17.0%) · deep-in-the-money stock substitutes $2.61B (12.8%) · same-second same-size multi-leg packages $7.14B (34.9%) · structure-clean directional residue $7.23B, 35.3%. Nearly two thirds of the premium on this tape carried no directional information at all.

Two thirds of the tape was financing, and the index was the worst offender

SPX alone was $9.79B of the $20.46B file — 47.9% of all premium. Of that, 67.1% printed inside same-second, same-size multi-leg packages: 7000 calls against 8000 puts, repeated across the September, December and December-2027 expirations, in sizes from 200 to 7,000 lots. Those are jelly rolls and boxes. They are how large institutions finance a book. They are not a view.

Once that is removed, SPX’s structure-clean residue is 11.6% of its gross — below the threshold at which any directional statement is permitted. The honest description of index options flow on 08/03 is not bearish and not bullish. It is absent. SPY’s structure-clean net was +$1.2M and QQQ’s was −$2.0M, on a day the two vehicles traded more than $1.35B of gross premium between them. That is zero, twice.

This matters because there is a standing temptation to read the index tape as the market’s opinion. On this session it was a balance-sheet operation. The September 7600 line carries 53,261 calls against 48,671 puts — a ratio of 0.91, which is the signature of a financing box, not a wall. The open-interest panel that shows 297,851 September 7000 puts and no corresponding calls is showing one leg of a two-leg trade. It makes a loan look like a fortress, and it did exactly that once before.

TAPE · the largest package on the file 15:47:56, 7,000 lots, four legs, $1,351.2M — Dec-17-2027 7000C sold below bid $888.40M (21.33 vol) + Dec-17-2027 8000P bought to ask $428.75M (17.57 vol) + Sep-18 7000P bought $20.05M + Sep-18 8000C sold $14.00M. Repeated at 15:18:56 on 5,950 lots ($760.9M) and 5,850 lots ($754.5M), at 16:01:57 on 2,000 and 2,050 lots, and at 16:05:53 across eight legs. $6.57B in packages above $50M each. Three further identical $102.9M packages at 13:12 bought and sold Jan-2027 4200 and 4250 calls at 45% in the money against open interest of 3 and 10 — cash financing dressed as a call trade.

Where the real money went was single names, and there it was one-directional. Structure-clean nets: AMZN +$103.57M, NVDA +$32.75M, GOOGL +$12.82M, AAPL +$10.73M, MU +$10.07M, KWEB +$9.80M, META +$2.85M. Money did not leave equities on 08/03. It left the index vehicle and went into names. Hedge funds independently posted their largest net purchase of US technology since December 2022, concentrated in software and semiconductors — the same signature from the opposite direction.

MSFT printed red on a 4.93% day. It was a roll, and it is bullish.

Microsoft was the second-largest negative tile on the day’s flow map after rising 4.93%. The panel nets sold-call premium against bought-call premium and colours the result, and MSFT sold $291.3M of calls against $196.7M bought. On the surface that is distribution into strength.

It is not, and the moneyness settles it before the pairing does. Of that sold-call premium, 64% was struck in the money and a further 33% within fifteen percent of spot. Only 3% sat beyond fifteen percent out of the money. In-the-money call selling is a roll or a harvest. Near-money selling is an overwrite against stock. Ninety-seven percent of Microsoft’s sold calls were one of those two things.

Then the pairing — and this is a matched structure, not two separate trades. Four times during the session, in the same second and at identical size, someone sold the August-21 460 call and bought the October-16 510 call. The open interest tells you which is which: the August 460 line carried 59,187 contracts of established position; the October 510 line started the day at 253 and finished with volume above 53,000, every print flagged as opening. 21,435 contracts were moved fifty dollars higher in strike and two months further out. That is an institution paying premium to extend upside exposure, executed in four tranches so as not to move the market.

TAPE · the roll, four tranches · matched same-second legs, counted as one structure 10:46:32 — Aug-21 460C sold at bid 7,435 lots $22.79M (OI 59,187) against Oct-16 510C bought at ask 7,435 lots $13.31M (OI 253, opening) · 13:34:32 — Aug-21 460C sold below bid 10,000 lots $32.50M against Oct-16 510C bought above ask 10,000 lots $19.40M (opening) · 14:50:22 — same pair, 2,000 lots, $6.80M against $4.06M · 15:49:44 — same pair, 2,000 lots, $6.42M against $3.87M. Separately at 12:37:08 a 465 straddle printed 10,000 by 10,000 for $33.00M, both legs opening against zero open interest — pure volatility, zero direction, correctly excluded from the directional figure.

Three independent sources agree. Microsoft’s dark-pool net was +$4.54B on participation up 33.2% — the only large positive net on the entire board printing on rising volume, which is the difference between a campaign and a single transfer. Its three-day cumulative slope is rising by $5.49B. And the pipeline’s own fifteen-day tag, which reads distribution, is flagged as contradicted by price and as stabilising, which removes it from the bearish column outright.

Read: the largest identifiable holder of Microsoft upside now owns October 510 calls instead of August 460 calls. The near-term risk is digestion — MSFT closed through the top of its zone, through its two-standard-deviation daily band, and spent 65.3% of a full month’s expected upside in one session. But nothing in this tape describes a seller.

CRWV printed red on a 19.49% day. That one is real.

CoreWeave rose 19.49% and also printed a negative tile. Same colour, opposite meaning, and the test that separates them is the same one: where were the sold calls struck?

On a structure-clean basis, after matched legs and stock substitutes are removed, seventy-four percent of CRWV’s sold-call premium sat beyond fifteen percent out of the money — against three percent for Microsoft. Three prints carry almost all of it, and each satisfies every condition that distinguishes genuine short-upside selling from a roll or an overwrite: struck far out of the money, flagged opening, and printed in size exceeding the existing open interest. Roughly 83,000 contracts of September upside between the 110 and 115 strikes were sold into the spike.

TAPE · the ceiling 10:00:40 Sep-18 110C sold below bid, 26,980 lots, $11.16M, 106.69 vol, open interest 8,011, opening · 10:17:27 Sep-18 110C sold below bid, 25,285 lots, $9.66M, 106.31 vol, same line, opening · 10:30:11 Sep-18 115C sold at bid, 30,701 lots, $9.52M, 105.48 vol, open interest 8,143, opening. All three print size above open interest and 36% to 44% above spot. Structure-clean net for the name −$26.56M on an 87.2% directional residue — one of the cleanest reads on the board.

The darkpool tape does not corroborate this, and that matters more than a confirming number would. CRWV’s reported dark-pool net was −$367.7M, but 98.8% of that is the 16:00 auction cross — seven late prints worth −$363.3M, whose side tag is mechanical noise rather than a decision. Strip the cross and the intraday net is −$4.4M, essentially flat, and the cumulative series sits mid-range rather than at a new low. So the ceiling stands on the options structure alone: September upside written in size above open interest is the trade, and it needs no darkpool confirmation. The naive “new low on a 19% up day” read was the cross tag masquerading as distribution — the exact false-divergence this framework flags rather than buries.

The volatility context sharpens rather than softens it. Those calls were written at 106 implied vol, with the far wing pricing at 0.96 times the at-the-money contract on the same expiration. Sellers were harvesting an extreme premium at a discount to the body, not paying up for protection. There is an institutional ceiling at 110 to 115 through September 18, and roughly 83,000 contracts of supply sit under it. Anything that requires CoreWeave above 115 by September is trading against that.

Why names that closed green sit in a red panel

The put-volume-change panel showed META, BA and OWL in red on a session all three closed higher. The colour is not a verdict. It is the identity of the chart — green is the calls panel, red is the puts panel, and every bar in each carries the same colour, which no directional encoding would produce. PFE and NKE appear in both panels on the same day, and one stock cannot be up and down at once.

The companion dashboards prove it outright: the all-green calls table carries SPY at −$36.56M and SPX at −$133.19M of net premium, while the all-red puts table carries SPX at +$112.06M and SPY at +$26.39M. The shading has no sign attached to it.

Once the side data is applied, the intuition inverts completely. On the up day the put volume in those names was mostly sold: OWL puts 35.3% bought, BA 42.6%, META 52.3%. Selling puts into strength is a bullish footprint, and the structure-clean options nets agree independently — OWL is net positive, carried by puts sold roughly two and a half to one against puts bought, and BA is net positive with calls bought better than two to one. The panel ranks names by change in volume, which is a magnitude with no intent attached. A put-volume spike can be protection, premium harvest, or one leg of a straddle. The sharpest illustration on the same panel: VIX puts printed 71.7% bought — a bet that volatility falls further — inside a chart coloured red.

Where price sits against its own bands, and what each one does to Tuesday

Daily. Measured against the band the market carried into the session, SPX finished 19.38 points through its two-standard-deviation upper, SPY 1.44 through, and RUT 0.66 through; NDX, QQQ and IWM cleared the one-sigma line. Among single names MSFT, META and GOOGL all closed beyond two sigma. A two-sigma break is roughly a one-in-twenty session and six instruments did it together. The consequence is that Tuesday opens with Friday’s distribution already void — every level derived from the prior close is stale, and the forward daily band is the only usable one: SPX 7636.17 above and 7564.83 below at one sigma, 7671.84 and 7529.16 at two.

Weekly — this is the binding constraint and it was spent on day one. SPY closed 757.67 against a weekly upper of 757.52. SPX cleared its weekly ceiling by 6.13 points, MSFT by 5.37, GOOGL by 3.62, META by 7.90. These bands were struck off the 07/31 close and were not reissued, so the market is now four sessions into a surface it exceeded immediately. The consequence is blunt: the weekly band now caps rather than guides for the remaining four sessions, the entire lower half of it is dead space through Friday, and any upside from here is being drawn from an allowance already spent. This is the single strongest argument against chasing on Tuesday, and it is geometry rather than opinion.

Monthly — intact, but the leaders have spent their allowance. Nothing breached the August band. But MSFT consumed 65.3% of a full month’s one-sigma upside on the first trading day, META 63.5%, GOOGL 59.5%, AMZN 57.4%, SPX 40.7%, QQQ 30.1%. SPX’s August range runs 7761.58 above and 7217.86 below. The consequence is that the remaining room genuinely lives here — about 2.1% to the ceiling — but the four names that would have to deliver it have already used two thirds of theirs. Fresh leadership is required for the month to extend, and the obvious candidate did not turn up on Monday.

Quarterly — the two boards disagree, and that disagreement is the trade. SPX is +1.35% for the quarter while QQQ is −4.93% and NDX −4.95%. The consequence is that the Nasdaq complex remains roughly 6.6% below its quarterly anchor even after a two-sigma up day — July’s damage has not been repaired and one strong session did not repair it. The only quarterly breach anywhere on the board is TLT, which closed 82.19 against a quarterly floor of 82.50, through it. And where the month’s ceiling meets the collar supply is worth marking: the quarterly collar carries short calls at 7890 against long puts at 7090, so 7761 to 7890 is where the monthly ceiling and the collar’s short-call supply stack on top of each other. That band, roughly 2.1% to 3.8% above Monday’s close, is August’s structural ceiling.

One more thing the bands did on Monday that has not happened in weeks. Trend validity was re-established across the entire index complex in a single session — the range reading on SPY roughly doubled from 74.5 to 111.8, SPX from 72 to 112, IWM from 35 to 102, DIA to 127.5. After a July in which most of these readings were too low to use, the daily bands are worth trusting again. The exception is AAPL, whose trend reading is outright reversed at −20, which makes its published upside an artifact and is consistent with it being the only red mega-cap on the session.

The names carrying levels, and where each one stands

The monthly sheet levels these fifteen, and all fifteen are marked here because the levels exist for all of them. Four closed beyond their weekly ceiling, one closed red, and the semiconductor group as a whole did not move.

MONTHLY-SHEET ROSTER · 08/03 closes MSFT 487.65 (+4.93%) through the weekly ceiling by 5.37, two-sigma daily breach, 65.3% of the monthly allowance spent · META 590.24 (+6.02%) through weekly by 7.90, two-sigma breach, 63.5% spent · GOOGL 373.51 (+4.88%) through weekly by 3.62, two-sigma breach, 59.5% spent · GOOG 372.47 (+4.44%) tracked as a separate slot, same direction as GOOGL on the session · AMZN 284.02 (+4.58%) one-sigma daily, 57.4% of the monthly spent · AAPL 303.42 (−1.78%) the only red name, trend reversed, band held · NVDA 206.64 (+2.93%) one-sigma daily, band held · TSLA 322.08 (+3.49%) one-sigma daily · AVGO 392.23 (+0.76%) inside its band, no breach · AMD 484.64 (+1.78%) inside its band, reports 08/04 after the close · INTC 91.00 (+0.89%) inside its band · MU 829.50 (+0.79%) inside its band, sitting just above the 800 line · NFLX 73.33 (+2.26%) inside · SPCX 114.53 (+5.68%) inside · MSTR 94.86 (+1.69%) inside.

Read down that list and the shape of the session is obvious. The four names that broke their weekly ceilings are the four that reported earnings the week before. The five that did not move — AVGO, AMD, INTC, MU and by extension the whole semiconductor complex — are the ones that would have to carry the next leg. The move was a post-earnings re-rating of four companies, not a market-wide risk appetite.

The semis did not come, and that is the most important negative on the board

It is being written elsewhere that Monday was the session where semiconductors, software and the hyperscalers finally moved together for the first time in months. At the resolution that matters, they did not. SOXX rose 0.55%. SMH rose 0.91%. MU rose 0.79%, AVGO 0.76%, TSM 0.46%. Against that, IGV — the software vehicle — rose 3.00%, and the mega-caps rose between 4.58% and 6.02%. That is a spread of more than four percentage points inside one sector on one session, and 12.45 points between the best and worst technology names on the board.

This is not a quibble about magnitude. The open question going into this cycle was whether July’s flat index had been a dispersion artifact — the index pinned while its constituents tore apart — or a genuinely quiet market. Broad simultaneous participation would have settled it in favour of the second reading. Participation was not broad. Dispersion printed one of its widest single-session readings on the day SPX broke two sigma.

There is a hard number for it this cycle, and it comes off the volatility surface rather than from commentary. At the August-21 expiration, QQQ’s at-the-money contract prices at 21.58 implied vol against SPX at 11.39 — a ratio of 1.89, where the normal relationship between the two runs between 1.15 and 1.25. Index volatility is being compressed while Nasdaq volatility is not. That is the dispersion regime measured directly, roughly 50% to 60% wider than usual, and it is the same mechanism that lets the index sit still while individual names move violently.

TERM STRUCTURE · at-the-money implied vol by expiration SPX — 0DTE 9.61 · Aug-07 10.40 · Aug-21 11.39 · Sep-18 13.17 · Dec-18 14.68 · Dec-2027 17.52. QQQ — 0DTE 23.23 · Aug-07 23.49 · Aug-21 21.58 · Sep-18 22.05 · Dec-18 23.50. SPX is in steep contango; QQQ is flat to inverted. Rolling an index hedge forward through the SPX surface means buying dear and selling cheap on every roll — a standing hedge there is a subscription, not a position. Own the convexity in a window and retire it outside one.

Index downside is expensive, and the wings are where you are being charged

Because the phrase “protection looks cheap here” is about to be everywhere, it is worth pricing it. At August-21, SPX puts struck 10% to 20% below spot carry 27.94 implied vol against 11.49 at the money — 2.43 times. At September-18, puts 20% to 35% below spot carry 34.36 against 13.19 — 2.60 times. Anything beyond twice the at-the-money contract means you are paying up, and both of those are.

The instruction that follows is the one this framework had to learn the expensive way: buy the body, sell the wing. The mispriced contract in this surface is the near-the-money one, and it is mispriced downward precisely because the structural volatility-supply complex — overwriting funds, income programmes, structured products — sells it on a schedule. Expressing a downside view through a far-out-of-the-money put here means paying more than double for the privilege.

Single-name upside, by contrast, is being sold at a discount to the body

The single-name surface says the opposite of the index surface, and this is the part of Monday’s tape most worth carrying forward. Wing-to-body ratios: CRWV 0.96, CAT 0.97, MU 1.05, PLTR 1.08, NVDA 1.28. In four of those five the far upside strike prices at or below the at-the-money contract on the same expiration.

Flat skew of that shape is not a neutral observation. It says nobody is hoarding upside — it is being written, in size, for yield. That is the condition under which a move becomes mechanically self-reinforcing, because the dealers who bought those calls have to hedge them, and the institutions that wrote them have to chase if the strikes come into play. It is also, on this framework’s own record, the configuration that marks a squeeze rather than a top — steep skew marks tops, flat skew marks squeezes.

Ten names that still have squeeze fuel

This screen is mechanical, not editorial. A squeeze requires dealers to be short upside, so that a rally forces them to buy. The measure below is therefore opening out-of-the-money call buying where the trade size exceeded the existing open interest — the print that leaves a dealer short gamma above spot — taken from the structure-clean residue only, after matched legs, stock substitutes and same-second packages are removed. It is then weighted against how flat the name’s skew is, and against how little the name has already moved. A stock that ran 19% on Monday has spent its fuel; a stock that ran 0.79% has not.

SQUEEZE FUEL · opening OTM call buys above open interest, structure-clean 1. MU 829.50 (+0.79%) — $8.4M across 23 prints, skew 1.03 · 2. SOXX (+0.55%) — $11.6M, the largest on the file, structure-clean net +$36.4M · 3. SKHY 142.72 (−0.70%) — $3.8M, skew 1.02 · 4. AVGO 392.23 (+0.76%) — $2.2M, skew 1.04 · 5. AMD 484.64 (+1.78%) — $7.0M across 48 prints · 6. CAT 830.03 (+1.87%) — $4.6M, skew 0.97, flattest on the board · 7. MRVL 193.77 (+3.31%) — $2.1M, dark pool +$717.5M on volume +41.5% · 8. SNDK 1,288.03 (+6.03%) — $2.4M, skew 1.01, reports 08/05 · 9. BE 218.32 (+6.08%) — $9.0M, skew 0.98 · 10. LITE 779.89 (+9.24%) — $2.7M across 12 prints, structure-clean net +$12.1M.

Seven of the ten are semiconductors or semi-adjacent, and the four best-ranked are precisely the names that did not move on Monday. That is the whole finding. The one part of this market where upside was sold rather than bought, where the skew is flat, where dealers are short gamma above spot, and where the move has not happened yet, is the complex that fell 21% in July with roughly 66% realised volatility — its worst month since October 2008. SKHY is the sharpest version: the only name on that list that closed red while institutions bought its upside.

Two honest qualifications. BE’s figure is ranked on fuel and skew alone — 48.9% of its options gross carries no side tag, so its net is not citable and it is listed for completeness rather than conviction. And SOXX has no entry in Monday’s per-name data set, a genuine gap that is declared rather than filled from another date, so its row rests on flow with no price claim attached.

The counterweight is not a footnote, and it sits outside the options structure entirely — this is fund assets, not residue. Leveraged and inverse exchange-traded assets of roughly $21B are concentrated in exactly these names — SK Hynix around $5.5B, MU around $5.1B, NVDA around $4.8B — and SOXL took in $6.9B in July, its largest monthly inflow on record. That is crowding, and crowding is what makes the squeeze violent in both directions. It caps conviction rather than adding to it. A squeeze list is not a buy list; it is a map of where forced buying would have to occur if a catalyst arrives. Without a dated catalyst these positions simply decay — a wall of open interest is fuel, never ignition. The dated catalyst is Friday.

The oil move is not deflation, and calling it deflation would have inverted the read

Crude fell about 5.3% to 80.03, closing through its two-sigma lower band, while the dollar held 99.97. Run through the standard dollar-and-oil matrix, that combination lands in the deflation quadrant — bearish for metals and for equities alike. It sat directly against a 1.42% up day, and it was the sharpest unresolved conflict on the board coming into this cycle.

That framing does not survive the rest of the board, and it is withdrawn. The matrix assumes a demand-driven oil move. This was a geopolitical risk-premium unwind: the Iran strike was paused pending a deal requiring the Strait of Hormuz fully reopened. Five things separate the two cases. Crude broke two sigma in a single session, and demand destruction walks a band down rather than gapping through it. Crude’s trend-validity reading collapsed from 48 to 22, which says the prior uptrend — the war premium — is what is dying, not that a downtrend is forming. Credit did not confirm: HYG fell 0.21%, where a genuine deflation impulse widens high yield. Metals did not sell: GLD +0.05%, SLV +0.19%, flat rather than liquidated. And the trigger is documented, dated and political rather than an inventory print.

The verdict is disinflationary and risk-positive. Lower energy costs with credit stable and metals flat is a tailwind, and it is genuinely part of why Monday looked the way it did. But the caveat is load-bearing rather than decorative: the pause is explicitly conditional and Iranian officials deny having requested it. This is a re-arming risk, not a resolved conflict. A premium that came out in one session can go back in on one headline. Treat the tailwind as live and revocable, and do not build anything whose thesis requires it to persist.

Of the three macro inputs that moved against the tape on Monday — oil down 5.3%, the dollar’s trend reading collapsing to near-dead, the 30-year easing from 5.275% to 5.231% — two resolve as supportive once that adjudication is made, and the dollar’s move is genuinely neutral. The one contradiction that survives is credit. HYG’s trend line sits above price, USHY printed −$663M on a volume expansion near ten-fold, and TLT printed −$370M. Credit and duration were both sold into a risk-on equity session, and that is the reason to keep the fragility count high.

Japan: the rescue is aimed at the wrong risk

The question worth answering is whether the US Treasury’s reported intervention to support the yen defuses the August-Japan volatility risk that has marked recent years. It does not, and the intuitive version of this is backwards.

August 2024 — the reference event — was not a yen collapse. It was a yen surge. The dollar fell against the yen from roughly 161 to 142 and the volatility gauge printed 65. A carry unwind means buying back the funding currency, so the damage arrives when the yen gets stronger, not weaker. Which produces an uncomfortable inversion: a yen rescue that works is the August-2024 risk. A yen rescue that fails is not. Supporting the currency pushes it in the direction that historically hurts.

On the reported scale — roughly $5B to $10B, relayed through wire reporting, with no primary document available and three commentary voices asserting it — this is a signal purchase rather than a level defence. Japan’s own 2022 and 2024 campaigns each ran several times larger and each decayed within weeks; the yen only genuinely turned in 2024 when the Bank of Japan actually raised rates. The one thing that would make an intervention stick is a policy change, and the bind Japan is in — defend the currency and damage the government bond market, or protect the bond market and let the currency slide — is exactly what rules that out.

The transmission channel to US equities is not the yen level. It is correlation. A carry unwind raises implied correlation; the index volatility complex is structurally short correlation through dispersion trades, overwriting and structured products; index volatility then rises mechanically and volatility-targeting strategies reduce exposure into it. So the series to watch is the three-month implied correlation index, not the currency pair. That is one series, it is the lowest-cost test available, and it is the same series the dispersion question turns on — which means these are not two questions but one.

Practically, for August 2026: the modal path is quiet decay of the intervention, which is a non-event. The tail that reproduces 2024 arrives through the failure rather than around it — a disorderly break running through Japanese government bonds, producing yen strength by panic with no policy authority left to spend. If that lands, it lands late — after the August expiration on the 21st, with the Jackson Hole window the highest-leverage date on the calendar. It is not a first-week risk and should not be sized as one. The lowest-cost falsifier for all of this: the correlation index falling while the yen strengthens.

Unusual structures

A billion-dollar financing package, printed five times in forty minutes

The largest single structure on the file was not a bet. At 15:47:56 someone printed four legs at 7,000 lots each for $1,351.2M — a December-2027 7000 call sold, a December-2027 8000 put bought, and a matching September pair in the opposite direction. The same shape repeated at 15:18:56 twice, at 16:01:57 twice, and again at 16:05:53 across eight legs. Read individually each leg looks like a violent directional statement; read as packages they are jelly rolls, rolling a financing position across the term structure. $6.57B of SPX premium printed in packages above $50M each, which is 67.1% of all SPX premium on the day. Any aggregate that counted those legs separately would have produced a headline bearish index number roughly five times the structure-clean figure.

A three-tranche package that bought October and sold August, in the same second

Microsoft’s roll is described above, but it is worth marking as a structure in its own right because of how it was executed. Four times — 10:46:32, 13:34:32, 14:50:22, 15:49:44 — the August-21 460 call was sold and the October-16 510 call bought at identical size in the same second, for 21,435 contracts in total. The open interest on the sold line was 59,187 and on the bought line 253. A net-premium panel nets those two legs and prints a negative number; the structure is an institution paying to move its exposure up and out. The lesson generalises: same-second, same-size, opposite-side prints across different strikes are one trade, and splitting them is how a roll gets published as distribution.

Eighty-three thousand contracts of September upside, written into a 19% day

Structure-clean, with no matched leg anywhere against them, three prints inside thirty minutes wrote CoreWeave’s September 110 and 115 calls in sizes of 26,980, 25,285 and 30,701 against open interest of roughly 8,000 on each line, all flagged opening, all at 105 to 107 implied vol, all between 36% and 44% above spot. The volume-to-open-interest relationship is what makes this unambiguous — these were not closing trades and not overwrites of an existing book. Someone sold the top of a 19.49% move at an extreme premium, and the far strike priced at 0.96 times the at-the-money contract while they did it.

Half a billion dollars of downside sold on the September and November lines

The put side of the index book ran the other way from the equity book. SPY’s two largest prints of the session were November-20 620 puts at 150,000 contracts for $53.55M and November-20 500 puts at 300,000 contracts for $34.20M — 450,000 contracts of far downside, struck 17.8% and 33.7% below spot, at 26.81 and 38.93 implied vol against an at-the-money contract near 11.84. Those wings priced at roughly 2.3 and 3.3 times the body, which is precisely the premium the supply complex exists to harvest, and it was harvested.

The forward call supply nobody is discussing

The expiration map carries a detail that cuts against every bullish reading of Monday, and it survives the structure adjustment. The August-07 line is the largest premium row on the board at roughly +$41.0M of calls against +$8.5M of puts — but its dealer delta is negative. Heavy call premium against negative dealer delta is what an overwrite looks like, not a chase. And September-18 carries roughly −$33.5M of negative call premium — calls sold at size into the quarterly — against +$7.5M of puts bought, with September-04 showing the same pattern at −$16.0M. Someone is monetising upside and buying protection across September while buying August calls. The bullish flow has an expiration date on it, and the date is the third Friday.

The prints MAV flagged, checked against our own tape

A new standing section. MAV calls out unusual-activity prints in his evening commentary; from tonight forward each one gets looked up in our own options file and either confirmed, corrected or marked unverifiable. He named four on Monday. Two are futures options and do not appear in an equity tape at all. Of the two that do, one confirms exactly and one is real but materially incomplete — and the incomplete one is his headline.

QQQ — the $28M call package is real, and eight minutes later a bigger one printed the other way

His biggest call: a trader bought the August-31 700 and 705 calls for roughly $28M, which he read as a bet on QQQ reaching 735 by month-end. The print is real and our file has it to the second. At 10:31:08, with QQQ at 695.76, the August-31 700 call went up at the ask for 10,000 lots and the 705 call above the ask for the same 10,000 lots — same second, same size, $15.15M plus $12.65M = $27.80M, the 705 leg flagged opening against an open interest of just 2,094. It classifies in our decomposition as a single same-second, same-size structure rather than two independent bets — and because both legs are calls on the same side, the premium is genuinely the sum rather than a spread net, so his figure is the right way to count it. His number is accurate and his reading of it as bought is correct.

What is missing from his account is what happened eight minutes later. At 10:39:14 the identical two-strike package printed again — same two strikes, same expiry, same second, same size on both legs — at 12,000 lots apiece and below the bid on both. That is $18.72M and $15.66M, $34.38M sold, larger than the purchase it followed. Across the whole session those two lines take in $38.08M of buying against $36.13M of selling. The net is about $2M on a pair of lines that traded more than $78M of premium. His $28M bullish signal is one leg of a two-sided flow that nets to roughly nothing. This is also why QQQ's session-level directional figure is an ARTIFACT in our decomposition and is not citable as direction: the same-second packages dominate it.

TAPE · the two sides of the same package 10:31:08 — Aug-31 700C at ask 10,000 lots $15.15M (21.81 vol, OI 25,061) + Aug-31 705C above ask 10,000 lots $12.65M (21.35 vol, OI 2,094, opening) = $27.80M bought · 10:39:14 — Aug-31 700C below bid 12,000 lots $18.72M (21.56 vol) + Aug-31 705C below bid 12,000 lots $15.66M (21.12 vol) = $34.38M sold. Full-session totals on the two lines: bought $38.08M against sold $36.13M. Both prints classify as same-second same-size packages, so the premium is genuinely the sum of the legs rather than a spread net.

The stated target is also wrong, and the transcript's own polish notes flag it. Paying 15.15 for the 700 strike and 12.65 for the 705 gives a combined debit of 27.80 for one of each. The package breaks even where the two intrinsic values together cover that debit — at 716.40, not 735. His 735 figure only works if the whole $30 debit is charged against the 705 strike alone, which double-counts. QQQ closed 700.07, so what this package actually needs is +2.3%, not the +5.0% he described. That is a far more achievable trade than the one he presented, and it sits just under the 720 area where dealer gamma turns positive again.

PANW — confirmed to the second, but the “financing” is a rounding error

He described a trader buying the September-4 390 calls against selling the August-7 390 calls, with the near leg financing the far one, on a stock at $346. Our tape confirms it exactly. At 15:20:10, same second and same 920-lot size on both legs: the September-4 390 call lifted at the ask for $1.08M against an open interest of 10, flagged opening, and the August-7 390 leg at the same size. His structure, his strikes, his direction — all correct.

One correction to the framing. The August-7 leg fetched $37K. That is 3.4% of the $1.08M he paid for the September leg, because a 390 strike sits 13% above a $346 stock with four sessions left and is close to worthless. Calling it a financed calendar overstates what the near leg does — this is 96.6% an outright long call with a token strike sold against it. The direction is right and the financing is decorative. Our own PANW read agrees with him on direction: the structure-clean citable net is +$4.36M.

Crude — unverifiable here, and the largest energy structure on our tape leans the other way

He flagged roughly $800K of September-17 91.5 calls on the October crude contract, against a $77 contract price, tying it to an $80 support level and a cup-and-handle that targets the 90s. Futures options do not appear in an equity options file, so this cannot be confirmed or denied from our data. Said plainly rather than glossed.

What our tape does carry is the equity energy complex, and the largest structure in it runs the other way. At 15:32:53, XOP printed a 25,000-lot October-16 165/150 put spread — the 165 leg bought to the ask for $14.40M against the 150 leg sold for $4.53M, both legs opening, both sizes enormously exceeding an open interest of 279 and 951. XOP closed 173.72, so the shape is a bet on energy equities falling more than 5% by mid-October, and the legs together are an order of magnitude larger than the crude print he cited. One honesty caveat our own rules force: because it prints as a single same-second structure, the five-stage strip reduces XOP’s directional residue to essentially zero, so this is a structure we can describe but not a directional number we may cite. USO's own flow was mixed at $5.08M bought against $4.80M sold. The two names he called out cut his way but barely register. XOM traded $2.23M of premium all session with its August-21 145 puts sold, a mildly bullish footprint; CVX $7.81M, most of it a matched 186 straddle carrying no direction at all; XLE $3.90M and mixed. His bullish energy view may still be right on the commodity, and the single names lean gently his way, but the biggest identifiable money in energy equities on this session paid up for downside and it dwarfs everything else in the complex.

Treasuries — also futures, and our proxy is shape-consistent but not citable

He flagged roughly $8.5M of September-25 106 puts on the December bond contract, reading it as a bet on a yield surge to levels last seen in 2023 and before that 2011. Again a futures option, so not directly checkable here.

The nearest equity proxy is TLT, and its two largest prints are shape-consistent with him: at 09:36:33 the September-18 82 and 81 puts printed 40,000 lots each in the same second, $5.08M and $3.60M, a vertical against an 82.19 close. But both legs printed with no side tag, and TLT carries 59.5% of its gross premium unsided across the session, which is why our own gate marks the name low-confidence and refuses it as a citable directional number. The shape agrees with him. The evidence does not rise to confirmation, and it is listed here as unresolved rather than as support.

Scoring him, and what this section is for

Credit first: he does not claim these are signals. His own framing on the QQQ print is that it is a trader with an opinion, no more, and he decodes each one by naming the condition under which it works rather than taking its direction at face value. On QQQ that condition is a rotation back into semiconductors, which is exactly the cohort our own squeeze screen flags. That is a more careful method than most unusual-activity commentary.

The tally: two of four checkable here, one exact, one materially incomplete, one contradicted by proxy, one unresolved. The pattern worth carrying is not that he is wrong — his PANW read was precise and his QQQ print was real. It is that an unusual-activity screen shows you a print, not a position. The QQQ case is the clean demonstration: the same screen that surfaced a $28M purchase at 10:31 would have surfaced a $34M sale of the identical package at 10:39, and only the pair tells you the flow netted to about $2M. Reading one and not the other converts a wash into a conviction call. That is the same failure mode our own decomposition exists to prevent, and it is why every print quoted here gets looked up rather than repeated.

Grading last week’s report against Monday’s tape

The 07/31 report covered this session, so this is its verdict, and it is a poor one. Six clear hits, twelve misses, four partials and three items that could not be graded — 27% strict, 36% counting partials. The mid-day supplement published during Monday scored 20% on its five directional calls, though its bookkeeping across all nineteen gradeable items ran 63%.

The biggest miss was a standing instruction to avoid the AI-financing complex, reaffirmed on Thursday and held into Monday. CRWV closed +19.49%, NBIS +11.64%, IREN +8.02%, APLD +7.67%, and all four returned accumulation verdicts with an explicit flag that the fifteen-day distribution tag disagreed with price. The best call was Microsoft, read as a roll rather than distribution while the naive number showed $89.2M of written calls; it closed +4.93%. Second best was the AMZN long, whose invalidation at 267.21 was never approached against a 284.02 close.

The structural lesson is one error made twelve times, and it is worth stating plainly because it recurs. Label-derived flow was allowed to outrank the price-and-volume verdict, on a fast tape, on the exact names where the contradiction had already been flagged in bold. Every instrument at the centre of last week’s bearish case — SPY, QQQ, IWM and the four financing names — returned an accumulation verdict with tape speed fast and label reliability low. At-bid is not selling in a fast rising tape; that adjustment is applied automatically upstream and none of it reached the document. The correction layer worked and was ignored, which is a worse failure than not having it.

Two secondary patterns. Invalidations were placed where they could not fire — the index hedge required both a close above the zone high and a Nasdaq trend reclaim, and price cleared the first on day one while QQQ closed 700.07 against a trend near 704.90, leaving a position wrong by every measure that mattered and technically un-invalidated. And multi-month structural evidence was used to set a next-session probability. Every hit in that report was a single-name structure read; every miss was a direction or regime read. The decomposition machinery is working. The layer above it is where the losses were.

The August map

The calendar that matters is short and the resolution point is not Tuesday. Friday 08/07 is July payrolls, and it lands on the most negatively positioned expiration of the month — dealers carry roughly −$2.44B of put delta against −$0.23B of call delta into it, with cumulative premium negative and falling. A soft print into that positioning is the squeeze case; a hot one is the compression case. Wednesday 08/05 brings the services survey and SanDisk’s result, the most contested single call in this week’s commentary. AMD reports Tuesday after the close with roughly an 8.58% move priced — a single-name event, not an index one.

Beyond that: the bought-volatility cluster expires 08/10 to 08/12, inflation data lands 08/12, and August expiration on Friday 08/21 carries the largest gross book of the cycle at +$2.95B of call delta against −$4.58B of put delta, with size concentrated in the MSFT 460 line, QQQ 680 puts, SPY 735 puts and the SMH 530 and 475 puts. The projected timing overlay puts a local high on Wednesday 08/05, a major trough around Friday 08/14 to Monday 08/17, and a peak around Tuesday 08/25 — and after one session its opening leg has been confirmed while its mirror has been invalidated. Treat those as inflection dates only; that source carries no magnitude information and none is taken from it. Wednesday 08/05 is the cheapest test on the calendar: a high there makes the mid-month trough the buy window, a low there hands the month to the inverse shape. Two dates on this map are genuinely unconfirmed — NVDA’s report, placed at 08/26 by one source and 08/19 by another, and the Jackson Hole window — and nothing should be sized against either until they are confirmed.

Where this differs from Monday’s mid-day read

Three places. On CoreWeave the mid-day call — institutions selling the rip — was arithmetically correct and the full session confirms it, but it never reached a written document and so could not be acted on. On the index the mid-day read treated the negative dark-pool prints as directional; the full session shows those tapes closed fast with low label reliability and an accumulation verdict, so they are not citable as direction. On small caps the mid-day read stands without amendment: IWM’s structure-clean residue shows $48.8M of puts bought against $5.5M of calls bought, which is genuine absence rather than a labelling artifact.

The trap this framework wrote down before it sprang

There is a pattern worth naming because it recurs and because it is a positioning failure rather than a forecasting one. Heavy protective positioning accumulates ahead of a seasonally feared month; a catalyst arrives; the hedges have to come off at once and the move is violent. The weekend note before this session observed that a commentary batch agreeing this completely is itself a crowding signal, and that if it fired with breadth the entire batch — and anyone positioned with it — would be offside together. It then fired with breadth.

What the record actually teaches about this shape is narrower and more useful than the folklore. A wall of open interest is fuel, not ignition — a squeeze thesis requires a dated catalyst, and one on 07/24 sat inert for three sessions because nothing forced covering. Quality leadership marks a bottom; most-shorted leadership marks a squeeze — and Monday’s leaders were the beaten-down financing names, not the quality semis. Flat skew marks squeezes and steep skew marks tops, and the skew is flat. All three read the same way tonight, which is why the squeeze list above exists. The failure mode on record is not missing the setup. It is trading an “avoid” as a short and holding it through the catalyst.

BOTTOM LINE

The bid was real, the direction is up, and the room is nearly gone. Nine independent bullish inputs against eight fragility inputs is a net of one — and four of those eight are structural ceilings rather than opinions: the exhausted zone, the weekly band spent on day one, negative dealer gamma sitting at the close, and September call supply. Structural ceilings do not care about conviction. They cap distance, not direction. The modal Tuesday is a narrow, jumpy session between 7565 and 7650 that resolves nothing, and the real resolution is Friday's payroll print into the most negatively positioned expiration of the month. Own the rotation into single names, not the index. Do not chase a band that has already been spent.

Top Trades to Follow

MSFT · SUPPORTIVE Close 487.65. The largest identifiable holder rolled 21,435 contracts from the August 460 line to the October 510 line, paying premium to do it, while the dark pool printed +$4.54B on volume up 33.2% and the three-day slope turned up $5.49B. Near-term risk is digestion after a two-sigma break, not reversal. Fails below 460.

CRWV · UPSIDE CAPPED Close 85.76 after +19.49%. Structure-clean residue −$26.56M on an 87.2% directional read; roughly 83,000 contracts of September 110 and 115 calls written at 105–107 vol, opening, in size above open interest. The darkpool net is flat once the 16:00 auction cross is stripped, so the ceiling rests on the options structure alone. A genuine ceiling at 110–115 through September 18. Fails on a sustained close above 115, which would mean the writers are being run over.

KWEB · SUPPORTIVE Close 28.74. The cleanest signal on the board: the only name whose tape was normal rather than fast, a 100% structure-clean residue at +$9.80M, the tallest bar on the call-volume panel at 4.9 times baseline, 81.8% buy-side with a four-and-a-half-month average expiry, and a three-day slope turning up. Small flow, high evidentiary quality. Fails below 27.

AMZN · SUPPORTIVE, WITH A NAMED OFFSET Close 284.02. Structure-clean +$103.57M, the most robust bullish figure on the file — it survived every stage of the decomposition and got stronger. Offset explicitly: a $4.07B pre-scheduled insider filing landed the same day and the dark pool printed −$1.75B. Absorbing known supply on a strong close is constructive. Fails below 267.21.

THE SEMI COMPLEX · COILED, NOT CONFIRMED MU 829.50, AVGO 392.23, SKHY 142.72, SOXX, AMD 484.64. Flat skew, dealers short gamma above spot, and no move yet — the configuration that squeezes. But roughly $21B of leveraged assets sit on top of it and SOXL took a record $6.9B in July, so this is crowded in both directions and conviction is capped accordingly. Needs the 08/07 catalyst. Dies if semis do not lead within three sessions.

INDEX FLOW · NOT CITABLE AS DIRECTION SPX structure-clean residue 11.6% of gross, below the threshold for any directional claim; SPY +$1.2M and QQQ −$2.0M once financing is removed. The September 7600 line is 0.91 calls to puts — a box, not a wall. Anyone quoting an index flow number from this session is quoting a jelly roll.

THE NEAREST INVALIDATION ON THE BOARD The 30-year closed 5.231%. A close below 5.20% — 3.1bp away — kills the read that the long end has stopped leading this market, and would put duration back in charge of the equity tape. This is the single lowest-cost thing to watch on Tuesday.

SOURCES

Every file opened this cycle, not a highlights sample. No URLs were fetched and no web search was run. Every external fact used above — the hedge-fund technology purchase, the Bezos filing, the PLTR results, the leveraged-ETF concentration figures, the July semiconductor statistics, the Iran strike pause, the reported yen intervention, the server-market forecast, the positioning views, and the single-country vehicle and delivery numbers — arrived through the operator or the commentary layer and was not independently verified. None of it overrides a flow or price figure anywhere in this report. The yen intervention in particular has no primary document behind it: three commentary voices assert it and the reported scale is a wire relay.

EXPECTED MOVES · 15 artifacts, all opened Forward set stamped 08/04 carrying the 08/03 closes: base, zones, range & trend, ZONE DOCUMENT, Zone Visual · retrospective set stamped 08/03 carrying the 07/31 closes: the same five, from the archive · weekly 08/03–08/07 · monthly August 2026 · quarterly July–September 2026 · the Q3 collar sheet (read; strikes cited only — the capture is a 1 July screenshot so every band and gamma overlay on it is stale) · and the 08/03 sentiment gauge, three pages.

DASHBOARDS · every panel read as an image Options dashboard 08/03 — 21 pages, all 21 read individually · darkpool dashboard 08/03 — 9 pages, all 9 read individually. Rendered at 150dpi with PyMuPDF (pdftoppm is not installed on this machine), 33 images in total including the sentiment pages.

RAW DATA Live options flow 08/03 (43,775 rows, deduped to 42,383) · darkpool market summary 08/03 (3,425 names) · darkpool summaries 07/31 and 07/30 for the day-over-day volume overlay · two Savino August projection charts, normal and inverse, opened for the first time and read for timing, direction and structure only.

PER-TICKER CENSUS · 2026-08-03, 37 files opened and quoted AAPL · AMD · AMZN · AVGO · BE · CAT · CRWD · CRWV · DELL · GLW · GOOG · GOOGL · IGV · INTC · IWM · KWEB · LITE · META · MRVL · MSFT · MSTR · MU · NBIS · NFLX · NVDA · NVO · ORCL · PLTR · QQQ · SKHY · SMH · SNDK · SPCX · SPY · TLT · TSLA · WULF. SOXX has no census file for this date — declared as a gap, not substituted from another session.

COMMENTARY · 16 files, all read in full Mike Silva 08/03 transcript and the matching 32-page slide deck, read as one input · MAV 08/02 Sunday futures · Cem Karsan 08/03 · Arete Trading 08/03 · FX Evolution 08/03 · Tom Gentile 08/03 · Conquer Trading 08/03 · Wall Street Millennial 08/03 · James 08/03 · Mike Jones 08/03 · Krown 08/03 · Geeks of Finance 08/03 · two tastylive pieces 08/03 · MAV 07/30 for continuity · and the prior Arete analysis covering 06/26–07/30.

PRIOR WORK GRADED OR CARRIED The published 07/31 daily report · the 08/03 mid-day flow supplement and its first-day-of-month roster · the 07/31 comprehensive analysis for the existing August calendar · the 08/01–08/02 weekend commentary batch · the bottoming-framework synthesis · the regime snapshot · the rolling tracker at version 61 · the correction log · the thesis ledger · and the session logs for 08/04, 08/03 and 08/02.

DECLARED GAPS · not substituted, not inferred No census file for SOXX · no darkpool sentiment histogram in the 08/03 capture, so no cumulative-slope claim was made from the panels · the forward daily band sheet carries no single-name rows, so there are no forward daily bands for any individual stock this cycle · no daily band row for DIA or VIX on either stamp · the weekly sheet is defective in places, with four sector rows printing their upper below their lower and two index closes wrong, so those rows were flagged and not graded · no census exists yet for 08/04, so tonight's post-close reporters cannot be graded on price · and two forward dates — NVDA's report and the Jackson Hole window — are unconfirmed across sources and are flagged rather than asserted.