📊 Ainvest Option Flow Digest — 2026-08-06

🔁 Updated 2026-08-07 pre-market — this session's own open-interest check is now in, and five of the eighteen names changed meaning. All 35 legs are settled: 22 opens, 5 closes, 8 flat/transfer. 🔄 Inversions: TTWO, IGV, AMD, ECHO, TLT. TTWO was not a new bear call spread being sold into earnings — it was an existing bull call spread being closed the day before the print. IGV was not a roll — both legs opened, making it a new short calendar. AMD, ECHO and TLT created no new open interest at all. Every affected article has been rewritten with a ✅ RESOLVED box; the bullets, table and closing section below now carry the corrected reads. Full detail: _OI_RESOLUTION_NOTES.md for this session.

Eighteen names, ≈$508M net — and one trade is 61% of it

---

🎯 Quick Read

Every package today was negotiated — floor trades and facilitated auctions. Not one lit sweep. And the board is dominated by a single print.

• ≈$508M net across 18 names, of which SanDisk alone is $310.5M. Strip that out and it is an ordinary session; leave it in and it is the largest single day we have covered.
• ⭐⭐ SNDK sold $286.5M in ONE put — and unlike two days ago, this one has no hedge. 5,000 January-2027 $1,660 puts at $573, against prior open interest of just 660. On August 4 the near-identical trade was matched by 105,000 shares against a 104,550-share delta equivalent — a 0.4% fit, making it delta-neutral. Today we looked and could not find it: 134,152 shares in the whole window, no block above 20,000, against ≈252,500 expected. Inference: this one is directional. It went on the day after SanDisk beat but guided lower and fell 4.5%, at an effective purchase price of $1,087 — 15.7% below spot.
• ⭐ GLD sold gold upside for the second session running, 2.3× bigger — and yesterday's is now confirmed live. The September $400 call's open interest rose 38,167 → 67,993 overnight, proving Tuesday's spread was genuine new short risk. Today: sell 55,292 September $410 calls, buy the $430, $20.4M credit, both legs proven opens. A desk adding, not covering. The $410 strike sits exactly on a "Very Strong" gamma resistance shelf.
• 🔄 TTWO — CORRECTED: this was a bull call spread being CLOSED, not a bear spread being opened. We published $8.9M collected on the September $270/$310 as premium sold into a confirmed print. Next-day open interest collapsed on both strikes — $270 call 27,182 → 6,025 (−21,157), $310 call 28,250 → 6,603 (−21,647) — so the sale was a sell to close and the purchase a buy to close. A desk that was bullish took its September position off the table the day before earnings. The $8.9M is exit proceeds, and the $70.7M tail we quoted does not exist.
• 🔄 AMD — CORRECTED: the $54.4M of December-2028 calls created no new exposure. Open interest fell on both strikes — $760 call 2,351 → 1,498 (−853) and $860 call 2,744 → 2,526 (−218) — on day volume of 2,001 and 2,018, essentially just this package. A buy that reduces open interest is a buy to close. Not one new long call was created; the net position at both strikes shrank. The structural read holds — two outright calls, not a spread, proven by $14 of price difference across a $100 strike gap — but the "long-dated bullish" framing does not.
• MSFT put on the cleanest directional structure of the day. A $19.0M September $525/$570 bull call spread, both legs proven new against prior open interest under 2,000. Max profit $88.4M, breakeven +7.1% — but there is no earnings print inside the expiry, and the $525 strike sits above every resistance shelf the gamma model finds.
• S (SentinelOne) built a bearish synthetic on a strike that did not exist this morning. Long 4,500 December-2028 $30 puts against short $35 calls, prior open interest zero on the put. Delta −411,390 shares, opened within 4.5% of the 52-week high — and notably, the analyst average target of $19.68 sits below the $20.48 stock.
• 🔄 TLT — CORRECTED: the $20.4M put block created ZERO new open interest. Four floor-traded legs of deep in-the-money January puts, and the next-day count came back essentially unchanged on all of them: $110 put 7,776 → 7,776 (0), $107 put 500 → 500 (0), $120 put 140 → 140 (0), $105 put 2,626 → 2,646 (+20) — twenty contracts of net change against 14,090 printed. We re-checked the tape and found no cancellations, so this was a pure transfer: an existing short-duration position changing hands, not a new bet being placed. The position still carries −554,170 shares of delta and three FOMC meetings still sit inside the expiry — but nobody expressed that view on August 6.
• UBER bought $2.3M of December $80 calls, 13% out of the money. Delta 0.34 means it controls ≈$15.7M of stock-equivalent exposure for the $2.3M paid — but breakeven needs +18% on a stock −14.4% over the year. Note the open is proven by a narrower margin than most of today's board (6,500 against 4,699), so tomorrow's number matters more here.
• PANW is the smallest ticket on the board at $1.29M — and the most interesting piece of context. Palo Alto is IGV's largest holding at 9.61%, and IGV itself put on a $10M short calendar today. Add Microsoft, Salesforce and Adobe and ≈28% of that fund traded on this page — pointing in different directions. PANW's own call confirmed opening, and by more than the flagged block: open interest went 202 → 2,541 (+2,339) on a 1,258-lot print.
• 🔄 ⭐ ECHO — CORRECTED: the $12.57M put package opened nothing, and the directional read was backwards. Five floor-traded legs, and the next-day count settled all five against us. The two big legs were transfers — $140 put 2,700 → 2,700 (0), $135 put 2,510 → 2,503 (−7) — with no cancellations on the tape. The three smaller strikes were closes: $125 put 396 → 34, $130 put 160 → 0 (strike extinguished), September $145 put 187 → 8. A sale that reduces open interest is a sell to close, so what we called "unmatched short puts carrying real downside risk" were long puts being sold to close — the opposite exposure. The financing arithmetic still holds ($4.90 collected on a $5.00 width, delta only +215,953 shares); what does not hold is that any of it was new.
• 🔄 ⭐ IGV — CORRECTED: not a roll. A brand-new $10M short calendar, both legs opened. We published the falsifiable test — a roll required August open interest to collapse toward ≈0–1,000 — and it did the opposite: 19,308 → 38,739 (+19,431, 97.2% of the print). The old August position was never touched. The November leg opened too (1,750 → 21,764). So this is long the August $100 call, short the November $100 call for a $10M credit: a bet that back-month software volatility is expensive, with the front-month protection expiring August 21 and an uncapped short call left behind it. Worth knowing: MSFT (9.47%), CRM (5.38%) and ADBE (3.68%) are ≈18.5% of the fund — and all three confirmed as genuine opens in the same snapshot.
• ⭐ CELH bought a 15-to-1 bull call spread hours after a double earnings miss. Celsius reported today — revenue $817.9M vs $870.07M, adjusted EPS $0.36 vs $0.42 — and the stock sits at $23.74 against a 52-week low of $23.66. Into that, somebody paid $5.5M for the September-2027 $47.50/$65 spread, both legs proven new against prior open interest of 34 and 243. Breakeven requires the stock to double.
• IBIT bought $5.7M of bitcoin crash puts, 59% below spot, expiring 2028. That is ≈$115.8M of notional hedged for 5.0% of its value — expensive as insurance goes, and it pays only on a 64% decline. Note the fund is already −43% over the past year: this is protection bought after a fall, against a further one.
• B — that single letter is Barrick Mining, and it makes today's gold picture two-sided. A modest $1.57M of March-2027 $60 calls, 45% out of the money, four days before confirmed August 10 earnings — while GLD saw somebody sell gold upside for the second session running.
• ADBE sold at-the-money November calls for $8.1M — a $260 strike just 1.2% above the stock, at 10.6% of the share price for three and a half months, on a name −22.8% over the year. Rich premium on a falling stock is compensation for risk, not a bargain.
• ASX bought BOTH a call and a put — and it is not a spread. 10,000 March-2027 $55 calls and 5,000 $30 puts, both bought, prior open interest of 74 and 5. That makes it a strangle, but at a 2-to-1 ratio it tilts bullish (delta +269,300 shares) on a chip packager up 122.7% over the year.
• CRM and DELL are the quieter two. Salesforce sold 5,000 November $195 calls for $7.96M — on a stock already −27.4% over the year — with the strike parked on its strongest gamma resistance. Dell bought $7.68M of June-2027 $480 calls into a ≈15% pullback on a name up 248%.

The honest frame: one number on this page is worth more than the other seven combined, and its meaning turns entirely on a stock trade we could not find. That is worth more of your attention than the headline dollar figure.

---

🔁 OI Review — Last Session's Provisional Flags, Now Resolved

Yesterday's board has been through the next-day open-interest test, and five of thirteen names came back meaning something different than they looked. All 26 legs from the August 5 session are settled: 15 opens, 5 closes, 1 partial close, 2 partial opens, 3 pure transfers. Two of the biggest dollar figures on the page turned out to establish nothing at all.

🔄 Inversions

🔄 TSM — the $19.75M September $400 call buy was a short being COVERED. Open interest fell 11,647 → 7,711 (−3,936) against a 5,000-lot print. We published both branches — rise toward ≈16,647 means opening, fall toward ≈6,647 means closing — and it fell. That strike has now declined four consecutive sessions (21,661 → 16,725 → 11,647 → 7,711), a steady ≈4,000-to-5,000-per-session unwind. Corrected: only the $20.5M risk reversal is a new bullish position. The "$40M bullish combo" headline overstated the day's actual commitment by roughly half.

🔄 XLP — calendar or double-long? Neither. It was a ROLL — and our published test called it to four contracts. We wrote that September open interest falling to ≈5,526 would prove the near leg closed. It printed 5,522. September was wiped out (85,629 → 5,522, −80,107) and December opened (25,080 → 110,112, +85,032). Corrected: a put roll-out at a true net cost of ≈$6.57M, not a $10.1M new bearish bet. Same $78 strike, same size, three more months. That is a maintained conviction, not a new alarm.

🔄 WMT — the "$105 floor" was not a floor, and the position is riskier than we published. That strike lost 15,930 contracts against a 23,702-lot buy, so it was predominantly a buy-to-CLOSE. The $119 sale confirmed opening cleanly (2 → 20,950). Read together this is a short-put roll UP from $105 to $119 — a bullish escalation with no long put underneath it. Corrected: our published payoff table (worst point −$15.37M at exactly $105, recovery near ≈$64.83) described a ratio spread that does not exist. The real shape is an unhedged short put: breakeven ≈$111.72, and the loss simply keeps growing below it — roughly −$23M at $100, −$43M at $90. It carries the estimated August 20 earnings print.

🔄 SPY — the $32.2M collar created ZERO new open interest. Both legs printed flat against a 20,500-lot size ($707 put −14, $783 call +105). That is the signature of a pure transfer — one side opened, one side closed, in equal measure. The open-interest history pins the collar's birth to the July 31 close; it sat untouched August 3–4 and simply changed hands August 5. Corrected: there is no new hedging demand to read here. The separate $53.0M August $600 call package did confirm opening (26 → 3,009).

🔄 GDX — the two $67 put sales never opened a short, so the payoff is not capped. November fell −17,278 and December −1,193; we had written that a decline would mean "an unwind/roll on the $67 side layered under two fresh $75 longs, not a clean new spread." Both $75 put buys confirmed opening decisively (+42,339 and +44,410). Corrected: a roll of protection UP from $67 to $75. You cannot cap a payoff with a short you never established — the "max value $8.00, defined-reward" framing was withdrawn, and the surviving ≈85,000 long $75 puts pay all the way down.

✅ Confirmations

• ⭐ NVDA — the $65.3M question is answered: it was an ADD, not an exit. Open interest rose 91,722 → 105,697 (+13,975) on a 14,200 print — ≈98% new contracts. The buy-to-cover branch is dead. Context: that strike went 2,935 → 90,997 in one session on July 23, held flat for nine sessions, then grew again. A deliberate, staged accumulation.
• ⭐ WYNN — the roll is proven on both sides. Predicted September open interest ≈100–150; it printed 235 (16,627 → 235). December confirmed 630 → 18,043. The provisional STC marker is retired.
• INTC — a trim, within 14 contracts. OI fell 83,141 → 78,127 (−5,014) on a 5,000 print, total strike volume 5,056. The open-interest history (built in one burst July 28, then untouched) was the right evidence to lean on.
• GLD — a genuine brand-new bear call spread. Both legs rose above the print size (+29,826 and +26,961), ruling out the "bullish trader cashing out a winner" alternative we flagged.
• HYG — the full 75,000 lots became new open interest (1,842 → 77,359), day volume matching the gain exactly. No transfer leakage: the $596M-notional tail hedge is as big as the headline said.
• SPCX — both legs opened (call +13,044, put +10,600), resolving the provisional put in favour of a new long and confirming a clean synthetic short. HUBS — confirmed 171 → 3,248, above the predicted range.
• ⚠️ CC opened, but only ≈81% of it. The $14 strike landed at 13,540 against a predicted ≈16,700 and the $17 strike at 13,687 against ≈16,900 — about one contract in five was matched against a closing counterparty. Structure and direction unchanged; the market's net new position is ≈13,500 spreads, not 16,666.

What changed

The dollar totals are untouched — open interest never moves the cash. What moved is what those dollars mean. TSM's $19.75M was an obligation being retired. SPY's $32.2M established nothing whatsoever. WMT's, GDX's and XLP's packages were protection being moved rather than protection being bought. Against that, NVDA's $65.3M is confirmed as genuinely new exposure, and HYG's tail hedge is confirmed at full size.

The standing lesson, with a new wrinkle: a flat open-interest print is a real answer, not a failed test. We offered readers only two branches on the SPY collar — open or close — and reality took a third. A 20,500-lot cross can be economically enormous and leave the market's aggregate position exactly where it started. From now on the ⏳ callouts will name flat/transfer as a third possible outcome alongside open and close.

<!-- PRE-DRAFT NOTES (internal — delete before publishing) - Written 2026-08-06 pre-market by the morning OI check, per the pre-draft-next-day-oi-review rule. - Source of truth: agent/UOA/20260805/_OI_RESOLUTION_NOTES.md (full 27-leg table + per-inversion detail). - Place this section between the Quick Read and the At-a-Glance table, separated by ---, per the newsletter-OI-review-section rule. - Glance-table status cells for the 2026-08-06 session must reflect resolved ✅/🔄 verdicts where applicable. - Carry-over CLOSED OUT: MSFT Aug-21 $460C (from the Aug-3 session) is now permanently INCONCLUSIVE — strike churn (+22,191 then −51,440) dwarfs the 7,435 print. No further re-checks scheduled; do not re-list it. - Known web-app discrepancy to avoid contradicting: CC's two legs render CLOSED on the site because the deployed position_threshold() measures against day volume at 90%, and CC absorbed ≈80%. They are genuine opens (OI 0 → 13,540). Do not describe CC as closed in the newsletter. -->

---

📋 At a Glance

Net, not gross. Every figure is net — premium paid minus premium collected. GLD's two legs gross $51.9M and net to a $20.4M credit; MSFT's gross $31.4M and net to $19.0M.

---

🔍 Three Worth Your Time

⭐⭐ SNDK — the same trade as Tuesday, with the hedge missing

Two days ago SanDisk saw 2,750 January $1,650 puts sold, and the equity tape carried 105,000 shares against a 104,550-share delta equivalent — a 0.4% match. That made it a volatility sale: the seller stripped the direction out and kept only time value.

Today's is bigger and looks different. 5,000 January $1,660 puts at $573 — $286,500,000 — against prior open interest of 660, so almost all of it is new. We went looking for the same stock hedge and found 134,152 shares in the entire 11:17–11:19 window, with no block above 20,000, against roughly 252,500 you would expect. The January leg also prints as a plain floor trade, without the stock-plus-options marking the September legs carry.

So the inference is that this one is directional — and it is an inference, not proof. A hedge worked at another venue or over several hours would not appear.

The timing is what makes it interesting. SanDisk reported yesterday: a beat, but guidance that "failed to impress", and the stock is down 4.5%. Somebody stepped into that with an effective purchase price of $1,087 — 15.7% below the current price — and was paid $286.5M to wait.

Worth knowing before you form a view: analyst targets are being cut today and the disagreement is extreme — an average of $2,148.68, but with same-day revisions from RBC at $1,300 to Evercore ISI at $2,800. A 2.15× spread means the street has no shared view of the NAND cycle at all.

⭐ GLD — the second bear call spread in two days, and the first one is confirmed

On Tuesday a desk sold the September $400 call and bought the $425, 23,839 contracts. We flagged both legs unprovable, because each traded below existing open interest.

This morning settled it. The $400 call's open interest went 38,167 → 67,993, a rise of 29,826. Contracts are only created when a position is genuinely new, so Tuesday's spread was real new short risk — not an unwind.

Today the same shape returned, 2.3× larger and $10 higher: sell 55,292 September $410 calls, buy the $430, for a $20.4M credit. Both legs are proven opens.

Two things make it worth reading closely. First, the $410 strike sits precisely on a "Very Strong" gamma resistance shelf — the level where dealer hedging pushes back hardest against upside. Second, the chain's own expected range through September tops out at $418.98, above the short strike but well below the $430 wing. The spread threads exactly between likely and extreme.

And the tail is real: $90.2M if gold reclaims $430. Collecting $20.4M against that is roughly one dollar of income for every four and a half at risk.

🔄 TTWO — the correction that matters most on this board

Four minutes into the session, a desk sold 19,905 September $270 calls and bought the $310s, keeping $8.9M. Take-Two reported August 7 — confirmed.

We published this as premium being sold into a dated event, and flagged both legs ⏳ provisional because each traded below existing open interest. We wrote that it might be a desk fading the print, or a desk clearing risk before it — "opposite signals from an identical tape" — and that tomorrow's check would decide it.

It decided it, and it was the second one. Open interest at the $270 strike fell 27,182 → 6,025, and at the $310 strike 28,250 → 6,603: drops of 21,157 and 21,647 contracts, both larger than the 19,905-lot print. Open interest can only fall when both sides are closing. So the sale was a sell to close a long $270 call, and the purchase a buy to close a short $310 call — long the lower strike, short the higher one, same expiry. That is a bull call spread, and on August 6 it was taken off.

The story inverts from bearish to bullish-de-risking. Nobody faded Take-Two into its print. A desk that had been positioned for a move toward $270–$310 by September decided not to own the gap risk, and banked $8.9M of exit value one session before the news. The elevated pre-event implied volatility that we described as making the sale attractive was real — it just made the exit better, not the entry.

And the scariest number we printed evaporates. There is no $70.7M tail, because there is no open short spread. We are retracting that figure explicitly.

---

📅 Upcoming Catalysts — and Which Expiry Each One Lands In

⚠️ A catalyst only matters to a contract still alive when it happens. One name here has no event inside its expiry at all.

On rates, because it cuts against the reflex: the Fed held at 3.50–3.75% on July 29 on a 9–3 vote, and the statement names Hammack, Kashkari and Logan as preferring a quarter-point HIKE (Federal Reserve). That is the confirmed record. A gold thesis built on imminent cuts is arguing against the Fed's own minutes.

Four earnings dates we deliberately did not guess: SanDisk and AMD have both already reported and neither has published a forward date, and we could not source confirmed forward dates for Adobe or ASE Technology either. We would rather leave a blank than print an estimate you might trade an expiry against.

---

👥 Four Ways to Read This Board

🎲 The YOLO Trader

TTWO is the only dated binary here — and it is the classic trap. The chain says ±9.96% tomorrow, which means the move is already in the price. Buy options into that and you can call the direction correctly and still lose, because implied volatility collapses the moment the news is out.

AMD looked like your trade and was not even a trade in the sense the headline implied. A $54.4M call buy printed in the first ten seconds of the session and created zero new open interest. Size on a scanner is not conviction — check whether the contract count actually grew.

📈 The Swing Trader

MSFT is the cleanest structure on the page — both legs proven new, defined risk, 4.6-to-1 payoff. The honest caveat is large: no earnings inside the expiry, and the $525 strike sits above every gamma resistance shelf. You would be betting on drift through a wall.

GLD's $410 level is now worth watching whichever way you lean — it is both a very strong gamma shelf and where a desk has sold twice.

💰 The Premium Collector

Three of today's eight are your trade, and the contrasts are the education:

• GLD bought the $430 wing. That wing cost $15.8M of the $36.2M collected — expensive, and the only reason the position is survivable.
• TTWO turned out not to be a premium sale at all — the resolution shows a bullish spread being closed. The lesson that survives: elevated pre-event implied volatility helps whoever is selling an option, and that includes someone selling a long call to get out.
• CRM sold a call with no wing at all. If those 5,000 contracts are not covered by stock, the loss above $195 has no ceiling. We cannot tell from the tape which it is — and neither can you.

🌱 The Beginner

Two ideas from today, both worth more than any single trade:

First: "sold a put" does not mean bearish. SNDK's $286.5M put sale carries positive delta — it is a commitment to buy the stock at an effective $1,087. Read the delta, not the word.

Second: "BUY" on a flow scanner does not mean "opened." Buy-to-open and buy-to-close print identically — same direction, same size, same premium. Only the next morning's open-interest count separates them, and today it separated five of these eighteen names from the story we first told. AMD is the cleanest example: two legs both bought, $14 of price difference across a $100 strike gap (so genuinely not a spread) — and open interest fell anyway.

---

✅ Every Read Here Has Now Been Through the Open-Interest Check

Updated 2026-08-07 pre-market. The ≈06:30 ET OPRA snapshot reflecting the August 6 close has published. All 35 legs across these eighteen names are settled — 22 opens, 5 closes, 8 flat/transfer — and five names inverted.

• ✅ Confirmed opens (22 legs, 13 names): all three SNDK legs (the January $1,660 put landed on 5,660, exactly the predicted number), both GLD legs (≈103% of the print each), both MSFT legs (≈124%), both S legs (the $30 put went 0 → 4,500, a 100.0% match from an empty strike), both ASX legs, both CELH legs, both IGV legs, CRM, DELL, ADBE, B, PANW, IBIT (opened at ≈double the flagged block, 1,312 → 66,083) and UBER (92.1% — the narrowest margin on the board, exactly as flagged).
• 🔄 Closes (5 legs): both TTWO legs (−21,157 and −21,647 — a bull spread taken off) and three ECHO strikes (−362, −179, −160; the $130 line went to zero).
• ⚠️ Flat / transfer (8 legs): both AMD LEAP calls (open interest fell), both large ECHO legs, and all four TLT legs. In each case we re-checked the OPRA tape for cancellations and found none — these were genuine prints that created no new position. $87M of headline premium across three names that established nothing.
• 🔄 The five inversions, in order of consequence: TTWO (bear-spread-opened → bull-spread-closed), IGV (roll → new short calendar), AMD, ECHO and TLT (all opens → transfer or close).

The standing lesson, stated again because today proves it three times over: a large print is not a large position. Volume and premium tell you money moved; only the next morning's open-interest count tells you whether exposure was created, destroyed, or merely handed to somebody else. Come back next trading day pre-market (≈06:30 ET) for the same check on the current session.

---

⚠️ Risk & Reality Check

A large trade tells you somebody with capital took a position. It does not tell you they were right, and it is never a recommendation.

What today should make plain: the biggest number on this page — $286.5M — means one thing if it is hedged and something quite different if it is not, and we could not find the hedge. We have told you that rather than picking the tidier story. Yesterday's review is the reason why: five of thirteen names changed meaning overnight, and two of the largest dollar figures turned out to establish no new position at all.

Three specific cautions:

• A credit is not a profit — and it is not always risk being taken on. GLD collected $20.4M and accepted a $90.2M tail; that one is confirmed new. TTWO's $8.9M credit turned out to be a bullish position being closed, so no tail was assumed at all.
• A big number is not a big position. AMD's $54.4M, TLT's $20.4M and ECHO's $12.6M created no new open interest between them.
• "No earnings inside the expiry" is a real weakness. MSFT's spread has no catalyst to work with, only time and drift.

Position sizes you can hold through a bad week, and the patience to skip what you do not understand, will do more for your results than any flow alert. Nothing here is investment advice.