📊 AInvest Option Flow Digest — 2026-09-09: Nine Days to Triple Witching, the Desks Are Clearing Out of September

$181.4M net premium across 12 names. Five are the same trade. Four are bitcoin proxies pointing in different directions. And on three of them, the number you saw in the feed was the wrong number.

⚡ The Quick Read

Today's board has an unusually clean shape, and it is worth understanding before any individual ticker.

Five of the twelve names are rolls off the September 18 expiration. NVDA, AMD, PLTR, MSTR and COIN all involve a deep-in-the-money September call being moved, closed or unwound. September 18 is quarterly triple witching and the S&P 500 quarterly rebalance — the most flow-distorted session of the quarter. None of those five has a company-specific catalyst on that date. This is desks tidying up before a messy expiry, not a wave of fresh conviction.

The September calls being rolled are nearly all intrinsic value. NVDA's $195 call traded at $29.70 against $29.45 of intrinsic. AMD's $470 call: $54.61 against $52.21. PLTR's $135 call: $35.95 against $35.68. Contracts with almost no time value left behave like stock, not like options. Rolling them is maintenance.

Four crypto-linked names printed today, and they do not agree with each other. MSTR reversed a bitcoin-proxy position it had just put on. COIN rolled its exposure out two months. IBIT bought a 15-month bull call spread on bitcoin itself. CRCL sold downside puts more than two years out. Same theme, four different answers — which is a much more honest picture of "institutional positioning" than any single trade.

Three trades were not what the feed reported — and that is today's most useful lesson.

• AFRM was shown as a $1.2M call purchase. The full option chain for that same millisecond held a second leg the feed never displayed — a short $100 call of identical size. It is a spread, and the money at risk is $862,500, not $1.2M.
• M was shown as a $2.07M bullish call buy. A 510,000-share block printed 108 seconds later, sized to the exact delta. It had essentially zero directional exposure on day one.
• NXPI was shown as two call purchases. The paired stock hedge was 18,540 shares — about twelve times too small to cover both legs being long. The arithmetic says one leg is almost certainly short.

Three different failure modes: a missing leg, a hidden hedge, and a mislabelled side. In each case the tape held the answer and the feed did not.

Two rate-sensitive fintechs, opposite exposures, both got bullish structures. AFRM funds a loan book with borrowed money, so a Fed hike hurts it — and someone bought a capped call spread into an 11% slide. CRCL earns 95% of its revenue as interest on reserves, so a hike helps it — and someone sold downside puts after a 53% August. Both trades are defined-risk. Neither is a pure rate bet. But the pairing is a good reminder that "rate-sensitive fintech" is not one trade.

Gross premium was $670.0M. Net was $181.4M. That gap is the whole lesson. Gross counts both legs of a spread; net counts the money actually at risk. Anyone quoting today's flow as "over $650 million" is counting the same dollars twice.

The one-day reversal is the story of the day. Yesterday a desk put on a $74.5M bullish-looking roll in MSTR. Today the same two contracts traded the opposite way, leg for leg. That position lasted 24 hours.

And the cleanest signal was the smallest trade. Almost every large print today was negotiated — floor blocks, facilitated auctions, crosses, all with a known counterparty. Exactly one was a genuine lit, liquidity-taking order: LYFT's $1.11M put buy, the smallest premium on the board.

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

The September 9 pre-market open-interest snapshot settles all thirty-three legs from yesterday's board. Twenty-eight opened, three closed, two were transfers. Eighteen of the twenty legs where we published a lead branch landed on it. One structure inverted outright.

🔄 The inversion

DRAM's "$3.89M barbell" was a position coming OFF, not going on. We published all three branches and refused to pick one, and the CLOSE branch won on both legs at essentially full capture: the January 2028 $90 call fell 45,963 → 34,185 (−11,778 on 12,000 printed, 98.2%) and the $40 call 24,455 → 18,468 (−5,987 on 6,000, 99.8%). Four untraded control strikes on the same chain were flat, so this is a real unwind, not an exercise sweep. Both lines were built on a single day, July 31, 2026, in a 1.96:1 ratio — and yesterday took them back off in a 1.97:1 ratio at roughly 27% of the original size. Same structure, in reverse. Corrected DRAM analysis

📉 The two calls we told you to hold us to: right direction, wrong size

Leg	Size	Baseline → Resolving	Δ	Capture	Our published band
AMZN Sep-18 $245C	9,010	14,297 → 10,864	−3,433	38.1%	171 contracts above the band
MSTR Sep-18 $100C	11,550	36,119 → 33,397	−2,722	23.6%	≈1,900 above the band

Neither rose, so neither reading was inverted — but both fell by far less than our own close branch required. Score them as partials. Roughly a quarter to a third of each block genuinely retired old exposure; the rest met counterparties who were opening. AMZN · MSTR

That MSTR line matters today: only 23.6% of yesterday's September $100 call block actually closed — and today the same desk came back and bought 14,400 more of it.

⚠️ One pure transfer

MU's January 2027 $1,200 call added 50 contracts on a 500-lot block — 4,851 → 4,901, 10.0% capture. One desk bought what another sold. MU's other three legs opened normally, so the short-volatility package is real; this leg was simply a change of ownership. MU

✅ Confirmations

• ROIV's roll confirmed on both legs — and bigger than the feed showed. February 2027 $45 call opened on a virgin strike (0 → 8,848), December $39 call closed at 100.8% capture. A control check found the neighbouring December $40 call also fell by 7,609. The unwind covered two strikes, not one. ROIV
• IEF's jelly roll was built from scratch, all four legs, 108-110% capture. IEF
• GOOGL's missing leg is corroborated. We published it as a $370/$440 spread after finding the short wing on the tape; both lines grew by within two contracts of each other, +27,509 and +27,507. Two lines moving in that lockstep are one order. GOOGL
• MOD landed to the contract — two lines that held 1 and 0 contracts for 26 sessions now hold exactly 750 each. MOD
• AMD, SMH and MU all opened, so the semiconductor short-volatility program is genuinely new exposure. AMD · SMH
• GFL and NBIS both opened past their published figures. GFL's direction remains permanently unknowable — open interest cannot separate a spread from two outright buys. GFL · NBIS

What changed: the dollar figures hold at $217.65M net, but roughly $30.65M of yesterday's gross belongs to a position being retired rather than established. Eleven of thirty-three legs printed at or below existing open interest, and on every one we published no lead branch. Had we guessed on the two that closed, the DRAM headline would have been exactly backwards.

The standing lesson: a big headline premium tells you the size of a structure, not whether anyone just took a position. Only the next morning's open interest tells you which — and getting the direction right is not the same as getting the size right.

📋 At a Glance

Ticker	Net Premium	Expiry Class	Catalyst (and when)	The Option Play	What It Actually Means
NVDA	$28.4M credit	Quarterly → Monthly	Earnings unconfirmed: Nov 17 or Nov 25	Sep-18 $195C rolled up and out to Nov-20 $210C	Stays long, cuts delta, takes cash off the table
AMD	$0.5M credit	Quarterly → Monthly	Q3 earnings ≈Nov 3 —   after   both expiries	Sep-18 $470C rolled to Oct-16 $490C	Five more weeks bought for almost nothing
PLTR	$64.2M debit	Quarterly + Monthly	Q3 earnings ≈Nov 2-9 — inside the Nov expiry	Near-parity Sep-18 $135C + fresh Nov-20 $165C	Stock-replacement plus upside — or a roll
MSTR	$22.4M debit	Quarterly + Monthly	Fed decision Sep 16; MSCI result by Oct 16	Yesterday's roll reversed, leg for leg	Risk coming   off , not going on
COIN	$15.2M debit	Quarterly → Monthly	Q3 earnings ≈Oct 29 — inside the Nov expiry	Same $170 strike pushed Sep-18 → Nov-20	Buys two months and lands on earnings
TSLA	$31.4M debit	LEAP	Robotaxi, federal safety probe, governance	Sep-2027 $370 straddle, both legs	Pure volatility bet — no direction at all
IBIT	$5.3M debit	LEAP	Bitcoin price; Fed decision Sep 16	Dec-2027 $45/$55 call spread, 15,000 wide	Defined-risk bullish bitcoin, capped upside
CRCL	$3.6M credit	LEAP	Fed decision Sep 16 — for Circle a   hike helps	Dec-2028 $100/$65 ratio put spread	Short a moderate dip, long an outright crash
NXPI	$6.4M debit *	LEAP	Q3 earnings ≈Oct 27 (estimated) — lands on an FOMC day	Jan-2027 $290/$320 2:1 call ratio + stock hedge	Cheap bet on a big upside move, not plain bullish
M	$2.1M debit	LEAP	Earnings tomorrow, Sep 10 pre-open	Jan-2027 $23C paired with 510,000 shares	Delta-hedged —   not   a bullish bet
AFRM	$0.86M debit	Monthly	Rate-sensitive lender; Fed decision Sep 16	Dec-18 $80/$100 call spread (feed showed half)	Capped bullish bet bought into weakness
LYFT	$1.1M debit	LEAP	Q3 earnings ≈Nov 4-11; robotaxi competition	Mar-2027 $13 puts bought on the offer	Downside protection or a bearish view

Totals: $181.4M net · $670.0M gross · 12 tickers · 31 legs. Net is the number that matters — it is the capital actually committed.

* NXPI is carried at the figure the feed labels. If the hedge arithmetic is right and the $290 leg is short, the true net is closer to $46,000 and the day's net is ≈$175.1M. We flag it rather than quietly picking a side on a cross we cannot prove.

🍂 Theme 1: The September Exodus

Each of these is a deep-in-the-money September call being moved off the quarterly expiry. Read them as one behaviour, not several signals.

🔄 NVDA — $318.5M gross, and the desk got paid to roll

Four linked packages, 58,300 contracts total. Two printed as floor blocks, two as facilitated auctions — different mechanisms, same structure. The September $195 call was sold and the November $210 call bought, for a net credit of ≈$28.4M. Selling a near-parity call and buying a cheaper, higher strike takes cash out while staying long.

The wrinkle worth clicking for: NVIDIA has not confirmed its Q3 earnings date. One provider says November 17, others say November 25. If it lands November 17-19, those November 20 calls hold the earnings event. If November 25, they expire days before it. That is a genuinely open question sitting inside a $318M structure.

One honest caveat we put in the article: the largest package printed with SELL on both legs, which does not fit a clean roll. On a floor-negotiated print the side is not provable, and we say so rather than picking one. → Read the full NVDA breakdown

🔄 AMD — the roll that cost ≈$0.51M on $87.2M of gross premium

Essentially free. The September $470 call held only ≈$2.40 of time value against $52.21 of intrinsic, so giving it up cost almost nothing — and bought five extra weeks and a $20 higher strike. Neither expiry contains earnings (≈November 3). This roll buys news flow, not the event. AMD is up +144% year to date, which is exactly when desks start managing rather than adding. → See why this roll was nearly free

🎯 PLTR — $64.2M, and the ambiguity is the point

11,266 contracts on each leg: a September $135 call trading at essentially pure parity, plus a fresh November $165 call. The November leg is a provable open. The September leg is not — its size sits below existing open interest, so we cannot tell whether it opened or closed. Both readings are live, and we publish both. Worth knowing while you read it: the valuation debate in this name is loud, and a well-known short seller is positioned in March 2027 puts. → Package or roll? The evidence both ways

₿ Theme 2: Four Crypto Names, Four Different Answers

This is the most interesting cluster on the board. Four bitcoin-linked names printed size on the same afternoon, and if you were hoping "institutional flow" would tell you what smart money thinks of bitcoin, here is your answer: it disagrees with itself.

🔄 MSTR — put on Tuesday, taken off Wednesday

This is the one to read if you read only one. Yesterday: sell the September $100 call, buy the November $135 call, 11,550 lots. Today: the exact opposite, 14,400 lots. The proof is in the open interest — the November $135 line sat flat near 2,425 for weeks, jumped to 13,982 after yesterday's trade, and today the desk sold 14,400 of it.

Why it probably reversed: a September 8 filing showed zero bitcoin purchases and zero share issuance for the prior week, and expectations firmed for a Fed rate hike on September 16 — two days before the September expiry. Holding near-parity September calls with ≈0.95 delta into a two-sided Fed meeting is close to holding the stock outright. → The 24-hour round trip, with the open-interest proof

🔄 COIN — the cleanest roll on the board

Same $170 strike, simply pushed from September to November for ≈$15.05 per contract. The November leg opened against prior open interest of just 498 — roughly 20x, about as clean an open as this data offers. The logic is legible: September 18 holds no Coinbase catalyst but sits right behind a Senate vote and an FOMC meeting, while the November contract contains Q3 earnings. → Stepping around the Fed, landing on earnings

₿ IBIT — a patient, capped bet that bitcoin recovers

15,000 contracts on each leg of a December 2027 $45/$55 call spread on the iShares Bitcoin Trust — roughly 15 months out. Net cost $3.55 per share of spread against a $10-wide structure: $5.3M at risk, $9.7M maximum profit, a little better than 1.8 to 1. Breakeven is $48.55, about 9% above spot, and it pays in full at $55, about 24% higher.

What makes it interesting is the timing, and the picture here is more nuanced than the usual "crypto is weak" line. Bitcoin is ≈$78,600 — up ≈28% over three months from a late-June low near $57,700, with August alone up ≈25% — yet still ≈38% below its October 2025 record, and softer over the last week. So this is neither bottom-fishing nor chasing a high.

There is a genuinely odd thing in the data worth your attention: the 10-year real yield rose ≈23 basis points over those same three months while bitcoin rallied 28%. A non-yielding, long-duration asset is not supposed to do that, and some analysts now describe bitcoin as trading more like gold than like a tech stock. Note the limit of that claim, though — on September 9 itself bitcoin fell alongside equities on geopolitical headlines. The decoupling is from the interest-rate channel, not from risk appetite generally.

And note the contrast: on the same afternoon, MSTR's desk was reducing its bitcoin-proxy exposure while this one was adding. → The 15-month bitcoin spread, and what it needs to pay

🧊 CRCL — short a moderate dip, long an outright crash

Three prints in the same second, all expiring December 2028 — more than two years out. Short 2,500 of the $100 put, long 3,375 of the $65 put. That is a ratio put spread: more long wings than short bodies, for a net credit of ≈$3.6M.

The shape is the story. Selling the $100 put with the stock at $94.30 says "I am comfortable owning this on a moderate decline." Buying 1.35 cheap $65 puts for every short body says "but if it truly collapses, I want to be paid." It is a structure that loses on a drift lower and profits on a crash — an unusual and quite specific opinion.

The company behind it makes that opinion legible. 95% of Circle's revenue is interest income on the reserves backing USDC — $667.7M of $701M last quarter. So unlike almost every other financial on today's board, higher rates help Circle, which makes a September hike a tailwind here rather than a headwind. Set against that: USDC supply has essentially stalled, growing ≈1.5% in ten weeks, and the stock has run ≈53% in August alone. Selling downside puts after a run like that is a different proposition from selling them into weakness. → The two-year structure that wants a crash

🌱 Theme 3: Everything Else

🎲 TSLA — $31.4M on a one-year straddle

2,250 calls and 2,250 puts, same $370 strike, same September 2027 expiry. A bet on a big move in either direction with no directional opinion whatsoever. The cost: ≈$139.68 per share of straddle against a $368 stock, so Tesla must move roughly 38% either way just to break even. We pulled the implied volatility straight off the tape — ≈48% blended, above shorter-tenor implied and well above recent realized. Whoever bought this paid up for vol. → What a 38% breakeven actually requires

🔬 NXPI — the hedge that gives away the structure

Two January 2027 call lines printed in the same instant at an exact 2:1 ratio: 9,270 of the $320 strike and 4,635 of the $290 strike. Both marked as purchases. NXP was trading at $224, so those strikes sit ≈29% and ≈43% above the stock — cheap, far out-of-the-money calls.

Then the equity tape settles it. A 18,540-share block printed 43 seconds later carrying the contingent-trade marker. Work out what that hedge implies: if both call legs were long, the package would carry ≈221,553 shares of delta and need a hedge roughly twelve times larger than the one that printed. If the $320 leg is long and the $290 leg is short — a 2:1 backspread — the two legs largely cancel and the delta is ≈16,129 shares, which is close to what actually traded.

So the hedge size, not the feed label, tells you the shape. A 2:1 call backspread costs almost nothing up front, is close to delta-neutral at inception, and pays only on a large move higher. It is a cheap lottery on a big upside move, not a conventional bullish position. We present it as strong evidence rather than proof — on a cross the sides are never provable, and the desk's own delta assumptions may differ from ours.

Why someone might want that here: NXP has been taken apart. It is down ≈19.7% since the end of June and ≈33.6% off its 52-week high, after falling 16.3% in five sessions on a quarter that grew revenue 19%. Buying far out-of-the-money upside for almost no net premium, in a name that has already been crushed, is a coherent way to own a recovery without paying for one. Note too that the estimated October 27 earnings date lands on a Fed decision day — two variance events on one morning. → The 12x hedge mismatch, worked through

🤝 M (Macy's) — the "call buy" that is not a bet

The feed shows a $2.07M call purchase the day before earnings. The tape shows something else: a 510,000-share block printed 108 seconds later. The call's delta was 0.51, and 10,000 contracts × 100 × 0.51 = exactly 510,000 shares. The package was delta-hedged at inception — essentially zero directional exposure on day one. With two ex-dividend dates before the January expiry and ≈11.5% short interest, this looks like a financing or volatility structure, not a view on the stock.

Macy's reports tomorrow, September 10, before the open. That is the fact to carry into any position here. → Why the headline is misleading, and what the hedge proves

🧩 AFRM — the feed showed us half the trade

Reported as a straightforward $1.2M purchase of 2,500 December $80 calls. Scanning the whole option chain for that exact millisecond turned up a second leg that never appeared in the feed: 2,500 December $100 calls sold at $1.40, same size, same instant, same floor-negotiated print.

So it is not an outright call buy. It is an $80/$100 bull call spread, and the money actually at risk is the net debit of $862,500, not $1.2M — about 29% less. The upside is capped: maximum profit $4.14M, roughly 4.8 to 1. Breakeven is $83.45 against $68.87 at the print, so it needs a ≈21% move in about 100 days just to break even.

The setup around it is the interesting part. AFRM was down ≈4.4% on the session and ≈11.4% over eight sessions, straight through seven analyst target raises and two new Buy initiations — and no company-specific news explains it. What does fit is rates: this is a lender that funds its loan book with borrowed money, the July Fed meeting drew three dissents preferring a hike, and the two-year Treasury sits well above the funds ceiling. The Fed meets September 15-16. The December expiry also contains the next earnings report, estimated November 5 and not company-confirmed.

Buying a capped structure rather than an outright call, into that macro, is the considered version of being bullish.

This is the second time this week the displayed feed has understated a structure. It is worth internalising: what you see in a flow feed is a report, not the trade. → The leg the feed didn't show you

🛡️ LYFT — the smallest trade, the cleanest signal

$1.11M — the smallest premium on the board — but the only genuinely lit, liquidity-taking print of the day. That matters, because lit is the one mechanism where reading the aggressor side is actually valid. A buyer took 9,995 March 2027 $13 puts on a day the stock was already down ≈6.7%.

The tape detail we liked: a large resting offer had been worked all afternoon — 5,000 lots, then 4,997, then 9,995 — and the trade took exactly the displayed size. Implied volatility barely moved. So it is a real buy, but not a panicked one. Someone was waiting to sell to them. → How to read the only lit trade of the day

📅 Upcoming Catalysts — and Which Expiry Actually Holds Them

Catalysts and option expirations are two different calendars. Confusing them is one of the most common ways retail traders lose money following flow. Here they are kept separate.

Date	Event	Which of today's expiries contains it?
Sep 10 (pre-open)	Macy's Q2 earnings	Inside M's Jan-2027 LEAP — but that position is delta-hedged
Sep 11	AMD at Goldman Sachs conference	Inside AMD's Sep-18 leg (the one being sold)
Sep 15 (one source says Sep 10)	Senate CLARITY Act cloture test — imminent, and our sources disagree on the exact date	Either way it falls   before   Sep-18, the expiry COIN rolled away from
Sep 15	Macy's ex-dividend date	Inside M's LEAP window (one of two before January)
Sep 16	Federal Reserve decision   (hike expectations rising)	Two days   before   Sep-18. A key reason MSTR reversed, and a live risk for AFRM, CRCL and IBIT
Sep 18	Quarterly triple witching + S&P 500 rebalance	The expiry five of eleven names are leaving
Oct 12-15	OCP Global Summit	Ends one day before AMD's Oct-16 leg expires
Oct 16	MSCI consultation result   on non-operating companies	Straddled by MSTR's Nov-20 leg — the one just sold
≈Oct 27	NXP Q3 earnings (estimated, not company-confirmed) — and it   collides with the Oct 27-28 Fed meeting	Inside NXPI's Jan-2027 structure. Two variance events stacked on one date
≈Oct 29	Coinbase Q3 earnings (not company-confirmed)	Inside COIN's Nov-20 leg
≈Nov 2-9	Palantir Q3 earnings (not company-confirmed)	Inside PLTR's Nov-20 leg
≈Nov 3	AMD Q3 earnings (not company-confirmed)	Outside   both AMD expiries
≈Nov 4-11	Lyft Q3 earnings (sources disagree)	Inside LYFT's Mar-2027 LEAP
Nov 17 or Nov 25	NVIDIA Q3 earnings — genuinely unconfirmed	Nov 20 holds it only if the earlier date is right
≈Nov 5	Affirm FQ1 earnings (estimated, not confirmed)	Inside   AFRM's Dec-18 call spread
Sep 15	Affirm risk + capital-markets executives at Barclays conference	Inside AFRM's Dec-18 spread, and the day before the Fed
Dec 18	—	AFRM's call spread expires. It needs the move by then
Oct 19	Comment deadline on the Treasury stablecoin rulemaking	Inside CRCL's Dec-2028 structure
Jan 18, 2027	GENIUS Act statutory effective date.   The affiliate-yield presumption that would touch Circle's distribution economics is still   proposed, not final	Well inside CRCL's Dec-2028 structure
Dec 2027 / Dec 2028	No scheduled events this far out	IBIT and CRCL are macro bets, not event bets

Note how many of these are estimated rather than company-confirmed. We flag that in each article rather than presenting a guess as a date.

🎯 For Four Kinds of Reader

🚀 The YOLO Trader

Be honest about what today actually offers you: not much that is urgent. Most of the board is negotiated flow with a known counterparty — no one is rushing. The trades with real directional content are LYFT's put buy (small, lit, genuinely aggressive), AFRM's call spread (defined risk, bought into weakness) and IBIT's bitcoin spread (defined risk, 15-month horizon).

Notice that all three of those are defined-risk structures, not naked long options. That is not an accident — it is what people do when they want exposure without betting the outcome on timing. If you cannot resist, size at 1-2% and copy the structure, not just the direction. MSTR is the cautionary tale: a $74.5M position that lasted one day.

📈 The Swing Trader

The tradeable observation is the calendar, not the direction. Five desks decided September 18 was not worth holding through — triple witching, the S&P rebalance, and a Fed meeting two days prior. If institutions are stepping around that session, treat mid-September price action in these names as noise rather than signal.

COIN and NVDA give you the cleanest roadmap: both moved exposure onto expiries that actually contain earnings. And the crypto cluster is a useful lesson in humility — four desks, four different views on the same underlying theme. Watch whether next-day open interest confirms these rolls before assuming any position is real, and remember AMZN and MSTR yesterday: direction right, size much smaller than expected.

💰 The Premium Collector

Today is unusually instructive for you. NVDA and AMD both rolled for a credit, which tells you what near-parity September calls were worth: almost nothing in time value. That is the same edge you harvest — you just want to be the seller of time, not the buyer.

CRCL is the one to study. Someone sold 2,500 two-year puts and used part of the credit to buy 3,375 much cheaper puts far below. That is premium selling with the tail bought back — it gives up some income in exchange for surviving a crash. If you sell puts naked, look hard at that structure; the wing is what turns a good year into a survivable bad one.

The other side of TSLA's straddle is worth thinking about too: someone is now short a year of Tesla volatility at ≈48% implied. That is above recent realized, which is the setup you want — but Tesla's realized volatility comes overwhelmingly from overnight gaps, and gaps are exactly what short-vol positions cannot manage.

🌱 The Beginner

Four things from today are worth more than any single trade idea.

Gross is not net. Today's flow was $663.6M gross but only $175.1M net. A spread has two legs; adding them up counts the same dollars twice. Always ask which number you are being shown.

What you see in a feed is a report, not the trade. This happened three separate ways today. AFRM was displayed as a $1.2M call buy, but a second leg was missing and the real risk was $862,500. M looked like a bullish call buy but had a 510,000-share hedge attached that cancelled the direction. NXPI showed two call purchases, but the hedge that printed was twelve times too small for that to be true. Different failure modes, same lesson: the feed is a summary, and summaries lose things.

A big premium is not a conviction signal. MSTR proves it: a desk put on a $74.5M position and took it off the next day. And M's "$2.07M bullish call buy" had zero directional exposure once you find the 510,000 shares attached to it.

How a trade prints matters as much as its size. A negotiated floor block or cross has a known counterparty and expresses no urgency. A lit order that takes liquidity does. Today's smallest trade was the only lit one — and arguably the most informative.

⚠️ Risk Control — Read This Part

Do not follow unusual options activity blindly. Today's board makes the case better than we could.

• You cannot see the whole position. A desk rolling a September call may be hedging stock you cannot see, or unwinding a structure built months ago. Open interest is side-blind — it can tell us whether contracts were created or destroyed, never who bought.
• Negotiated flow is not a signal of urgency. Floor blocks, auctions and crosses all have a willing counterparty who agreed on price in advance. That is position management, not someone rushing to get in.
• Size below existing open interest proves nothing about intent. On several legs today we genuinely cannot tell open from close, and we say so rather than guessing. Come back tomorrow ≈06:30 ET, when the open-interest snapshot resolves them.
• The feed can show you part of a trade, or the wrong side of it. Three of today's twelve names were misreported: a missing leg (AFRM), a hidden stock hedge (M), and a side the hedge arithmetic contradicts (NXPI). If you size off a headline premium, you are sometimes sizing off the wrong number entirely.
• Positions can be reversed. MSTR's lasted 24 hours.
• Patience beats participation. There is no requirement to trade any of this. The most useful thing today's flow tells you is a calendar fact — September 18 is a session institutions chose to step around — and that is information you can use without putting on a single position.

Never risk more than you can lose entirely. Options can and do expire worthless.

Every figure in this digest was verified against the OPRA tape, official open-interest snapshots and the equity tape. Where something could not be proven, we say so rather than filling the gap with a guess. Trade dates and prices reflect an open trading session on September 9, 2026.

Not investment advice. Options involve substantial risk and are not suitable for every investor.