AInvest Option Flow Digest — 2026-09-08

Ten names, twenty-nine legs, $385.13M of premium — and the three biggest are the same trade.

Between 10:26 and 10:44 this morning, MU, SMH and AMD printed the identical structure in the identical leg ratio: sell January-2028 puts and calls, buy January-2027 puts and calls, all at one strike above the money, weighted 2.5 : 1.5 : 1 : 1. Three correlated semiconductor names, eighteen minutes apart, the same arithmetic each time.

We measured the delta on each package and all three come out flat — MU −1,758 shares, AMD +970, SMH +9,335, on packages of $50M to $89M. That ratio is not a preference. It is the weight that cancels direction. What is left is a short position in long-dated semiconductor volatility: ≈$2.08 million per volatility point, for $120.38 million of credit collected. A five-point rise in long-dated chip vol costs them ≈$10.4M.

⚠️ And the feed dropped legs on THREE names — each time changing what the trade is. MU's package was reported as three legs; the tape shows four, with a $5.23M January-2027 call missing. IEF's was reported as two call legs and looked like a bet on bonds; the tape shows four — both expiries carry a matching put — making it a financing trade with almost no view on bond prices at all. And GOOGL was reported as $22.3M of calls bought outright; the tape shows a matching 20,000-lot short leg at $440, which turns it into a spread costing $18.5M instead — and it is the cleanest directional bet on the board.

⏳ Nothing is settled until tomorrow morning. Ten of today's twenty-nine legs printed at or below the open interest already standing on their strike. On those we publish no lead branch at all — the tape genuinely cannot separate an opening trade from a closing one, and last week taught us twice what happens when we guess.

🔁 OI Review — Friday's Predictions, Scored

The September 8 pre-market open-interest snapshot is in — the first one since Friday, because September 7 was Labor Day — and it settles all twenty-two legs from Friday's board. Every one of the eighteen legs where we published a lead branch landed on it. But the four legs we deliberately refused to call went two ways, and two of them inverted the structure we had named: QQQ was not a calendar, and URA was not a roll.

🔄 The inversions

1. QQQ's "$18.3M call calendar" was a long call rolled out to year-end — and it was directional all along. We led with "the geometry looks bullish. It isn't," on the strength of a package that netted to −0.05 delta. The October $700 line FELL 19,586 → 9,738 (Δ −9,848 on 10,000 printed, a 98.5% close) while the December line opened cleanly (2,198 → 12,205). A calendar creates a new short in the front month; this destroyed the front month. The arithmetic leaves no room: that line traded 10,213 contracts all session, so the block itself had to take out at least 9,635 — and a block only shrinks open interest when both sides close. Neighbouring October strikes were flat (−1, +4, −4, +284), so no exercise sweep. What survives the print is 10,000 long December $700 calls and nothing else: ≈653,200 shares of long delta. The −0.05 measured the change in delta from moving 76 days out, not the resulting position. Our own falsification test fired — we had written that a falling October leg would change the read. Corrected QQQ analysis

2. URA's "$50 call rolled out a month" was a brand-new calendar spread — nothing was closed. We called the roll obvious from 28,304 contracts of standing September interest against a 9,052-lot print. The September $50 line ROSE 28,304 → 30,988 (Δ +2,684). A roll needed it to fall toward ≈19,252. Again the arithmetic is airtight: with 10,211 contracts of line volume and a 9,052-lot block, the block must have contributed at least +1,525 to open interest, and a block cannot add open interest while closing. Meanwhile the October $50 line opened hard, 2,287 → 17,231 (+14,944, 165% of the block). Both legs are new: short September, long October — a position on time, not a rescue of a losing bet. The $353,028 collected is premium on a fresh short, not salvage from an old long. Our own falsification test fired here too. Corrected URA analysis

⚠️ These two are the same mistake in mirror image, and it is worth naming precisely. On QQQ we assumed a near-month sale was a new short, so we called it a calendar. On URA we assumed a near-month sale was an old long coming off, so we called it a roll. A near-month sale against a far-month buy prints identically either way. The only thing that separates them is which way the near line's open interest moves the next morning: grows ⇒ calendar, shrinks ⇒ roll. Neither can be read off the tape, and neither belongs in a headline before the snapshot.

The lesson, and it is a different one from last week's. On September 3 we got burned declaring fifteen legs settled opens in advance, from size against open interest. This time every open/close call we made was right — the errors were structure calls, made from the geometry of a two-leg print. Where we refused to pick a branch on open versus close, the snapshot did the work honestly. Where we named a structure in a title, we were wrong twice out of eleven.

🔀 Two forks the snapshot settled the other way

• SPY's $760 put OPENED, and the inversion we warned about did not happen. Size sat below prior open interest so we published no lead branch, and we flagged that a rising $745 line against a falling $760 line would turn the trade into a short-put buyback. Both rose, and both past their ceilings: the $760 line took on 43,751 new contracts (97.7% of its 44,784 session volume) and the $745 line 36,165 (92.9%). The put debit spread is genuinely new bearish exposure into September 18. SPY
• GE's December $340 put CLOSED, confirming the roll down and out. No lead branch was published; the close branch won at 82.7% capture (14,744 → 4,741), inside the 40-100% band. The January 2027 $300 put opened at 119.8% (862 → 15,353) — past the full-open figure, so other buyers were building the same line. A month-long hedge that began on GE's all-time high was banked and replaced with thinner cover. GE

✅ Confirmations

• EWY was the call we asked you to hold us to, and it held. Three legs that each printed below standing open interest all FELL, two of them by the printed size to the contract: $160 call 8,243 → 2,008 (93.5%), $245 call 11,212 → 1,212 (100.0%), $120 put 5,575 → 2,242 (100.0%). Not one line rose. A control strike on the same June 2027 chain was flat. Tranche two of a two-day unwind — never a new Korea bet. EWY
• VIX was the widest margin we have ever published, and it landed past the band. Both dormant lines absorbed essentially their entire session volume as new open interest: the $31 call 462 → 128,234 (99.7% of volume) and the $34 call 804 → 143,761 (99.6%). The failure mode we explicitly braced for — heavy volume, open interest barely moving, the way PDD went — did not recur: PDD produced −430 contracts on 19,060 printed; this produced +270,729 on 271,602. VIX
• LQD opened three-for-three, each within two contracts of the predicted figure — $104 put 0 → 20,000 (exact), $105 put 398 → 10,400, $107 call 1,153 → 11,153 (exact). The arithmetic floors we published were never tested. LQD
• SMH opened on both legs, so the fresh risk reversal is real — $535 put 6,328 → 18,753 (95.6%), $615 call 2,883 → 15,472 (96.8%). Neither leg lagged, which was the published test. The 702,000-share hedge is attached to a genuinely new position. SMH
• IGV opened past the top of the band — 5,097 → 19,943 (+14,846, 121.7%), above the ≈17,297 ceiling, meaning the block opened in full plus ≈2,600 more from others on the line. IGV
• TYRA opened on both legs above their upper bounds — the $30 line, frozen at exactly 1,164 for six straight sessions, went to 3,702; the $45 wing 1 → 2,717. The event spread was real and live into Wednesday's readout. TYRA
• KWEB opened at 97.1% — 28,706 → 125,758 — and direction is still unknown, permanently. The print carried no reported side and crossed at the mid; open interest is side-blind. A confirmed open makes that unknown bigger, not smaller: there is now a real 100,000-contract January 2027 line where there wasn't one, and nobody can say who is long it. KWEB

What changed, and the standing lesson

Friday's board published as $114.48M net across eleven tickers. Nothing about the money moved — every price and every size was right, so the net figure is unchanged. What moved is what two of those structures are: QQQ's $18.26M bought a directional long call rolled to year-end rather than a delta-neutral calendar, and URA's $823,732 opened a fresh calendar rather than rescuing an existing position. The count of non-directional structures on the day goes from seven to six.

The score to remember: size against open interest tells you about open versus close, and it told the truth on all eighteen legs where we used it. It tells you nothing at all about what the structure IS. A near-month sale paired with a far-month buy is a calendar or a roll depending entirely on which way open interest moves the following morning — and the tape, the premium and the geometry are identical either way. Only the next-morning snapshot separates them.

📋 The Board at a Glance

Ticker	Premium (net)	Expiry tag	The trade	What it means	Catalyst
MU	$51.08M credit   (of $88.82M gross)	LEAP   — Jan 2027 + Jan 2028	Sell the 2028 straddle 2.5:1.5, buy the 2027 straddle	⚠️   Volatility, not direction   — delta −1,758 shares	Earnings confirmed Sept 30
AMD	$38.95M credit   (of $65.89M gross)	LEAP   — Jan 2027 + Jan 2028	The same structure, same ratio, 18 minutes later	⚠️   Delta +970 shares — flat	Q2 beat and raise,   stock fell 8%
SMH	$30.35M credit   (of $50.81M gross)	LEAP   — Jan 2027 + Jan 2028	The third ticket in one program	⚠️   Delta +9,335 shares — flat	Micron Sept 30, NVIDIA ≈Nov 18
DRAM	$3.89M credit   (of $30.65M gross)	LEAP   — Jan 2028	A 2:1 ratio call spread, $90 against $40	A barbell   — pays on a bust or a melt-up	A memory ETF   five months old
AMZN	$4.96M debit   (of $29.28M gross)	Quarterly → monthly	A call   rolled up $10 and out two months	Buys the Q3 print, gives up 31% of its delta	Q3 earnings ≈Oct 29
MOD	$4.17M debit   (of $9.68M gross)	LEAP   — Sep 2027	A   synthetic short   on two empty lines	⚠️   The $260 strike is not the breakeven — $204.40 is	A pro-rata spin-off due by year-end
GFL	$2.06M debit   (of $4.89M gross)	Quarterly   — Dec 18	A $45/$50 call structure, 12,500 each	⚠️   Direction unproven — both readings published	Take-private interest disclosed July 29
IEF	$2.68M credit   (of $11.54M gross)	LEAP   — Apr 2027 + Jan 2028	A   jelly roll   — two synthetic forwards	Financing, not bonds. Delta ≈ zero	Six Fed meetings inside the window
GOOGL	$18.49M debit   (of $26.13M gross)	Monthly   — Nov 20	A   $370/$440 call spread , 20,000 wide	Directional and bullish — +514,600 shares of delta	Ad-tech ruling landed Sept 4; earnings ≈Oct 27
MSTR	$19.87M credit   (of $67.45M gross)	Quarterly → monthly	A call   rolled up $35 and out two months	Cuts delta 38%   — the second roll today	MSCI decision due by Oct 16
	$176.48M net		$385.13M gross

Net is the debit paid or the credit collected per package. Gross double-counts every spread.

🔍 What's actually interesting today

The same trade, three times, in eighteen minutes

This is the clearest example we have published of why headline premium misleads. A flow alert reading "someone sold $92 million of premium across Micron, AMD and the semiconductor ETF" invites exactly one conclusion — somebody is bearish on chips. The delta says otherwise, on all three.

Selling 2.5 puts against 1.5 calls at a strike above the money, and buying one of each at a nearer expiry, produces a package whose directional exposure is close to nil. On $205.52 million of gross premium, the three packages together carry less than 9,000 shares of net delta. You cannot construct that by accident.

So what is the actual bet? That a year from now, semiconductor volatility will be cheaper than it is priced today. They sold January-2028 vol and bought January-2027 vol. In AMD and SMH the back month was genuinely richer — 58.1% against 56.3% in AMD, 38.7% against 36.7% in SMH — so they sold the more expensive expiry. In Micron the curve was flat at ≈68% and the trade is simply short vol outright.

⭐ The sharpest argument in the structure is about which events they own. Everything with a date on it falls inside the long January-2027 legs: Micron's September 30 report, eight semiconductor earnings between October 14 and November 12, NVIDIA around November 18, Broadcom in December, and the November 2026 China rare-earth licensing decision. The window where they are naked short — January 2027 to January 2028 — contains roughly four earnings cycles nobody has scheduled yet. They own the events you can put in a calendar and are short the ones you cannot.

⚠️ Two honest caveats. We cannot prove this is one desk — three near-identical packages in correlated names inside eighteen minutes is a strong inference, and the tape carries no account identity. And the three tickets overlap by design: Micron and AMD together are roughly 10–11% of SMH, so this is not three independent positions.

And the timing deserves saying out loud: every semiconductor name on this board topped in late June and is 13% to 24% below that high. MU peaked June 25, SMH and DRAM June 22, AMD June 30. This is a volatility sale placed after a two-and-a-half-month drawdown, not at the top.

IEF: the trade that looked like a bond bet and isn't one

The reported flow showed two call legs on the 7-10 year Treasury ETF and read as a directional rates position. The tape shows four legs — both expiries carry a matching put that the feed left out entirely.

That changes everything. A call and a put at the same strike and expiry combine into a synthetic forward. Two of them, at different expiries, make a jelly roll — a financing trade with essentially no exposure to where bond prices go.

The arithmetic is vol-independent, so you can check it yourself. April-2027: $90 + $3.45 − $0.77 = $92.68. January-2028: $88 + $6.34 − $0.98 = $93.36. Spot is $92.19. This desk is trading the 68 cents between those two forwards — what it costs to carry the fund from April 2027 to January 2028 against the income it pays out over the same window.

Two details make it more interesting than it sounds. IEF's 30-day yield is 4.58% while its trailing twelve-month distribution yield is only 4.01% — a 57 basis-point gap, because the portfolio keeps rolling into higher coupons. Any forward has to forecast a rising payout. And the January 21, 2028 expiry lands four days before the January 2028 Fed meeting, so the position's policy exposure stops cleanly after the December 2027 decision. That looks deliberate.

⚠️ What we cannot reconcile, and will not pretend to: a single 100,000-share block printed a minute later. A clean jelly roll needs almost no stock, and one synthetic leg alone would need a million shares. 100,000 matches neither. We report it and leave it unexplained.

MOD: the strike is not the number that matters

Someone bought 750 September-2027 $260 puts and sold 750 $260 calls — a synthetic short of ≈75,000 shares — on two lines that held 1 and 0 contracts and had not moved in 26 sessions.

The instinct is to read "$260 strike, stock at $196" as an aggressive bearish call. That reading is wrong by $55.60. Because the put cost $92.30 and the call brought in $36.70, the effective short is struck at $260 − $55.60 = $204.40 — about 4% above the current price across 374 days, or ≈3.9% a year. Modine pays no dividend, so that entire premium is financing cost. The package priced at the fair forward: no volatility edge, no skew edge. It makes money below $204.40, not below $260.

⭐ And there is a clean, non-directional reason to build it in options rather than stock. Modine is roughly three months from spinning off its Performance Technologies business to shareholders pro rata and merging it with Gentherm. A pro-rata distribution is the case where the options clearing house adjusts the contracts. Someone short the actual stock across the record date would owe the distribution — ≈$15 a share — to whoever lent them the stock. Options sidestep that. The September-2027 expiry also clears the deal's outside date and its extension.

⭐ A test you can check: any company insider putting on a short like this must file a Form 4 by September 10. As of today, none exists.

GFL: a live take-private, and a structure that argues against it

On July 29 GFL disclosed in a filing that it had received unsolicited preliminary expressions of interest to take the company private and formed a special committee. That is a primary-source fact, not a rumour. Today, 12,500 December $45 calls and 12,500 December $50 calls printed together.

The obvious read is a takeover bet. The structure argues against it. Whoever built this gave away everything above $50 — and a bid for a $43 stock would plausibly clear $55 to $60, where the freshest analyst targets sit. Capping at $50 looks like a bet on a grind higher into the November guidance raise, with a buyout as a free option — not a bet on a deal landing before December 18. There has been no update in the six weeks since. The process is live but dark.

⚠️ The reported side says BUY on both legs and that cannot be right at face value. Equal size at two strikes in one expiry, printed as a single package, is the shape of a spread. As a $45/$50 spread it costs $2.06M and can be worth $6.25M. If both legs really were bought, it cost $4.89M. We publish both.

GOOGL: the feed showed you half a trade

The alert read as $22.3 million of November $370 calls bought — five prints, one strike, 20,000 contracts. The tape carries a second leg the feed left out entirely: 20,000 contracts of the $440 call, split into the identical five pieces, sold at $1.91.

That single missing leg changes the arithmetic completely. It is a $370/$440 call spread costing $18.49 million, not $22.31 million — and because the upside is capped at $440, the position has a defined maximum: $140 million of value on $18.5 million paid, a maximum profit of $121.5 million, about 6.6 to 1. Break-even is $379.24, or 11.7% above the current price — and worth noting, that break-even sits below the 52-week high of $408.61 set in May. It is the cleanest directional bullish position on today's board, carrying +514,600 shares of delta.

The strikes are not a wild guess — they bracket the analyst consensus. Of seventeen dated price targets we could enumerate, sixteen sit above $370 and five sit at or above $440; consensus of $420 to $428 lands neatly between the two strikes, with no sell ratings at all.

⚠️ But two things argue the other way, and they deserve equal space. First, there is no confirmed dated catalyst inside this window except earnings — Cloud Next has already happened, and the whole thesis rests on one print estimated for October 27 or 28 that Alphabet has not itself confirmed. Second, the buyback is gone. Alphabet repurchased no stock at all in the first half of 2026, issuing roughly $49.6 billion of convertible preferred and a $40 billion at-the-market program instead. Shares outstanding have fallen just 0.56% in a year. A bet on a per-share move is harder when the company has stopped shrinking the share count and has contracted future dilution.

And the most recent evidence is not encouraging. The ad-tech remedies ruling landed four days before this trade and went Alphabet's way — no forced sale of AdX, behavioural remedies only. The stock fell anyway, −1.17% on the day. Good news has not been moving it.

MSTR: rolling a winner that stopped winning

At 13:56 a floor block rolled 11,550 September $100 calls into 11,550 November $135 calls. The September leg is at parity — $37.19 of intrinsic value against a $37.80 price, sixty-one cents of time value with ten days to run — sitting on a line that has held ≈36,100 contracts flat for 26 sessions. That is what a stale, deep-in-the-money long looks like just before it gets rolled.

Read as a roll it collects ≈$19.87 million, and it cuts exposure by 38% — from ≈1,109,840 shares of equivalent delta down to ≈687,918. Like today's Amazon roll, it buys time and gives up directional weight. (The reported side says buy on both legs, which cannot be right for a roll. If both really were bought it cost $67.45M. We publish both.)

⭐ The context most people will miss: MSTR is down 62% from its high, while bitcoin is down 37% from its own. Bitcoin peaked at $126,296 last October and trades near $79,092 today — off 29% year over year. Strategy peaked the same week at $365.21 and trades at $137.39. That gap is leverage running in reverse, and it is the single most useful thing a retail reader can take from this name.

⭐ Why November, specifically? Because the September leg expires into nothing and the November leg spans everything. MSCI's feedback deadline is September 30, its decision is due on or before October 16, and implementation would come at the November index review — with Strategy's $23.9 billion float-adjusted weight making up 86.9% of what is at stake. Q3 earnings lands in the same window. The expiry being closed is catalyst-empty; the expiry being bought holds all three.

⚠️ And two facts that retire the popular story about this company. It sold bitcoin — 3,588 coins for $216 million between June 29 and July 5 — under a monetisation programme authorised to $1.25 billion. And in the week to September 7 it bought no bitcoin at all, instead repurchasing $176 million of its own preferred, with that authorisation just doubled to $2 billion. This is deleveraging, not a flywheel.

⚠️ On the "discount to bitcoin" figure you will see quoted: be careful. Market value against the gross value of the coins is 0.79 times — a 21% discount. But once $6.7 billion of convertible debt and ≈$14.7 billion of preferred are counted, the same company trades at roughly 1.09 to 1.14 times its net holdings. The headline discount is arithmetically true and analytically wrong.

👥 What this means for you

🎰 If you trade for the big score. DRAM is the shape you like and the one that most deserves a warning. The reported side profits if the memory ETF is below $46.48 or above $133.52 in January 2028, and its worst case is −$26.1 million at exactly $90 — a level only ≈11% above the fund's own record high. It is a barbell: it pays on a bust or a melt-up and loses most in a boring grind. Note also what today is: both of those lines jumped on the same day, July 31, in the same 2:1 ratio. Today is that trade again at a quarter of the size. And on a negotiated cross the side is not provable — reversed, the same package has a maximum gain of $26.1M.

📈 If you swing trade. Levels worth writing down: GOOGL's spread needs $379.24 to break even — below its own 52-week high of $408.61. AMZN $245 and $255 both sit on real gamma walls, which is a large part of why the roll was struck where it was. SMH's $600 is the strongest resistance shelf on its board. And the dated calendar in front of everything: Micron September 30, the Fed September 15–16, CPI September 11, AMD ≈November 3, NVIDIA ≈November 18. Two names have chains too thin to produce gamma levels at all — MOD and GFL returned none, and we would rather say so than invent them.

💰 If you sell premium. Today is a masterclass and a caution in one. Three desks' worth of premium — $120 million — was collected by selling volatility a year and a half out, with the direction deliberately removed. That is the professional version of what you do. Notice the two things that make it different from selling a naked strangle: the ratio was chosen so the position does not care where the stock goes, and they bought the nearer expiry to own the events they can see coming. Notice also the risk: short ≈$2.08 million per volatility point means a five-point move in long-dated chip vol costs ≈$10.4 million. Semiconductor volatility does move five points.

🌱 If you are new. The lesson today is that the biggest number in a flow alert is usually the least informative one — and sometimes it is not even the whole trade. Three times today the feed showed fewer legs than actually printed, and each time the missing leg changed the answer: Micron's package was a leg short, IEF's looked like a bond bet until two puts turned up, and Alphabet's "$22 million of calls" was really an $18.5 million spread with the top sold off. "Someone sold $92 million of premium in chips" sounds like a giant bearish bet; measured properly it carries almost no directional exposure at all. "Someone bought $3.5 million of Treasury calls" sounds like a bet on bonds; with the two missing legs restored it is a financing trade. "Someone is short Modine from $260" is wrong by $55.60. In every one of those cases the correction came from the same three habits: count all the legs, work out the net delta, and check what the strike actually costs after premium. None of that requires a terminal — just the patience to not stop at the headline.

🧯 Risk control and patience

• All twenty-nine legs were negotiated blocks — floor trades, crosses and one multi-leg auction. There was not one lit print on the board today. Nobody swept anything; every one of these had a known counterparty who agreed the price in advance. The buy/sell labels are allocation conventions, not observations of urgency.
• Ten legs get no open/close call from us at all, because they printed at or below the open interest already standing. Last week we were wrong twice in one session by naming a structure before that snapshot existed — a calendar that was a roll, and a roll that was a calendar. The tape prints them identically.
• Seven of today's ten structures are not directional bets on price. Three are volatility positions, one is a financing trade, one is a spin-off mechanic, and two are rolls that reduce exposure. Only GOOGL's spread is a clean, sized bet on a stock going up. Headline premium measures the size of a structure, not the strength of anyone's opinion.
• Prices in this issue are intraday marks taken around 14:35 ET, not closing prices — the session was still open when we wrote. Two names had chains too thin to produce gamma levels, and one fund is five months old. Where the data does not support a number, we have left the number out.

These positions run into the tens of millions and are hedged, financed and modelled in ways a retail account cannot replicate. Yours does not need to be. The most valuable habit in this business is being willing to do nothing on a day when you do not fully understand what you are looking at — and being honest with yourself about which days those are.

⏳ Tomorrow morning, ≈06:30 ET

Leg	Prior OI	Size	Lead branch	What would refute it
AMD Jan-28 $520P	153	2,500	open (85–100%)   — widest margin today	a fall
AMD Jan-28 $520C	713	1,500	open (40–100%)	a fall
AMD Jan-27 $520P	374	1,000	open (40–100%)	a fall
MU Jan-28 $1200P	457	1,250	open (50–100%)	a fall
SMH Jan-28 $600P / $600C	456 / 519	2,500 / 1,500	open (50–100%)	a fall
MOD Sep-27 $260P / $260C	1 / 0	750 / 750	open (95–100%)	anything but ≈+750
GFL Dec-18 $45C / $50C	283 / 1,925	12,500	open both (60–100%)	a fall on either
IEF — all four legs	98 / 64 / 35 / 389	10,000 each	open (80–100%)	a fall
AMZN Nov-20 $255C	2,915	9,010	open (50–100%)	a fall
GOOGL Nov-20 $370C / $440C	2,288 / 686	20,000 each	open (70–100%)	a fall on either
MSTR Nov-20 $135C	2,428	11,550	open (60–100%)	a fall
AMZN Sep-18 $245C	14,297	9,010	CLOSE — first call to hold us to	the line RISING
MSTR Sep-18 $100C	36,119	11,550	CLOSE — second call to hold us to	the line RISING
MU Jan-28 $1200C · MU Jan-27 both · SMH Jan-27 both · AMD Jan-27 $520C · DRAM both legs	all ≤ OI	—	NO CALL	—

Two roll-closes are the calls to hold us to: AMZN's September $245 line and MSTR's September $100 line. In both cases we are saying a line that printed below its own standing open interest will fall — on the strength of a dormant line, a near-parity price and clean roll geometry. If either rises, that roll reading is wrong and we will say so here tomorrow.

This newsletter describes trades that appeared on the public options tape. It is not investment advice and nothing here is a recommendation to buy or sell. Options can lose their entire value. The tape cannot tell us who traded, why, or what else they hold — and we say so every time it matters.