AInvest Option Flow Digest — 2026-09-03 · $328M Crossed the Tape.

Thirteen names, thirty-four legs, $328.66M of premium — and the single largest package on the board contains $532,460 of actual optionality.

PDD printed fifteen deep-in-the-money puts for $69.6 million, every one at or within pennies of parity. Strip out the intrinsic value — the part that is just arithmetic, not opinion — and $532,460 is left. The headline overstates the trade 131 times. Two legs actually printed below intrinsic value.

That is the theme of the whole session: today's big numbers are not what they appear to be. The open-interest column was wrong on three names, one trade's direction is genuinely unknowable, and the largest premium figure on the board is 99.5% arithmetic.

The reported open interest on BABA's straddle came through as 2.4K on the call and 630 on the put. The tape says 30,383 and 33,852 — understated 12.7× and 53.7×. On the reported figures that trade looked like a clean new position; on the real ones both legs are smaller than the open interest already sitting there and prove nothing at all. OKLO ran the other way: reported at 11K, actually 2,932, which turns a marginal-looking trade into an unambiguous new position.

And NKE made it three: reported at 26K, actually 126,956 — understated 4.9×. On the reported figure an 88,000-lot put buy looked like a clear new position; on the real one it is smaller than what was already there.

That is why we re-derive every open-interest figure from the tape rather than trusting the column. Today it changed the reading on three of eleven names.

⚠️ The feed also dropped legs on three more. SPCX showed one deep-in-the-money put; the tape has a second leg sold against it, which turns an open-ended $17.9M position into a $15.8M spread floored at $110. RARE showed 4,000 contracts when 5,000 printed — 25% larger than reported. And NKE's cross carried a second leg, 72,000 calls, that never appeared.

⏳ And for the fourth session running, not one leg on this board was a lit print. Every trade was negotiated away from the public order book. On those, the buy/sell label is a bookkeeping convention, not an observation.

🔁 OI Review — Wednesday's Predictions, Scored

The September 3 pre-market open-interest snapshot is in, and it settles all twenty legs from Wednesday's board. Fifteen landed on the branch we published. Five did not — every one of them QQQ, including the single largest package of the session.

🔄 The inversions: QQQ's $73.5M ladder was the right thesis with four wrong mechanics

1. The December $700 put opened — it did not close. We published a HIGH-confidence archive-based closing read, and alongside it the exact test that would refute it: "if OI on this line RISES, our closing read is wrong." It rose +4,199, to 51,608 — clear of the entire prior 11-session band of 46,539–48,942. The verdict is a partial open, 24.7% capture on a 17,000-lot print. So the biggest package of the day added December downside rather than retiring it. Corrected QQQ analysis

2. The September-11 $705 put closed — it did not open. Open interest fell −9,489, to 24,782, against a 30,000-lot print that was most of the line's session volume. That buy retired an existing short put. The honest size of the September-11 hedge is 30,000 new long $695 puts, not the 60,000 we headlined.

3. The September-30 $705 put opened as a fresh short. Open interest rose +5,917, to 27,413 — we leaned close. Paired with the October-30 $710 put that opened at 99.5% capture in the same 10:45 cross, that makes the package a put calendar spread put on, not a roll.

4 and 5. There was no call buy-back. We read the 12:33 September-30 $720/$725 call package as buying back 7,500 of the 9,500 calls sold that morning. Under that reading those two prints could have created at most ≈2,000 contracts of open interest per strike. Observed: +18,367 and +18,517 — roughly 89% of each line's full session volume. The morning's short calls are still on in full, and a separate long-call position opened on top of them.

The standing lesson, and we have now paid for it twice this month: an archive-matched prior position tells you what a desk once did, not what today's print does to it. Only the next-morning open-interest snapshot settles open versus close. The same boundary applies to price geometry — the side of the 12:33 reprint was correctly derived from arithmetic; open-versus-close never can be.

✅ Confirmations

• GLD's roll is confirmed on all four legs. Both September-11 legs fell (−37,880 and −38,030) and both September-18 legs rose (+91,675 and +108,581). This was our most falsifiable call of the session and it held. One honest amendment: closing capture came in at 57.6%/57.8%, just under our predicted 60–100% band, so ≈29,442 of the old $430 calls are still outstanding into CPI morning. GLD
• PANW's roll-or-double-short question resolves to ROLL. The dormant December-2026 $240 line fell 2,047 (68.2% capture) while the June-2027 $260 line rose 2,856 (95.2%). The "$56M of brand-new short calls" alternative is retired. The risk reversal opened on both legs, and the September-2027 $240 line went 0 → 1,102 exactly as the arithmetic required. PANW
• ADBE opened at 98.5% capture — 112 → 2,255 on a 2,175-lot print. The 2.3-year delta-hedged put sale is a real, fresh position. ADBE
• PCG opened at 75.4% capture — 20,639 → 58,356 on a 50,000-lot print, landing mid-range exactly as we said to expect rather than at a clean 100%. PCG
• QQQ's other two put legs held: September-11 $695 opened at 108.1% capture and October-30 $710 at 99.5%.

What changed, and the standing lesson

Wednesday's board published as $184.03M net across five tickers. Nothing about the money moved — every side and every price was right, so the net figure is unchanged. What moved is the shape of the largest name on the board: QQQ's protection ladder now reads as downside bought at three expiries (September 11, October 30, December 18) and sold at the fourth (September 30), with the December leg confirmed as an addition rather than an exit.

The score to remember: on the three QQQ legs where this newsletter explicitly refused to pick a branch, two went the way we would have guessed wrong. Restraint worked. Where the article did pick — on an archive override rated HIGH — it was wrong. A big headline premium tells you nothing about open versus close, and neither does a confident archive match. Only the next-morning OI snapshot does.

📋 The Board at a Glance

Ticker	Premium (net)	Expiry tag	The trade	What it means	Catalyst
IREN	$65.17M credit	LEAP   — Sep 2028	A   short straddle at $120 , two years out	Short volatility.   Profits only if IREN stays between ≈$28.85 and ≈$211.15	AI/HPC pivot;   dilution is the live risk
QQQ	$53.77M credit	LEAP   — Jun 2027	Sold 9,500 in-the-money $740 puts	Bullish-to-neutral   — the opposite of yesterday's flow	Jobs Sep 4, CPI Sep 11, Fed Sep 15–16
MSFT	$51.13M	Monthly   — Nov 20	5,000 deep-in-the-money $420 calls	Stock substitution   — 0.90 delta, only $4.46M is time value	A projected Nov 19 ex-dividend, one day before expiry
SPCX	$15.78M	LEAP   — Sep 2027	A   $245/$110 put spread   — synthetic short, floored	Bearish, bounded.   Effective entry ≈$152.20	Expires   after every remaining lock-up , including Musk's
OKLO	$12.27M credit	Monthly   — Oct 16	50,500 calls sold   at $45	Short 1.99M shares of delta   — the biggest real risk today	16–17% short float; no earnings before expiry
SPY	$12.30M	Quarterly   — Sep 18	37,500 calls,   97.3% delta-hedged	Not directional.   A bet on movement and financing	Jobs, CPI and the Fed all inside;   triple-witching
BABA	$10.65M	Quarterly   — Dec 18	A   $100 straddle, 99.2% delta-hedged	Pure volatility   — a bet it moves, not which way	November earnings (undated); Trump–Xi summit
PCG	$9.07M	Monthly   — Oct/Nov	490,000 contracts   at a $20 strike	Lottery tickets   at 12 and 25 cents	Third session of unusual flow since SB 492 died
NN	$1.71M	Monthly   — Nov 20	7,425 $18 calls,   hedge marker but no hedge found	Directional, probably unhedged	FCC 900 MHz petition —   no dated milestone
NKE	$17.38M	Quarterly   — Sep 18	87,995 $40 puts + 72,000 $45 calls	Short ≈$231M of delta   — but open vs close unprovable	Earnings Oct 1 — 13 days AFTER expiry
PDD	$54.01M   (of $69.61M gross)	Quarterly · LEAP   — 4 expiries	15 deep-in-the-money puts at parity	⚠️   Mechanical, not directional — $532,460 of time value in $69.6M	No dated catalyst inside 134 days
GLD	$1.40M	1-day   — Sep 4	A 175,000-lot $395/$392 vertical   expiring tomorrow	⚠️   Direction unknowable — a 36.5:1 lottery ticket, or its exact opposite	August payrolls, 08:30 ET tomorrow
RARE	$1.40M	Quarterly   — Jan 2027	5,000 at-the-money calls, at the ask	Directional, event-driven	UX111 PDUFA confirmed Sep 19 — 16 days out
	$306.02M net		$328.66M gross

Net is the debit paid or credit collected per package. Gross double-counts spreads, and on the deep-in-the-money trades it is mostly intrinsic value rather than premium.

🔍 What's actually interesting today

RARE: someone bought the crash, sixteen days before a decision

Ultragenyx filed an 8-K at 16:04 on September 2 disclosing that apazunersen failed its Phase 3 primary endpoint in Angelman syndrome. The stock went $26.53 to $14.94 — down 43.7% in one session.

The next day, 5,000 at-the-money January calls printed at the ask. And there is a dated reason: UX111 has a confirmed FDA decision date of September 19 — sixteen days after the trade.

So this is a $1.4M position on a $1.46B company, bought into a different programme's regulatory decision the week after an unrelated failure gutted the stock. It is the clearest event-driven trade on the board — and also the one where being wrong costs 100% of the premium.

PDD: $69.6 million of premium, $532,460 of opinion

Fifteen separate blocks printed in two bursts twenty-three seconds apart, across four expiries. Every leg is a deep-in-the-money put — strikes from $89 to $135 against an $81.94 stock.

At parity, a deep-in-the-money put is not really an option. It behaves like short stock. A $130 put on an $81.94 share is worth at least $48.06 no matter what happens; that part is arithmetic. Only what you pay above intrinsic value is an opinion. Across all fifteen legs, that came to $532,460 — 0.49% of the ≈$109 million of exposure involved. Two legs printed below intrinsic value, which is a financing signature rather than a view.

What makes this one genuinely interesting is what we could NOT find. We checked every mechanical explanation for taking synthetic short exposure through parity puts instead of shorting the stock:

• No dividend, and no plan for one.
• No Hong Kong or secondary listing — so the ADR-versus-local-share arbitrage that produces exactly this pattern in other Chinese names is structurally unavailable here.
• No index event, spin-off, buyback or corporate action. A Form 144 filed the same day covers 1,705 shares worth $140,236 — three orders of magnitude too small to matter.
• Short interest is 3.49% with 96% of the float unshorted, arguing against a borrow squeeze — though the actual borrow cost is the one thing we could not source, and it remains the best open question.

And there is no confirmed dated catalyst anywhere inside the 134-day horizon. Two of the four expiries contain no expected earnings at all; the other two share the same single expected print.

So we can tell you what this is not — not directional, not an earnings play, not an arbitrage we can identify, probably not a borrow squeeze — and we cannot tell you what it is. That is where the evidence stops, and we would rather say so than invent a story.

GLD: the same two prints are either a lottery ticket or the opposite of one

A 175,000-lot $395/$392 put vertical printed at 14:22, expiring TOMORROW. Net cash either way: $1,400,000. Maximum value of the spread: $52,500,000.

And we genuinely cannot tell you which side the initiator is on, because it printed as a negotiated floor block. The two readings are opposite trades of identical size:

	If it is a debit spread	If it is a credit spread
Cash	pays $1.4M	collects $1.4M
Best case	makes $51.1M	keeps $1.4M
Worst case	loses $1.4M	loses $51.1M
Needs	GLD   at or below $394.92   tomorrow	GLD   above $394.92   tomorrow

What IS proven is that the position is new — 175,000 contracts against prior open interest of 1,462 and 2,694 settles that beyond doubt. Open versus close is proven; direction is not — and tomorrow's snapshot will not help, because open interest is side-blind.

⭐ The expiry date is the whole point: the August jobs report releases at 08:30 ET tomorrow, and these options expire that day. Nothing else is on the calendar — no Fed speakers, pre-meeting blackout. And hours before this block printed, Fed governor Waller said that "if the incoming data for August show this improvement has been fleeting, then it may be appropriate to raise the policy rate" at the September meeting. Gold had just rallied ≈2% on a soft private-payrolls estimate.

One myth worth killing: a 36.5-to-1 payoff is not free money. Eight cents on a $3-wide spread is 2.67% of maximum value, and against ≈44% implied volatility the $392 strike sits ≈2.04 standard deviations away — about a 2.1% chance. The price and the volatility surface agree with each other. Whoever is long is betting payrolls-day tails are fatter than the maths says; whoever is short is betting they are not.

NKE: 88,000 puts that expire before the thing everyone is waiting for

The largest position of the afternoon was 87,995 September $40 puts, plus 72,000 $45 calls at three cents that amount to 1.3% of the package. Net, it is short ≈6.0 million shares of delta — roughly $231 million of notional on a stock trading at $38.52.

Two facts make it more interesting than "someone is bearish on Nike."

First, Nike reports on October 1 — thirteen days AFTER this option expires. Whoever holds it is positioned for the fifteen days before the print, not for the print itself. And the window is almost entirely empty: the only scheduled corporate events before September 18 fall on September 4, when the outgoing CFO's last day coincides with the Chief Accounting Officer's resignation.

Second, the decline that got Nike here was a grind, not a gap. The stock is down ≈39.8% in 2026 and sits 53 cents above its 52-week low — but the largest single-day move since late June was +4.90%, and August fell 10.75% without a single crash day. A short-dated put on a stock that bleeds rather than gaps is fighting time decay with no scheduled event to rescue it.

And we cannot tell you whether this opened or closed. Both legs printed below open interest on a triple-witching expiry where each strike already carried 115,000+ contracts. A roll or an unwind is at least as plausible as a fresh $231M bet.

OKLO is the biggest genuine risk, and the numbers say so

50,500 calls sold at $45 collects $12.27M. Behind that: short 1,985,155 shares of delta — roughly $80M of notional — on a stock with 73.6% implied volatility.

The dangerous part is that the risk has no calendar. Next earnings is estimated November 10, after the October expiry, and no dated regulatory decision could be sourced inside the window. But 16–17% of the float is short at 2.44–2.71 days to cover, the lowest analyst target on the street is $51 — above the $47.43 break-even — and Oklo produced eight unscheduled announcements between June 11 and August 7. A heavily shorted, high-volatility name with a habit of unscheduled news and nothing on the calendar to hide behind is exactly where a short call gets hurt.

IREN collected $65M and only $8.6M of it is actually premium

A short straddle at $120, expiring September 2028. With the stock at $40.91, the $120 put is deep in the money — $79.09 of that $82.08 is intrinsic value. So of $65.17M collected, ≈$8.62M is time value and the rest is an obligation already worth $56.5M.

The position profits between ≈$28.85 and ≈$211.15 in two years' time. And here is the number that matters: the 52-week low is $25.31 — already below the lower break-even. The option market's own one-year range reaches down to $7.64. Essentially all the loss risk is on the downside, and the stock has already been there this year.

(One correction worth making: this is not a bitcoin trade. IREN holds zero bitcoin and targets a full exit by year-end; AI Cloud revenue overtook mining last quarter. The live risk is dilution — share count up 84% in eight quarters, with a $6B at-the-market programme running.)

Three trades with a hedge marker. Two had a hedge.

SPY paired 37,500 calls with 1,162,500 shares — 97.3% of the package delta. BABA paired its straddle with 230,000 shares — 99.2%. Both are therefore not directional bets at all; the stock leg cancels the direction and what is left is a bet on movement and financing.

NN carried the same family of marker — and we could not find the stock. A full hedge would need ≈357,291 shares. We scanned the entire session: 8,898 prints, largest 14,081. The hedge may have been worked in pieces, or reported elsewhere, or simply be small. We do not know, so we are not claiming it. Same marker, opposite result — which is the whole reason we check instead of assuming.

QQQ sold the downside it bought the day before

Yesterday this fund saw downside bought at four expiries, and this morning's snapshot confirmed the December leg opened. Today someone sold $53.77M of long-dated downside.

Either two desks disagree, or one book is financing its short-dated protection by selling longer-dated protection. There was no scheduled macro release on either day. We cannot tell which from the tape, and we are not going to pretend otherwise.

👥 What this means for you

🎰 If you trade for the big score. RARE is the shape you like — at-the-money calls, a dated catalyst sixteen days out, defined downside. But understand what "defined downside" means: if the September 19 decision disappoints, the premium is gone, not reduced. And note the professionals' own restraint elsewhere: SPCX's bearish position is floored at $110 rather than left open-ended, giving up the tail to cut the cost.

📈 If you swing trade. The levels that matter near-term: SPY $770 support / $775 resistance, with jobs, CPI and the Fed all inside the September 18 expiry; QQQ $715 support / $720 resistance; MSFT sitting right on its heaviest shelf at $510. For the small caps, note how thin the gamma maps are — RARE has only eight strikes carrying gamma at all, so do not read structure into noise.

💰 If you sell premium. Today gives you the case for and against in one session. Against: OKLO — 50,500 calls sold into a 17% short float with no scheduled events and the lowest street target above the break-even is the trade that ends badly on a headline nobody scheduled. Also against: IREN, where 87% of the "premium" is not premium at all, just an obligation that was already worth $56.5M. Both teach the same lesson — before you count a credit as income, subtract the intrinsic value and ask what has to go right.

🌱 If you are new. The single most transferable lesson today is subtract the intrinsic value before you get impressed by a headline. A $130 put on an $81.94 stock is worth at least $48.06 as pure arithmetic — that is not anyone's opinion, it is just the gap between strike and price. Only what is paid above that is a view. PDD's $69.6 million print contained $532,460 of view. Learn that subtraction and a lot of alarming-looking flow stops being alarming.

The other habit worth building is checking open interest — the number of contracts already outstanding at a strike. If a trade is smaller than the open interest already there, it might be opening a new position or closing an old one, and you genuinely cannot tell which from today's data. Today the reported open interest was wrong on two names, once by 53×. When a number does that much work in an argument, look it up yourself.

🧯 Risk control and patience

• Thirty-four of thirty-four legs were negotiated blocks, four sessions running. No aggressor exists anywhere on this board.
• On GLD that matters more than usual: the same two prints are a 36.5:1 lottery ticket or a tail being sold for income, depending on who initiated — and the tape does not say. We publish both readings and pick neither.
• Seven of seventeen legs get no open/close call from us at all. Yesterday we published a high-confidence closing read on a QQQ leg and it opened instead. Where the tape cannot resolve it, "we cannot tell yet" is the answer we print.
• Two headline premiums today are mostly intrinsic value — MSFT's $51.1M contains $4.46M of time value, IREN's $65.2M contains $8.6M. A big number is not a big opinion.
• Three trades are hedged packages. Their premium describes the size of a structure, not the size of anyone's conviction.

Position sizes here run to nine figures. Yours does not have to. The most valuable habit in this business is being willing to do nothing on a day you do not fully understand what you are looking at.

⏳ Tomorrow morning, ≈06:30 ET

Leg	Prior OI	Size	Lead branch	What would refute it
SPY Sep-18 $779C	3,298	37,500	open	a fall
QQQ Jun-2027 $740P	3,970	9,500	open	a fall
OKLO Oct-16 $45C	2,932	50,500	open	a fall
PCG Nov-20 $20C	0	245,000	open	anything but ≈+245,000
SPCX Sep-2027 $245P	1,000	1,700	open	a fall
IREN Sep-2028 $120P / $120C	2,744 / 2,775	7,150	open	a fall
NN Nov-20 $18C	211	7,425	open	a fall
RARE Jan-2027 $15C	1,343	5,000	open	a fall
GLD Sep-04 $395P / $392P	1,462 / 2,694	175,000	open (proven)	⚠️ nothing —   direction is not testable by open interest
PDD — all 15 parity put legs	83–1,605	240–3,300	open (proven on every leg)	⚠️ nothing —   direction is not testable by open interest
MSFT $420C · BABA $100C/$100P · PCG Oct $20C · SPCX $110P · NKE $40P/$45C	all ≤ OI	—	NO CALL	—

Those last seven get no branch on purpose. Yesterday, on the three legs where this newsletter refused to pick, restraint was right; on the one where a confident archive match pushed us into a call, we were wrong. A big headline premium tells you nothing about open versus close, and neither does a confident-looking match to a past position.

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.