AInvest Option Flow Digest — 2026-09-02 ·

Five names, twenty legs, $259.73M of premium — and the most useful thing on the board is a number most people never look at: the size of the paired share block.

Three of today's five names printed as stock-and-options crosses, meaning the option was negotiated together with a block of shares. In each case the share block matches the option's delta almost exactly:

Name	The option	Package delta	Share block	Match
PCG	50,000 January $17 calls	1,205,000 sh	1,200,000	99.6%
PANW	1,100 September-2027 $240 calls	88,957 sh	91,300	102.6%
ADBE	2,175 December-2028 $250 puts	56,615 sh	58,725	103.7%

When the share leg cancels the option's delta, the trade stops being a bet on direction. What is left is a bet on movement and on financing. So PG&E's "+28% call buy" is not a forecast that the stock rises 28%, and Adobe's "$10M put sale" is not a forecast that Adobe holds $250. If you screen flow for direction, all three of these would have fooled you.

⚠️ And the feed was wrong on three of the five names. GLD showed three legs when the tape has four — and the missing one turns a lone call sale into a two-spread roll. PANW showed a lone put sale when the tape shows a paired call in the same auction, making it a risk reversal. QQQ was missing an entire package.

⏳ And nothing is settled until tomorrow. Twenty of twenty legs were negotiated blocks — no lit prints at all, the third session running. On those, the buy/sell label is a bookkeeping convention, not an observation. Eight legs printed at or below existing open interest. Come back at ≈06:30 ET — we score every published prediction in public, including this morning's, where one of ours inverted.

🔁 OI Review — Tuesday's Predictions, Scored

The 06:30 ET open-interest snapshot is in, and Tuesday's board resolved 6 legs open, 6 closed, 1 close-plus-transfer and 3 as pure transfers. We publish a capture rate on every leg before the answer exists, then come back and score it in public. Fourteen of sixteen legs landed on the branch we led with — but the two that missed are both on the same name, and they carry 76.6% of its headline premium.

🔄 The inversion: SPCX's $100.87M put program created no position at all

We described Tuesday's SPCX blocks as a synthetic short of ≈980,000 shares being established ahead of the September 9 lock-up tranche — a hedge going on, dressed up in deep-in-the-money puts. Open interest says nothing went on.

The two legs that were bought moved by −245 contracts between them. The $205 put went 26,917 → 26,773 (Δ −144) against a 7,900-lot print — a 1.8% capture. The $200 put went 13,812 → 13,711 (Δ −101) against 5,000 — 2.0%. The $230 put did not move by a single contract (1,633 → 1,633). A print that moves open interest by a rounding error is existing holders passing positions to new holders, not a hedge being put on. The lock-up-hedge reading is refuted.

The one leg that moved meaningfully moved the other way. The sold $210 put fell 1,515 → 342 (Δ −1,173, a 75.7% close) — short-put exposure being retired. So the only net risk change on the whole SPCX board is a small reduction.

This was not a bust. September 1 end-of-day volume printed in full on all four lines — 7,900 / 5,003 / 1,650 / 1,550. And neighbouring September 18 put strikes were flat ($190P −12, $195P −4, $215P 0, $220P 0, $225P 0, $235P 0), so the tiny declines are strike-specific and real, not a chain-wide sweep. What survives untouched is the arithmetic that led the piece: $100.87 million of gross premium containing ≈$279,000 of time value. It just means something sharper now — the "bet" was not merely $279,000 of opinion, the position underneath it was not even new.

✅ The most falsifiable call on the board held: MSFT

MSFT's $35.97M "credit" was published with its own refutation condition stated in advance: four December-2027 lines each holding ≈8,000 contracts should FALL, and any one of them rising overturns the unwind read. None rose.

• $595C 8,172 → 2,761 (Δ −5,411, 67.6%) · $705C 8,093 → 2,683 (Δ −5,410, 67.6%)
• $610C 7,999 → 3,205 (Δ −4,794, 60.9%) · $720C 8,007 → 3,206 (Δ −4,801, 61.0%)

Look at the pairing. The two legs that printed at 10:12:45 fell one contract apart; the two that printed at 10:13:05 fell seven apart. Four independent lines do not move in matched pairs by accident — that is two packages unwound as units. The new-bear-call-spread reading is refuted and the credit is confirmed as recycled capital. This is also the direct payoff from the fix adopted after the August 28 MSFT package inverted our read: pull three weeks of open-interest history, not one session.

The honest correction: the closes are partial. We sketched a near-total evacuation at ≈98% capture; the reality is 61%–68%, so ≈2,700 to 3,200 contracts per strike are still outstanding. Control strikes on the same chain were flat ($600C −91, $620C −21, $700C 0, $710C 0, $730C 0).

✅ Confirmations

• DRAM — the ratio-spread unwind is real, and near-total. $45C 21,049 → 1,071 (Δ −19,978, 99.9%) — that line lost 94.9% of its entire standing open interest in one session. $100C 45,586 → 9,744 (Δ −35,842, 89.6%). Control strikes on the January 2028 chain were flat ($40C +11, $50C −48, $90C −3, $110C 0). The $14.5M credit is recycled profit, confirmed — not a new $70.5M opinion on memory.
• IWM — the largest confirmed new position on the board. $285P 17,144 → 180,182 (Δ +163,038, 98.2%) and $275P 588 → 166,710 (Δ +166,122, 100.0%). Worth noting because it nearly went the other way: the trade archive had flagged a prior 56,890-lot short at the same $285 strike and suggested this BUY was closing it. That older line had already fallen 56,715 → 17,046 a full session earlier — so Tuesday's block opened on top of what was left. The 7.5-to-1 spread is live into payrolls and CPI.
• NVDA — both legs opened, and the ratio survived the snapshot. $150P 42,144 → 135,599 (Δ +93,455, 93.5%) and $100P 46,501 → 185,341 (Δ +138,840, 92.6%). The trade was struck 1 : 1.5; the open-interest change came in at 1 : 1.486 — independent confirmation that the two legs are one package, not two coincidental prints. The shaped payoff peaking near $100 is exposure that now genuinely exists.
• BSX — both legs opened inside the published 80%–100% band. $60C 9,072 → 54,627 (Δ +45,555, 91.2%) and $75C 1,196 → 51,436 (Δ +50,240, 100.5%). The ≈8.8% shortfall on the $60 leg is transfer against holders who were already there, not a close.

What changed, and the standing lesson

Tuesday's board was published at $347.60M gross / $140.89M net. Those figures do not change — they come off the trade tape, not open interest. What changes is what the money did: ≈$77.22M of the SPCX headline moved risk between books without creating any, and MSFT's $35.97M credit is now confirmed as capital coming back rather than a new bearish structure. Six legs opened, seven closed or partly closed, three were pure transfers.

The lesson, and it is the mirror image of Monday's. Monday taught us that a low arithmetic margin means unprovable, not probably a transfer — we led with transfer on two legs that opened at 100%+. Tuesday taught the reverse: a dormant, calm open-interest line is not evidence of a coming open either. SPCX's $205 and $200 lines had been quiet for three weeks, and we leaned "open" on that basis. Wrong. When printed size is a small fraction of a large standing line — 7,900 against 26,917 — the arithmetic floor is zero and no lean is justified in either direction. Publish the three branches and stop. A big headline number is never, by itself, evidence that a position was created.

📋 The Board at a Glance

Ticker	Premium (net)	Expiry tag	The trade	What it means	Catalyst it points at
QQQ	$73.52M   (of $124.00M gross)	Weekly · Monthly · Quarterly	Downside bought at   four separate expiries   in one session	Layered protection , not one strike. Directional-bearish but bounded	Payrolls and CPI   inside the September 11 window; the Fed on September 15–16
PANW	$70.23M credit   (of $70.77M gross)	Monthly · Quarterly · LEAP	A risk reversal, a delta-hedged cross, and a $56M roll-or-double-short	Mostly hedged and structural   — only the $300 put is a live directional risk	Fell   −10.4%   on an NGS-ARR miss despite a beat-and-raise
GLD	$27.73M   (of $52.41M gross)	Weekly → Quarterly	A four-leg   roll : out of Sep-11, into a bigger, closer Sep-18 spread	Directional-bullish, and deliberately re-timed	CPI lands the morning the old expiry died; the Fed lands inside the new one
ADBE	$10.05M credit   (of $10.05M gross)	LEAP   — Dec 2028	Sold 2.3 years of downside volatility,   delta-neutral	A volatility and financing trade , not a view on price	Earnings confirmed September 10, eight days out
PCG	$2.50M   (of $2.50M gross)	Quarterly   — Jan 2027	50,000 calls   99.6% delta-hedged	Buying convexity   — pays on a big move either way, or on rising volatility	A wildfire bill died August 31; the stock fell over 20%
	$184.03M net		$259.73M gross

Net is the debit paid or credit collected per package — the money actually at risk. Gross double-counts every spread.

🔍 What's actually interesting today

GLD rolled its position over two landmines in one trade

A four-leg floor block did two things at once: closed a September 11 $430/$445 call spread and opened a September 18 $415/$435 spread — 1.68× bigger and $15 closer to the money.

Look at what sits between those two dates. August CPI releases at 08:30 ET on September 11 — the exact morning the abandoned expiry died. The old spread would have had to absorb that print at expiration, with no time value left to cushion it. And the September FOMC is September 15–16, carrying a full set of economic projections — after the old expiry, comfortably inside the new one.

So the roll escapes the CPI print and buys the Fed meeting. Break-even on the new spread is $417.65, +4.06% — and GLD's own 16-day implied move is ±5.00%. The break-even sits inside the market's cone while the $435 cap sits outside it. That is a considered position, not a lottery ticket.

(Verified backdrop: policy rate 3.50–3.75%, Kevin Warsh chairs, and the July 29 vote was 9–3 with three officials dissenting for a hike. We are not quoting hike odds — the probability data could not be sourced. The objective tell is that the 2-year yield at 3.80% sits above the 3.75% top of the target range.)

QQQ bought downside at four different expiries — and it's a rate hedge, not an AI panic

September 11, September 30, October 30, December 18. $73.5M net, layered across the curve, including a $41.3M single block of 17,000 December $700 puts — the largest single leg on the board.

The 60,000 September 11 puts are near-the-money, not tail insurance — struck at −0.5% and −2.0% from spot with nine days to run, against a 16-day implied move of only ±3.59%.

And the gamma underneath is unfriendly. Spot sits at $708.29 on a stack of heavily negative gamma: $708 (net −194.6), $700 (−170.8), $705 (−121.9). Negative gamma beneath the price tends to make a decline accelerate rather than settle. Someone bought 60,000 near-the-money puts into exactly that, on a fund already drifting down from ≈$721 a week ago.

But read the reason correctly. This week's dated evidence on AI spending runs the other way — a large beat-and-raise in the sector. These puts are a macro hedge against the September data, not a bet against artificial intelligence.

PG&E: a bill died, the stock fell 20%, and five hours later someone bought convexity

California's SB 492 died without an Assembly floor vote when the session closed August 31. PG&E filed an 8-K the same day calling it inadequate. The stock crashed more than 20%. At 08:30 ET on September 2 management announced a Strategic Review Committee and deferred $2 billion of 2027 capital spending. Five hours later this cross printed. The stock hit a fresh 52-week low of $12.59 today, and two large fires are burning uncontained in Monterey County, inside PG&E's territory.

The "+28% strike" framing is misleading: $17 sits only 11.3% below the stock's own 52-week high and 23% below the $22 analyst median. It is a recovery strike, not a moonshot.

But here is the honest counterweight, and it is why this is the weakest catalyst structure on the board: there is no confirmed dated catalyst between October 22 earnings and the January expiry. The 2027 rate-case decisions land in March and May 2027 — after this option expires. The buyer is paying for time in which an unscheduled legislative fix might appear.

Adobe sold 2.3 years of volatility eight days before earnings

$10.05M collected on December-2028 $250 puts, delta-hedged to neutral. The strike is 11% below spot with 2.3 years to run, so the entire $46.20 is time value — this is a bet on how much Adobe moves, not where it ends up.

The strike is calibrated to the bear case. The lowest Street target is $190 — sixty dollars below the strike. Break-even is $203.80, just above the 52-week low. And the 16-day implied move puts the lower band at $247.73 — essentially on the strike. The market thinks one earnings print could get there.

One thing that may matter more than the AI-disruption debate: the CEO notified the board on 2026-03-09 that he is stepping down and the search is still open, and the CFO resigned effective 2026-06-15. That explains June's wave of target cuts at least as well as AI does.

👥 What this means for you

🎰 If you trade for the big score. Today is a lesson in what isn't an opportunity. Three of five names are delta-hedged — the people on the other side deliberately removed the directional bet. Copying the option leg without the share leg gives you a completely different trade from the one that actually printed. The one genuinely directional structure is GLD's $415/$435 spread, and even that is capped. If you cannot see the whole package, you cannot copy it.

📈 If you swing trade. The levels worth writing down: QQQ's negative-gamma stack at $700–$710, which sits directly under spot and is the single most actionable thing on this page for a two-week horizon. GLD support $400, resistance $410. ADBE support $280, resistance $300 — and September 10 earnings sitting between now and any of it. PANW's chain is thin (64 strikes), so do not over-read its gamma map.

💰 If you sell premium. Adobe is the case study, in both directions. $46.20 collected on a 2.3-year put looks like rich carry — until you notice the lowest analyst target is $60 below the strike. And do not read a 0.26 delta as a 26% chance of finishing in the money; for a put, delta understates that probability, and over 2.3 years the gap is material. That is paid-for tail risk, not free money. The offset is real too — a $26.78B buyback authorisation, ≈24% of the market cap.

🌱 If you are new. Learn what delta-hedging is, because it explains three of today's five trades. An option's delta tells you how many shares it behaves like. If you buy a call that behaves like 1,205,000 shares and simultaneously trade 1,200,000 shares the other way, you have cancelled the direction on purpose. What is left is a bet on how much the stock moves, not which way. When you see a big option print, ask whether a share block printed with it — that one question changes the meaning of the trade.

🧯 Risk control and patience

• Twenty of twenty legs were negotiated blocks. No aggressor exists anywhere on this board, three sessions running. The buy/sell labels are allocation conventions.
• Eight legs printed at or below existing open interest, so today's tape cannot say whether they opened or closed a position.
• This morning our SPCX call inverted — a $100.87M program we described as a hedge going on turned out to create no position at all. The trades that read most cleanly are not always the ones that resolve cleanly.
• Three of today's names are hedged packages. Their headline premiums describe the size of a structure, not the size of anyone's opinion.

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

Three outcomes are always live on any leg: it opened in full, it closed, or it landed in the middle as a partial close or a pure transfer — and the middle is common, not exotic.

Leg	Prior OI	Size	Lead branch	What would refute it
GLD Sep-18 $415C / $435C	55,103 / 12,006	≈110,500	open	little or no rise
GLD Sep-11 $430C / $445C	67,322 / 66,968	65,817	CLOSE	either line RISING — this is our most falsifiable call
QQQ Sep-11 $695P	5,835	30,000	open	a fall
QQQ Sep-11 $705P · Sep-30 $705P · Dec-18 $700P	34,271 / 21,496 / 47,409	30,000 / 9,500 / 17,000	unresolved — we are not picking	—
QQQ Oct-30 $710P	573	9,500	open	a fall
PANW Nov-20 $300P	530	964	open	a fall
PANW Sep-2027 $240C	0	1,100	open	anything but ≈+1,100
PANW Jun-2027 $260C	235	3,000	open	a fall
PANW Dec-2026 $240C	3,539	3,000	CLOSE   (frozen 21 sessions)	a rise
PCG Jan-2027 $17C	20,639	50,000	open	a fall
ADBE Dec-2028 $250P	112	2,175	open	a fall

Note what we are NOT doing on QQQ's three size ≤ OI put legs: we are not picking a branch. Twice this week we attached a favourite to a leg the tape could not resolve, and twice it went the other way. When the honest answer is "we cannot tell yet," that is the answer we publish.

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.