AInvest Option Flow Digest — 2026-08-31

Ten names, thirty-one legs, $436.53M of premium — and the single largest fact on the board is not a bet at all. It is insurance.

Three separate collars printed on the two biggest index funds in the world today. Together they wrap $5.68 billion of exposure, and the total net cost was $26.52 million — 0.47% of the notional they protect. One of them, a $1.54 billion position in SPY, printed for a net debit of $320,128. That is two basis points.

A collar is the least glamorous trade in options and today it was the whole story. You buy a put to put a floor under what you own, and you pay for it by selling a call that gives away your upside above some level. When it is done well the two premiums nearly cancel and the cost rounds to nothing. Today's three were done at wildly different prices — 0.021%, 0.369% and 0.770% of notional — and the reason for the gap is the most useful thing a reader can take away from this session.

⚠️ Second: the feed dropped four legs today, and in three names the missing leg changes what the trade is. LPX was reported as a $1.15M outright put; the tape shows a $65/$55 put spread costing $825,000 with the downside capped. RUM was reported as a $1.66M outright call; the tape shows an $8/$11 call spread costing $1,287,500. IEF's midday print was shown as one leg on the sell side; it is actually a three-leg day and that print was a buy. And SPY's afternoon package had five legs, not four. In every case the correction made the trade smaller and more bounded than the headline.

⚠️ Third: TEX printed $6.36M and the economic content is minus $45,500. It sold deep-in-the-money calls thirteen cents below intrinsic value. There is no time value in that trade — there is negative time value. We flag it rather than counting it, and the reason it can print there safely is a calendar detail worth knowing: there is no ex-dividend date before the option expires, and without a dividend an American call is never worth exercising early.

⏳ Nothing here is proven until tomorrow morning. Thirty of today's thirty-one legs printed as negotiated blocks — crosses, auctions and floor trades arranged away from the public order book. On those, the buy/sell label is a bookkeeping convention, not an observation. Eleven legs printed at or below their existing open interest, which means we cannot tell from today's tape whether they opened a position or closed one. Come back at ≈06:30 ET on the next trading day — the open-interest snapshot is the only thing that settles it, and we score every prediction publicly.

🔁 OI Review — Friday's Predictions, Scored

The 06:30 ET snapshot is in for the 2026-08-28 board. Nine names, thirty-three legs — thirty-two resolved, one permanently unresolvable, and fourteen more thrown out by an exercise sweep we called in advance. The headline is a genuine inversion on the biggest package of the day.

🔄 The inversion: MSFT's $126M package was a CLOSE, not a bull call spread

We published two readings with opposite signs on Microsoft's four-leg, 17,000-lot auction and named next-day open interest as the only test between them. It picked the bearish one.

Leg	Baseline	Resolving	Δ	Verdict
Nov-20   $480C	21,709	11,937	−9,772	❗ partial close
Nov-20   $580C	18,648	10,246	−8,402	❗ partial close
Oct-16   $545C	2,091	18,931	+16,840	✅ open, 99.1%
Oct-16   $595C	234	17,389	+17,155	✅ open, 100.9%

The bull-call-spread reading needed both November lines to RISE by up to 17,000 each. Both fell, in near-matched size — the signature of closing a spread, not two unrelated trades. So this was a desk buying back a short November $480/$580 position that Microsoft's historic July 30 gap had run straight through, then re-writing the overwrite higher and nearer. The only new risk created on the day is ≈33,995 short October calls — delta ≈−646,000 shares, neutral-to-bearish into October 16. A reader who took "17,000-lot bull call spread" at face value had the direction backwards. The article's title and lead have been rewritten; the refuted branch is kept below them for the record.

One more thing in that name, and it cuts the other way: the separate lit $510 call buy — the only provable aggressor on the whole board — partially opened, 33,339 → 39,257 (+5,918, 63.2%). Same morning, same expiry: one participant opened ≈5,918 new long October calls while another opened ≈33,995 new short ones above them.

🔄 IGV — our published fork landed on neither branch

We wrote it as a clean test: "$103 OI rising ≈+15,000 with $97 OI falling ≈−12,500 = roll confirmed; both rising = spread confirmed." Neither happened. The $103 leg opened cleanly — 982 → 18,490 (+17,508, 116.7%) — but the $97 leg moved only 12,550 → 11,312 (−1,238), under a tenth of the size printed. So what entered the market is ≈17,508 new long November $103 puts and no net new $97 short interest. The "bounded, defined-risk" half of that spread does not exist as new market exposure.

That is the second consecutive session with the same shape. Thursday: Dec $103P +30,046 clean open, Dec $95P +36 transfer. Friday: Nov $103P +17,508 clean open, Nov $97P −1,238 transfer. ≈47,500 contracts of genuine new $103 put interest across two sessions — and the paired lower strikes are not building.

⚠️ B — the exercise sweep we called, plus a correction we owe

Barrick's fourteen deep-in-the-money legs all lost their open interest. That proves nothing on its own — and here is why we can say so rather than guess. The September $37 call had ZERO volume on Friday and still went 3,988 → 53, a 98.7% wipeout. A contract nobody traded cannot have been closed by a trade. That is exercise into the ex-dividend date, and it swept every call strike at or below $38 while the puts and everything at $39 and up stayed flat. All fourteen legs sit inside the swept zone, so none can be scored open or closed — and no future snapshot will recover it.

The correction: we split the ladder at the $0.175 dividend threshold and forecast that the eight higher-time-value legs would lead OPEN or TRANSFER. They were swept too — the $33 call fell 96%, and the January-2027 $10 call ended 461 contracts lower on a 2,600-lot print. The threshold was the wrong dividing line; how little time value remained in absolute terms was the right one.

✅ And the confirmations — including one roll that needed the same control test and passed it

GLW's roll confirmed on both halves: January $55 6,679 → 2,418 (−4,261), March $60 0 → 6,929. Corning went ex-dividend the same morning, so we ran Barrick's control test here too — every untraded deep-in-the-money January strike was flat on zero volume ($45C 227 → 227, $50C 624 → 624, $60C 922 → 922). No sweep, so the decline is a genuine close. Same test, opposite answer, and that is the point of running it.

SKHY all three collar legs opened (100.1% / 91.7% / 95.5%) — the $135 floor is live open interest, not an inference. AMZN both legs opened and over-captured (116.5% / 108.5%), so the breakeven-above-the-52-week-high problem belongs to a real carried position. VIX full open at 123.4% — the October $28 line nearly tripled, 66,815 → 195,627. ETHA both legs within 3% of forecast. SNDK's September leg opened at 113.5%, above its own arithmetic ceiling.

One honest non-resolution, disclosed in advance: SNDK's August 28 $900 call expired the same afternoon it traded. There is no Monday open-interest figure for it and there never will be — its open/close status is permanently unverifiable, exactly as we said on Friday.

What changed, and the standing lesson

Pre-resolution the board read as $190.03M of directional flow across five names. Post-resolution the largest piece of that — Microsoft's $126.47M package — carries the opposite sign, and IGV's capped structure turns out to be an uncapped long put line. A big BUY headline is not conviction, and a falling open-interest line is not automatically a close. Barrick and Corning fell on the same morning for opposite reasons, and the only thing that separated them was checking what happened at the strikes nobody traded. Do that check before calling any decline a close.

📋 The Board at a Glance

Ticker	Expiry tag	Structure	Gross	Net	Mechanism	Open/close status
SPY	Quarterly · Monthly · LEAP	Three packages: a $43.19M put sale, a five-leg protection roll, and a   $1.54B collar for $320,128	$115.81M	$33.73M	Cross · floor · auction	6 legs proven open · 2 provisional · 2 expire today
QQQ	Monthly · Quarterly	Two collars,   $4.15B   of notional; the December floor is struck   at the money	$250.78M	$26.20M	Multi-leg cross	3 legs proven open · 1 marginal
SNOW	LEAP (Jan-2027)	$340/$420 call spread for a   $12.81M credit   —   earnings in two days	$34.80M	$12.81M	Floor block	⏳   Both legs below prior OI — genuinely unresolved
NRG	Quarterly (Dec-18)	Four legs, one expiry: strikes rolled   DOWN   into a 52-week low	$9.22M	$0.38M	Floor block	2 proven open · 2 ⏳ likely close
IBIT	Monthly (Nov-20)	30,000 calls a leg at $45 and $57 —   exposure coming off, not going on	$10.74M	$10.74M	Multi-leg auction	⏳   Both legs below prior OI
IEF	Quarterly (Sep-18)	At-the-money calls   sold at the 52-week low , then a small spread bought back the other way	$2.00M	$1.04M	Auction · floor block	1 proven open (prior OI of   1 ) · 1 ⏳
XEL	Long-dated (Mar-2027)	10,000 puts,   delta-hedged to 95.8%   — the cleanest pairing on the board	$3.30M	$3.30M	Stock-and-options cross	✅ Proven open (98% margin)
LPX	Monthly (Oct-16)	$65/$55 put spread, $825,000   — capped, not open-ended (feed showed one leg)	$1.48M	$0.82M	Multi-leg cross	✅ Both proven open
RUM	Weekly (Oct-9)	$8/$11 call spread, $1,287,500   — brand-new strike lines (feed showed one leg)	$2.04M	$1.29M	Stock-and-options cross	✅ Both proven open (prior OI   0 )
TEX	Quarterly (Sep-18)	Deep-in-the-money calls sold   13 cents BELOW parity   — a financing print, not a view	$6.36M	$6.36M	Cross (one print, after a cancel)	✅ Proven open (prior OI   0 )
		TOTAL	$436.53M	$96.67M

Net is the debit paid or credit collected per package — the money actually at risk. Gross double-counts every spread. Where the two diverge by a lot, the gross number is telling you about plumbing, not conviction.

🛡️ The main event: what $26.52 million bought

	SPY Nov-30	QQQ Nov-20	QQQ Dec-31
Contracts	20,008	20,000	38,000
Notional protected	$1.536B	$1.43B	$2.71B
Floor	$727 ( −5.3% )	$680 ( −5.2% )	$715 ( at the money )
Floor stops working below	$612 (−20.2%)	—	—
Upside sold above	$806 (+5.0%)	$760 (+5.9%)	$740 (+3.6%)
Net cost	$320,128	$5.30M	$20.90M
Cost as % of notional	0.021%	0.369%	0.770%

Two things explain the 37× gap, and both are worth internalising.

1 · The deductible. The SPY floor does not start until the index has already fallen 5.3%, and it stops paying below $612 because a further put was sold underneath it. So it covers a defined 15-point band and nothing outside it. The Dec-31 QQQ floor starts immediately, at the money, with no deductible at all. Insurance with a large excess is cheap; insurance with none is not. That is not an options fact, it is a general one — it just shows up here with a number attached.

2 · The meetings. The SPY collar expires November 30 — eight days before the December 8–9 policy meeting. The Nov-20 QQQ collar expires before it too. The Dec-31 QQQ collar is the only one of the three that covers all three remaining meetings of the year. The Federal Reserve's own calendar puts them on September 15–16, October 27–28 and December 8–9. You pay for the meeting you keep.

⭐ And the macro backdrop is the opposite of the one most readers are carrying. The market is pricing a rate HIKE, not a cut — 64.4% odds of a move to 3.75–4.00% at the September meeting. The July 28–29 meeting held rates on a 9–3 vote with three officials dissenting , and Chair Kevin Warsh's August 28 Jackson Hole speech cited 12-month PCE inflation of 3.7% and said "we have work to do." These collars printed two trading days after that speech. The 10-year Treasury yield closed at 4.77%, its highest since January 2025.

🧵 The thread running through the small names

The five smaller trades are not five unrelated ideas. They sit on one axis: the price of money.

• IEF is the direct expression. It closed at $92.74 against a 52-week low of $92.62 — and someone sold 24,999 at-the-money calls on it, from an open interest of one, into an expiry that lands two days after the September policy meeting. That is a bet that yields do not fall. On this maturity the $0.51 collected is wiped out by roughly an 8 basis point move in the wrong direction. It is a thin cushion, and we say so.
• XEL and LPX are the same trade shape. Both are long puts struck just below the 52-week low — XEL $70 against a $71.29 low, LPX $65 against a $66.12 low — on rate-sensitive equities, bought on the day the 10-year hit a 19-month high. Someone is buying the break of the low on bond proxies.
• LPX has no company event in the window at all. It next reports around November 4 — nineteen days AFTER the put expires. It is a housing hedge wearing a single-stock ticker.
• RUM is the long-duration, unprofitable cohort. It signed a $13.7 billion six-year GPU lease on August 24 — and said the same day it still needs financing to fulfil it, on a $3.58B market cap. Earnings are a month after the option expires. The only plausible dated event inside six weeks is a financing announcement.
• TEX is financed industrial demand, placed into a window with no scheduled event of any kind — no earnings, no ex-dividend, no index change.

Three of the four single-stock names have no earnings before their options expire. Only XEL's March-2027 put spans a report. These are not earnings plays, and anyone trading them as such is trading a different thing than the institution on the other side.

🐍 The one trade with a date on it

SNOW reports Wednesday, September 2 — two days away. A floor block sold 5,690 January-2027 $340 calls and bought 5,690 $420 calls against them, for a $12.81M credit — and both legs printed below their existing open interest, so we genuinely cannot tell whether a new bearish spread went on or an old one came off.

If it opened, the position risks $32.71M to make $12.81M and breaks even at $362.52, +12.0%. That needs to be right roughly 72% of the time just to break even on expectation. The stock closed $331.43, +51.09% year to date, with an all-time high of $341.95 — the short strike sits 2.5% under it.

⭐ And the option market prices a 15.16% move on the earnings print alone — about $50. The stock only has to travel $16.20 to reach the short strike. One earnings report can settle this.

👥 What this means for you

🎰 If you trade for the big score. The honest read on today is that the institutions were buying protection, not lottery tickets. The two structures with real convex payoff are LPX's put spread (risk $825,000 to make $4,175,000, ≈5:1) and RUM's call spread (risk $1,287,500 to make $2,462,500, ≈1.9:1) — and both need a big move: LPX below $63.35 in seven weeks, RUM above $9.03 in six. Note what the professionals did that you probably would not: they capped the payoff to cut the cost. A capped trade you can afford to be wrong in five times beats an uncapped one you can afford once.

📈 If you swing trade. The levels worth writing down are the collar strikes, because they are where someone with real size has decided the world changes. SPY: floor $727, cap $806. QQQ: caps at $740 and $760, floors at $715 and $680. Those are not predictions — they are the boundaries a large book has chosen to live inside for the next three months. The gamma structure agrees on SPY: the heaviest shelves sit at $767 and $770, right where the index closed, with the next resistance at $810 — just above the collar's $806 cap.

💰 If you sell premium. Today's board is a lesson in what your premium is actually worth. IEF's seller collected $0.51 on an at-the-money call that an 8 basis point rate move erases. TEX's seller collected $18.16 on an option worth $18.29 — they paid thirteen cents for the privilege. And SPY's expiring puts returned $11.99M of which $30,000 was time value. A large credit is not the same as a large edge. Before you sell anything, subtract the intrinsic value and look at what is left.

🌱 If you are new. Learn what a collar is, because it is the most useful structure on this page and the least discussed. You own something. You buy a put below the price to stop the bleeding if it falls. You sell a call above the price to pay for that put. You have given up the gains above the call strike; in return you cannot lose more than down to the put strike. Today a very large institution did exactly that on $1.5 billion of S&P exposure and it cost them two basis points — $320,128. You can do the same thing on a hundred shares. The mechanics are identical; only the number of zeros changes.

🧯 Risk control and patience — the part that actually matters

Everything above is a description of what other people did with their money. None of it is a recommendation, and almost none of it is provable today.

• Thirty of thirty-one legs were negotiated blocks. On a cross, an auction or a floor trade, a broker matched a buyer and a seller away from the public order book. There is a known counterparty on the other side, and the "buy" or "sell" label attached to each leg is an allocation convention, not evidence that anybody was eager. Do not read urgency into a number that does not contain any.
• Eleven legs printed at or below their existing open interest. For those, we cannot tell whether a position was opened or closed — and the two possibilities point in opposite directions. Anyone who tells you otherwise from today's tape alone is guessing.
• Size is not conviction. Today's $250.78M in QQQ is $26.20M of actual money at risk. Today's $6.36M in TEX is negative $45,500 of economics. Always look for the net.
• The institutions on this board were mostly reducing risk, not adding it — collars, rolls, hedges and closes. That is worth sitting with before deciding today's flow is a green light for anything.

Position sizes here run into the billions. Yours does not have to. The single most useful habit in this business is being willing to do nothing on a day when you do not understand what you are looking at.

⏳ Tomorrow morning, ≈06:30 ET — what we will check, and what would prove us wrong

We publish a capture rate, not a single number. Three outcomes are always possible on any leg: it opened in full (≈100% capture), it closed, or it landed in the middle as a partial close or a pure transfer between two existing holders — and the middle is common, not exotic.

Leg	Prior OI	Size	Lead branch	What confirms it
SPY Nov-30 $727P / $612P / $806C	91 / 50 / 1	20,008	open	each line up ≈8,000–20,000
SPY Sep-30 $775P · Oct-30 $650P · Mar-27 $700P	544 / 1,717 / 5,837	11,000 / 12,000 / 32,600	open	40–100% capture on each
SPY Sep-18 $665P · Oct-16 $630P	106,542 / 36,356	91,400 / 30,000	transfer or partial	little or no net change
SPY Aug-31 $770P / $775P	—	—	no test exists	they expired today
QQQ Dec-31 $715P / $740C · Nov-20 $760C	1,006 / 3,728 / 5,667	38,000 / 38,000 / 20,000	open	40–100% capture
QQQ Nov-20 $680P	16,331	20,000	transfer first   (margin only 18%)	flat-to-modest change
SNOW $340C / $420C	6,200 / 6,247	5,690	either way — genuinely open	up ≈5,690 = new spread; down = old one closing
NRG $145C / $170C	240 / 2,167	20,000	open	both up ≈20,000
NRG $165C / $200C	26,489 / 26,358	25,000	close	both down; that confirms the roll
IBIT $45C / $57C	42,634 / 34,578	30,000	close	both down; a rise refutes the liquidation read
IEF $92.50C	1	34,999	open	up 25,000–35,000
IEF $93C	38,346	10,000	transfer or partial	little net change
XEL Mar-27 $70P	222	10,000	open	up ≈10,000
LPX $65P / $55P	9 / 4	5,000	open	both up ≈5,000
RUM $8C / $11C	0 / 0	12,500	open	both up ≈12,500
TEX Sep-18 $45C	0	3,500	open	up ≈3,500

Legs of a single package routinely diverge — that has happened in every session for a week, most recently on Friday, where one leg of Corning's roll fell 4,261 while the other rose 6,929. We predict each leg on its own, then read them together afterwards to see how a position actually evolved.

This newsletter describes trades that appeared on the public options tape. It is not investment advice, and none of it is a recommendation to buy or sell anything. 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.