Donald E. Graham liquidates Cable One on a fixed ten-day schedule
Cable One, Inc. (NYSE: CABO) · chairman emeritus, Graham Holdings
Primary document
Schedule 13D/A Nos. 6 & 7
Exit filing
- What he did
- Sold across five consecutive sessions, 13–19 October 2026, through seven vehicles. Beneficial ownership fell 390,361 → 282,882 shares, 6.88% → 4.99% of 5,673,925 shares outstanding. Amendment No. 7 states that on 19 October he ceased to be a five percent owner and that it “constitutes an exit filing.”
Shares sold per vehicle per session, from Schedule 13D/A Nos. 6 (13–14 Oct) and 7 (13–19 Oct). Weighted-average prices in brackets. Trust (d), 96,000 shares, sold nothing on any day and is unchanged in both amendments.
| Vehicle | 13 Oct | 14 Oct | 15 Oct | 16 Oct | 19 Oct | Tranche |
|---|
| Trust (c) | 11,683 | 11,683 | 11,683 | 11,683 | 11,683 | 11,683 |
| Trust (e) | 19,200 | 9,600 | 9,600 | 9,600 | 9,600 | 9,600 |
| Trust (f) | 500 | 500 | 500 | 500 | 500 | 500 |
| Trust (g) | 15,879 | 7,940 | 7,940 | 7,940 | 7,940 | 7,940 |
| Graham, direct | — | — | — | 5,459 | 1,820 | 1,820 |
| Revocable trust | — | — | — | 5,593 | 5,438 | ~5,593 |
| Trust (d) | 0 | 0 | 0 | 0 | 0 | — |
| WAP range | $11.72–13.02 | $9.00–9.44 | $8.34–8.52 | $7.23–7.98 | $6.07–6.63 | −53% |
Consensus read
CoBank sued on 9 October to block Cable One’s $480m purchase of Mega Broadband; a court paused the deal while a $2.45bn refinancing hung on the outcome; the shares lost 35%, then 22%, then 14.64% to $7.23 on 16 October.
Graham, who rode the stock down from its late-2020 peak at a reported cost of more than $1bn, finally capitulated: an eleven-year holder panic-selling into a broken deal, day after day, at steadily worse prices. The two amendments in six days are the signature of a decision unravelling in real time.
Contrarian read
This was not day-by-day capitulation. It was a pre-sized, ten-tranche, full-exit schedule — fixed on or before 13 October and executed without a single adjustment through a 53% collapse.
Amendment No. 6’s language, that he “intends in the near term to dispose of some or all” of his stock, was not an intention forming. It was disclosure of a programme already running, and already sized to zero.
And it was not indiscriminate: one trust holding 96,000 shares — the largest single block, 34% of what remains — was deliberately left out of it.
The evidence that discriminates
A capitulating seller varies size with the news. A pre-sized schedule does not. The divisibility settles it:
- Trust (c) held 116,830 before the first sale and sold 11,683 on each of five sessions. 116,830 = 11,683 × 10, exactly.
- Trust (e) held 96,000 and sold 9,600 a day after a doubled first day. 96,000 = 9,600 × 10, exactly.
- Trust (f) held 5,000 and sold 500 a day. 5,000 = 500 × 10, exactly.
- Trust (g) held 79,395 and sold 7,940 a day after a doubled first day. 7,939.5 × 10.
- The two accounts he controls alone — his own name and his revocable trust — started three sessions later, on 16 October, and run their own tens: 18,194 ≈ 1,820 × 10; 55,927 ≈ 5,593 × 10.
Four independent trusts and two personal accounts, six positions, all divisible by ten at the observed daily rate. That is not what a series of daily decisions produces. It is what one instruction produces.
Inference, labelled: neither amendment mentions a Rule 10b5-1 plan, and neither states when the schedule was set. That it was set before the first sale is my inference from the divisibility — the filings do not say it. What the filings do say is the quantity, the date and the price, and those are not in dispute.
What would falsify me
Amendment No. 8 showing the daily quantities changing, the programme stopping short of ten tranches, or any sale at all out of trust (d). Any of those and the sizes were being decided day by day, not in advance.
Equally fatal: a disclosed Rule 10b5-1 adoption date after 13 October, which would put the decision inside the collapse rather than before the first sale.
Pattern vs his own historyOriginal Schedule 13D 2 July 2015. Amendments 13 November 2015, 22 December 2017, 18 January 2018, 16 December 2019, 8 May 2023 — five in roughly eight years, none reporting a disposal programme. Then Nos. 6 and 7 six days apart, the second an exit filing. At the end of it, 10,915 shares remain in his own name.
Daniel Dines sold $33m of UiPath. The act that mattered was five weeks earlier
UiPath, Inc. (NYSE: PATH) · founder, CEO, chairman, 10% owner
Primary document
Form 4 + Form 144
10b5-1 plan 15 Jul 2026
- What he did
- Sold 2,500,000 Class A shares on 15–16 October: 1,250,000 at $13.4449, 1,202,400 at $12.9787, 47,600 at $13.8325 — $33.07m. Direct Class A 31,491,238 → 28,991,238. The Form 144, filed 15 October by ICEVULCAN INVESTMENTS LIMITED, puts the acquisition date at 21 April 2021 and the nature at “Founders Shares”, names Morgan Stanley as broker, gives the plan adoption date as 15 July 2026, and shows one prior sale in three months: 1,402,347 shares on 19 August for $22,539,362.39.
Consensus read
Routine, pre-planned founder diversification. The plan was adopted in July, the sales are mechanical, the Form 144 is a round number, and he still holds close to 29 million shares. A founder taking some money off the table at a depressed price, which is mildly bearish at the margin and otherwise not news.
Contrarian read
The sale is routine — see the consensus section below, where I say so. What is not routine, and what nothing I could find reported, is that the shares had to be manufactured first.
On 8 September he converted 5,000,000 super-voting Class B shares into Class A to feed this plan. The sale is reversible in principle; he could buy stock back tomorrow. The conversion is not. It retired 175,000,000 votes permanently, and it was sized to the plan, not to a view on price.
The evidence that discriminates
- The conversion is exactly the plan’s size. 5,000,000 converted equals exactly the Class A increase from 26,491,238 to 31,491,238, and exactly 2× the 2,500,000 on the 15 October Form 144. A diversification decision does not land on a round multiple of the selling tranche.
- The Form 4 names the plan inside the structural act. The 8 September filing ties the move to IceVulcan Investments Ltd. to “a trading plan adopted July 15, 2026.” The conversion document itself points at the sale document.
- It is a carve-out, not a collapse. Ice Vulcan Holding Limited retains 59,690,706 Class B. He did not unwind the dual-class structure; he cut precisely enough out of it to cover the plan.
- The August sale needed no conversion. 1,402,347 shares sold on 19 August came out of the pre-existing direct pool. The conversion appears only once the remaining tranche exceeds what was already Class A. That sequencing is my inference from the holdings arithmetic; the filings state the counts, not the reason.
- Charter terms, primary: Class B carries 35 votes per share and converted Class B “may not be reissued.” The vote loss is permanent by construction.
Attack the measurement. Every write-up of this filing says Dines “still holds 28,991,238 shares.” That is his Class A only. Set beside it: 59,690,706 Class B in Ice Vulcan Holding, and 9,615,297 Class A under the March 2026 divorce-related arrangement where he keeps voting rights but reports no Section 16 pecuniary interest. The number in every headline understates the economic position by roughly two-thirds and is silent on the voting position — which is exactly where the 8 September act landed.
What would falsify me
A prior Class B→Class A conversion at a similar cadence, or any conversion in his record with no plan reference attached, would make 8 September ordinary plumbing rather than a sized-to-plan structural act.
Concretely observable: the next Form 144 drawing on shares that required no conversion, or a Form 4 showing a conversion unconnected to a trading plan. Either and the claim fails.
Pattern vs his own historyThe only Class B→Class A conversion in his Form 4 record back to at least December 2025. Selling tranches over the same stretch: many at 45,000 shares through December 2025, 135,000 in late January 2026, 1,402,347 on 19 August, 2,500,000 on 15–16 October. The tranche has grown fifty-five fold in ten months, and the latest one is the first that his unconverted Class A could not cover.
Larry Ellison pledges a second public stock, and a private one
Oracle Corporation (NYSE: ORCL) · co-founder, CTO & chairman
Single-source
Oracle figures primary
Tesla proxy primary
- What he did
- The Wall Street Journal reported on 20 October 2026 that Ellison pledged shares of Oracle, Tesla and SpaceX as collateral for bank loans in the run-up to his son David’s roughly $110bn Warner Bros. Discovery acquisition, which closed 6 October 2026. Secondary accounts add that Citigroup and JPMorgan took SpaceX and Paramount Skydance collateral in the days before the close. The Oracle leg is documented in the DEF 14A filed 25 September 2026: 413,000,000 shares pledged at the 21 September record date, against 1,158,375,174 beneficially owned, up from 346,000,000 a year earlier.
Consensus read
The classic borrow-don’t-sell play at scale: Ellison levers his Oracle stake rather than realising gains, and the money bankrolls his son’s media empire. The risk is the standard one — an Oracle drawdown triggers calls, forcing sales into weakness, and a third of his Oracle position is already encumbered.
Contrarian read
The Oracle number is the part of this everyone can see and the least informative part of it. It exists only because Ellison is an Oracle officer and director. The legs that would actually change your estimate of his leverage — Tesla, SpaceX, Paramount Skydance equity — sit outside every filing regime that applies to him.
“How levered is Ellison” is therefore a question whose denominator is chosen by disclosure rules, not by him. Treat 413m/1,158m as a floor on an unknown, not as the ratio.
The evidence that discriminates
The clean test arrived in the window, from Tesla’s own board. The DEF 14A filed 19 October 2026 (record date 13 October, meeting 3 December) tightens Tesla’s pledging policy: it caps the aggregate loan collateralizable by “our CEO”’s pledged stock at the lesser of $3.5bn and 25% of the pledged stock’s value, and cuts the maximum borrowable percentage for other directors and officers from 25% to 15%.
Ellison left Tesla’s board in 2022 and holds below 5%. He is in neither category. The same board that just wrote tighter pledging limits has no visibility into, and no authority over, the Tesla shares he is reported to have pledged — which is precisely why no Tesla document can carry a number for them. SpaceX is private, so the same is true a fortiori.
And the one number that does exist is staler than it looks: 413,000,000 is a 21 September record-date snapshot, already two weeks old at the 6 October closing and a month old when the WSJ story ran. It cannot contain anything done to fund the deal.
What would falsify me
A UCC-1 financing statement naming the Tesla or SpaceX collateral, a Schedule 13D/G from Ellison on Tesla, or a Paramount Skydance filing that puts a share count on either pledge. Any of those and this is simply unreported, not structurally undisclosable, and my claim about the disclosure perimeter is wrong.
Also falsifying: evidence that the Tesla and SpaceX pledges predate 2026, which would make them habit rather than break.
Pattern vs his own historyOracle pledges ran 346,000,000 → 413,000,000 year on year, +19.4%, with a reported +67,000,000-share increment on 25 September. That is a slow, disclosed, single-asset habit of long standing. Yesterday’s edition looked for an October increment to it and found none — correctly. The change this week is not the size of the Oracle pledge. It is that the collateral stopped being only Oracle.
Applied Digital had one insider buyer, not two
Applied Digital Corp. (NASDAQ: APLD) · Jason Gechen Zhang, president · Wes Cummins, CEO & chairman
Primary document
Two Forms 4
No 10b5-1 plan
- What they did
- Both bought on 16 October 2026, open market, code P, no plan reported. Both Forms 4 were signed by the same attorney-in-fact and accepted by EDGAR five minutes apart on 19 October, at 07:06 and 07:11 ET. Zhang bought 300,000 shares at a $18.86 weighted average across a $18.40–$19.16 range — $5,658,000. Cummins bought 55,803 shares at a single price of $18.04 — $1,006,686. The stock rose 6.6% on 19 October.
Consensus read
Two Applied Digital insiders put $6.7m of their own money into the stock as it came off a 52-week low, with no trading plan to hide behind. Management is telling you the selloff overshot. The market agreed and marked it up 6.6%.
Contrarian read
There is one buyer here. Measured against what each man actually owns, Zhang committed 15.3% of his own shares and Cummins 0.25% of his.
Reporting the pair as a single $6.7m figure converts a 61:1 asymmetry into a 5.6:1 one, and lets the weaker signal borrow the stronger one’s credibility. Zhang’s purchase is a real commitment. Cummins’ is a rounding error on his own position.
The evidence that discriminates
The denominators are in the Form 4 footnotes, and they are not the reported holdings.
- Cummins reports 23,477,489 shares — 5,874,453 direct (including 1,500,000 unvested RSUs granted 6 January 2026 and a 742,166-share IRA), 17,590,238 in Cummins Family Ltd., 12,798 in 272 Capital. Strip the unvested RSUs and he owns 21,977,489 shares. 55,803 ÷ 21,977,489 = 0.25%.
- Zhang reports 2,839,227, of which 500,000 are RSUs granted 6 February 2026 with nothing vesting before February 2027, and 375,000 are the unvested remainder of a 750,000 grant from 8 August 2025 vesting March 2027 to March 2028. Strip those and he owns 1,964,227. 300,000 ÷ 1,964,227 = 15.3%.
- The dollar ratio is 5.6:1. The stake-weighted ratio is 61:1. Every headline used dollars.
- They were not the same order. Cummins filled at one price, $18.04, below every trade Zhang made ($18.40–$19.16). One block, early; one order worked through the session.
This is the denominator problem in its ordinary form: a widely quoted figure whose base is not a choice variable. “Dollars spent” measures how rich the buyer is at least as much as how convinced he is.
What would falsify me
A Cummins Form 4 in the next few sessions showing substantially more buying — that would make 16 October a first instalment rather than a token.
Or evidence that his 55,803 was sized by a company trading window, an ownership-guideline cap, or available liquidity rather than by choice. If the size was constrained, reading it as low conviction is wrong.
Pattern vs their own historyStated limit: I could not retrieve either man’s complete prior open-market purchase record from sources reachable here, so I am not claiming either purchase is a first. What the documents do establish is that both are coded P, both are dated 16 October, and neither reports a Rule 10b5-1 plan.
John Paulson’s $250m — and why nobody can yet say whether a dollar moved
Paulson Family Foundation · Hebrew University of Jerusalem
2+ independent sources
No primary instrument
Unresolved
- What he did
- Announced $250,000,000 to the Hebrew University of Jerusalem on the evening of Saturday 17 October 2026, at the university’s board of governors gala — the largest in its 101-year history. The Mount Scopus campus is to be renamed “The Alina and John Paulson Campus, Mount Scopus, Jerusalem.” Four stated destinations: a new engineering faculty, expanded medical and biomedical programmes, trauma and post-trauma recovery research, and a think tank. Prior gifts to the institution total roughly $46m, taking his lifetime giving there to about $296m.
Consensus read
The largest gift in the university’s history and an explicit solidarity statement; the philanthropic capstone of a fortune made in a single famous trade, now run as a family office. Naming rights on a whole campus mark it as a legacy commitment rather than a programme grant.
What I will not claim
I went looking for the contrarian read and the evidence to support one does not exist yet. No gift instrument, giving vehicle or payment schedule was disclosed, and until one surfaces nobody — including the people reporting the number — can say whether $250m changed hands on 17 October or whether this is a multi-year pledge.
That is a real gap, not a hedge, and the honest move is to name the hypotheses rather than pick one.
Three live hypotheses, and what would separate them
- Completed irrevocable transfer to the university or a supporting organisation, dated on or near 17 October.
- Multi-year pledge payable from the Paulson Family Foundation, in which case the 2026 cash outlay is a fraction of the headline and the rest is a receivable on the university’s books.
- Transfer into an intermediary vehicle — donor-advised or a dedicated supporting organisation — from which distributions run over years, which is legally a completed gift and economically closer to (2).
Two documents separate them, and neither exists yet: the Paulson Family Foundation’s next Form 990-PF grants-paid schedule, and whether Hebrew University’s audited statements book this as contribution revenue or as a pledge receivable. Until then the correct position is that the announcement is confirmed and the transfer is not.
What would resolve it
A 990-PF line showing a single $250m grant paid in the 2026 year settles it for (1). A pledge receivable in the university’s statements settles it for (2). A named intermediary in either document settles it for (3).
Pattern vs his own historyRoughly $46m to this institution before this week, against $250m now — a single gift 5.4× everything he had given it previously. That jump is real whichever hypothesis holds. It is the timing of the money, not the size of the commitment, that is open.