Two people broke their own pattern this week and the loudest headline on the tape — a co-founder’s holding falling 100% — is not one of them. It is a denominator.
The gaps between his GME purchases went 292 days (3 Apr 2025 → 20 Jan 2026), then 233 days (20 Jan → 10 Sep 2026), and then 11, 8, 3 and 11 days. That is a roughly twentyfold compression of his own buying frequency, and it starts seven days after the company closed a $1.4bn debt-for-stock exchange.
The price discipline broke at the same moment. Every GME purchase in his record from June 2023 onward sat in a band of $20.38 to $22.94. The last three are $23.48, $24.41 and $26.98 — the first prints above $23 since March 2022, and the most expensive shares he has bought in the post-2022 regime.
Item 5(e) states it plainly: as of 8 October 2026 the reporting persons ceased to be beneficial owners of more than 5% of the common stock. The group holds 10,866,132 shares against an assumed 220,678,623 — 4.9%. Before the block it was 16,543,084, or 7.50%.
The same filing discloses, in Item 4, that Lubin contributed pre-funded warrants to Permanent Highest Power Capital LLC and then gifted the LLC interests to a trust — the Gradient Ascent Trust, on which the Form 4 footnote names his son as co-trustee — expressly for estate planning. That line exists because a 13D amendment had to be filed. There is no obligation to file the next one.
He converted 200,000 Class B shares to Class A and sold 200,000 Class A in four tranches the same day. The Class A line therefore closes at zero, which is what “−100%” measures. He still holds 2,720,306 Class B. The sales ran under a Rule 10b5-1 plan adopted 12 June 2026. The real break in his year is elsewhere, in the GRAT line — see the log.
Nothing else in the window broke a person’s own pattern. Checked and found to be cadence: Duffield (Workday — 99,228 shares on 6 Oct and 99,835 on 9 Oct, both convert-and-sell from the David A. Duffield Trust dated 14 July 1988 under a plan adopted 2 Dec 2025, with 35,580,586 Class B untouched), Trøim (Borr — the 13 Oct tranche is the sixth of a continuous run, and the disclosure gap flagged in Editions 027 and 031 has now closed; see Where Consensus Is Right), Kurtz (CrowdStrike — 28,920 shares, 0% position change), Ullal (Arista — field work carried in Editions 027 and 031), Control Empresarial de Capitales (the Slim vehicle — Talos and PBF, carried in Edition 031), Isq Global Fund II (Kinetik — a fund, not a person). Checked and found empty for 13 and 14 October: Schedule 13D and 13D/A filings naming an individual other than Trøim and Lubin, named-buyer prime property, dated auction results with a named principal, Giving Pledge updates, and domicile changes. Nippon Life’s Corebridge purchase and Tether Global Investments’ Barrick purchase are both corporate, with no named individual as reporting person, and are excluded.
Three person-actions, each read against the filing fields rather than the headline.
Who · Ryan Cohen, President, CEO and Chairman, GameStop Corp | What · 67,130 shares at $26.75 and 777,000 at $27.00 on 13 Oct 2026, $22,774,728, all direct, no 10b5-1 plan | Holding after · 42,492,652 Class A
FORM 4 · COHEN RYAN · GME · accepted 2026-10-13 16:06 EDT
Table I 10/13/2026 P 67,130 (A) $26.75 owned after 41,715,652 D
Table I 10/13/2026 P 777,000 (A) $27.00 owned after 42,492,652 D
Table II — none — · Explanation of Responses: empty · 10b5-1 box: not checked
Both prices are flat, not weighted averages, and there is no stated range — so these are two limit levels, not a day’s VWAP. The filing was accepted the same afternoon as the trade, two business days inside the deadline. And on the same day, in a podcast published 13 October, Cohen called the GameStop investment “a screw up” and told listeners “bet against us.”
The strongest insider signal available: a founder-CEO putting $22.8m of his own money into his own stock, with no trading plan to hide behind, for the fifth time in five weeks, at ever-higher prices. Four other directors have bought alongside him. He is paying up because he believes the turnaround is working and the market has not caught up; the self-deprecating podcast line is swagger, not information.
The buying is percentage maintenance, not a price opinion. On 3 September GameStop issued about 55.5m shares to retire $1.4bn of converts. That diluted Cohen from roughly 8.54% to 7.60% of the company. Every purchase since has walked him back up, and he has now stopped 0.12 points — about 595,000 shares — short of where he stood before the exchange. The target is a share count, which is why he paid whatever it cost.
| Date | Shares | Price | Held after | % of co. |
|---|---|---|---|---|
| 20 Jan 2026 | 1,000,000 | $21.36 | 38,347,842 | 8.54% |
| 3 Sep 2026 — exchange closes, +55.5m shares | — | — | 38,347,842 | 7.60% |
| 10 Sep 2026 | 1,000,000 | $20.38 | 39,347,842 | 7.80% |
| 21 Sep 2026 | 1,150,680 | $22.94 | 40,498,522 | 8.03% |
| 29 Sep 2026 | 450,000 | $23.48 | 40,948,522 | 8.12% |
| 2 Oct 2026 | 700,000 | $24.41 | 41,648,522 | 8.26% |
| 13 Oct 2026 | 844,130 | $26.98 | 42,492,652 | 8.42% |
Share counts and prices from Forms 4. Percentages are Inference — computed on 504.50m shares outstanding and an approximate 55.5m issued in the exchange; neither the pre-exchange count nor Cohen’s percentage is stated in any of these filings.
Three facts separate the readings, and none of them is “he bought.” Both readings predict buying, so the volume and the dollar figure are worthless here.
One — the start date. The exchange closed on or about 3 September 2026. His first purchase of the run is 10 September, after a 233-day gap. A valuation view does not arrive seven days after a share issuance and not once in the eight months before it.
Two — he bought more as the price rose, not less. 1,000,000 at $20.38, then 1,150,680 at $22.94, then rising prices to $26.98 — a 32% markup paid against his own opening print in the same sequence. A buyer with a valuation view scales in on weakness. A buyer with a share-count target is price-insensitive by construction, which is the behaviour on the tape.
Three — where he stopped. Restoring 8.54% of 504.5m needs 43,087,943 shares. He is at 42,492,652. Of the 0.94 percentage points the exchange took, he has bought back 0.82. A conviction buyer has no reason to decelerate within a tenth of a point of a number that is only meaningful as a pre-dilution baseline.
What the documents do not support is any claim about motive. Inference Three hypotheses remain live and the filings cannot separate them: percentage restoration; a deliberate signal ahead of a further capital action, with $2.8bn of 0.00% converts still outstanding to 2030 and 2032; or genuine valuation conviction that happens to have begun the week after the dilution. The first is the only one with a dated trigger attached.
Cohen buying materially past 43.1m shares — roughly 8.6% — without a new dilution event in between. That would show accumulation, not defence, and the pre-exchange baseline would stop being the explanation.
Symmetrically: if GameStop converts any part of the remaining $2.8bn of notes into stock and Cohen does not buy again within a few weeks, the mechanical response the thesis predicts has not happened and the thesis is wrong.
And if a forthcoming 10-Q or proxy shows a pre-exchange share count materially different from 449m, the 8.54% baseline moves and the “stopped just short” observation dissolves. That number is my calculation, not a filed figure.
Six GME purchases on the record before this run: 100,000 at $101.76 (Mar 2022), 443,842 at $22.53 (Jun 2023), 500,000 at $21.55 (Apr 2025), 1,000,000 at $21.36 (Jan 2026). Roughly one a year, in the low $20s, from an indirect vehicle and then directly. The 2026 run is five purchases in 33 days at an average of $23.46, and it is the first time in the record he has bought at successively higher prices rather than on a decline.
Who · Joseph Lubin, Chairman and director, SharpLink Inc; CEO of Consensys Software Inc | What · Consensys Software Inc sold all 5,676,952 SBET shares in a block at $8.39 on 8 Oct 2026, $47.63m | Group after · 10,866,132 shares, 4.9%
SCHEDULE 13D/A No.2 · SBET · accepted 2026-10-13 · event date 2026-10-08
Item 4 “CSI effected the sale to liquidate their position.”
Item 4 pre-funded warrants exercised at nominal price once the share increase permitted it under the beneficial-ownership blockers; contribution of warrants to PHPC and gift of PHPC interests to a trust described as estate planning
Item 5(a) 10,866,132 4.9% (assumed denominator 220,678,623 = 217,223,604 outstanding at 3 Aug 2026 + 3,455,019 warrant shares)
Item 5(c) no transactions in the past 60 days other than as described
Item 5(e) ceased to be beneficial owners of more than 5% as of 2026-10-08
FORM 4 · LUBIN JOSEPH MICHAEL · Director · 10/08/2026 S 5,676,952 (D) $8.39 → 0 · I, by Consensys Software, Inc.
retained: 5,154,213 D · 1,200,000 I (PHPC) · 975,600 I (ConsenSys AG) · 81,300 I (Ethereal Ventures II)
Table II warrants, all exercisable since 05/30/2025, expiring 05/30/2030, no transaction reported:
1,382,007 @ $6.150 · 691,004 @ $6.765 · 691,004 @ $7.380 · 691,004 @ $7.995
An Ethereum co-founder dumping $47.6m of the Ethereum treasury company he chairs, with the stock down 51% over a year, is a vote of no confidence — or, more charitably, Consensys raising cash and cleaning a volatile crypto-equity mark off its balance sheet before the IPO it has already pushed to this autumn. Either way: the insider is out.
He is not out. The group still holds 10,866,132 shares and CSI kept every one of its 3,455,019 warrants — 61% of the share count it just sold, exercisable since May 2025, running to 2030, at a weighted strike of $6.888. What changed on 8 October is regulatory, not economic: the group fell under 5%, and Amendment No. 2 is the last document that will ever carry its stated purpose.
Against “no confidence”: the warrants were not touched, and that was a free choice. They have been exercisable since 30 May 2025. Nothing stopped a cashless exercise alongside the block. Selling the delta-one shares at $8.39 while holding 3,455,019 warrants struck 17.9% below that price, to 2030, is a position that gains if SBET rises and costs nothing if it falls. That is not how a seller with no confidence disposes of an in-the-money derivative.
Against “pre-IPO balance-sheet cleanup”: the warrants again. A derivative on a volatile crypto-treasury equity is exactly the mark-to-market line an issuer scrubs before an S-1. CSI kept it and sold the simpler asset instead. The cleanup story predicts the opposite ordering.
What the 4.9% actually buys. Lubin remains a Section 16 insider by virtue of his directorship, so Forms 4 still capture the numbers. What stops is Schedule 13D — the only filing in this set with an Item 4. The estate-planning architecture disclosed on the way out (warrants contributed to PHPC, PHPC interests gifted to the Gradient Ascent Trust, son as co-trustee) is now a snapshot with no update obligation behind it.
And the honest deflation of my own framing. Crossing 5% needed a sale of 5,509,153 shares. CSI sold 5,676,952 — its entire holding, an overshoot of 167,799 shares, or 3%. There is no evidence the threshold was engineered; the block was sized to the position, and the crossing fell out of it. Inference Intent is unknowable from these documents. The consequence does not require intent: the reporting regime changed on 8 October whether anyone planned it or not, and that is the part worth tracking.
One number I could not verify: whether the PHPC and trust transfers were already disclosed in Amendment No. 1 of 16 April 2026. If they were, this is a restatement rather than a first disclosure. The argument does not rest on it — the last-13D point holds either way — but the “surfaces only now” colouring would be wrong.
A Schedule 13D/A No. 3 from this group, or any Form 4 showing it back above 5%. Either returns them to the purpose-disclosure regime and makes “the last one” simply false.
A cashless exercise and sale of the 3,455,019 warrants in the next two quarters. That would show the warrants were retained for timing, not for convexity, and the whole “kept the optionality” reading collapses into a slower exit.
A Consensys S-1 that carries the SBET warrants as a disclosed asset. That would make the cleanup hypothesis compatible with the warrant retention after all, and my second discriminator stops discriminating.
The original 13D was filed 6 June 2025; Amendment No. 1 followed on 16 April 2026. Item 5(c) reports no transactions in the preceding 60 days other than this one. So on the record available, this is the first disposal by CSI of its SBET common since the position was established in May 2025, and it took the holding from the full block to zero in a single print. The price context: $8.39 against a last-reported book value of $6.50 a share, or 1.29× — while the retained warrants strike at 1.06× that book. The vehicle sold at a premium to net asset value and kept the right to re-enter at roughly net asset value for four more years.
Who · Gabriel M.I. Mecklenburg, Executive Chairman and co-founder, Hinge Health Inc | What · converted 200,000 Class B to Class A and sold 200,000 Class A in four tranches on 12 Oct 2026 at a blended $103.13, $20,626,371, under a plan adopted 12 June 2026 | Holding after · 0 Class A, 2,720,306 Class B
FORM 4 · MECKLENBURG GABRIEL M.I. · HNGE · 10/12/2026 · 10b5-1 box CHECKED, plan adopted 06/12/2026
Table I C 200,000 (A) $0 → 200,000 D
Table I S 8,642 $101.5308 → 191,358 | S 91,502 $102.5323 → 99,856
Table I S 59,988 $103.4755 → 39,868 | S 39,868 $104.3379 → 0
Table II Class B C 200,000 (D) → 1,644,004 D
Table II Class B 383,592 I, by Family Trust (no transaction)
Table II Class B 692,710 I, by GRAT (no transaction)
— compare FORM 4 · 04/01/2026 —
Table II Class B 1,793,102 D · 1,092,119 I by GRAT · 383,592 I by Family Trust · excludes 944,250 PSUs
A co-founder has taken his position to zero — a 100% cut — after $83.8m of selling in 2026 at prices running from $38 to $103. Read as a signal, it is the clearest kind: the man who built it is leaving at the top, with the stock near its high, and the aggregators are right to put it at the front of the insider tape.
“−100%” is measuring a transit account. Class A exists in his hands for hours. The number worth reading is in Table II: the GRAT has gone from 1,092,119 shares on 1 April to 692,710 on 12 October — −399,409, or −36.6% — while the Family Trust sits at 383,592 on both dates, unchanged. The estate plan is in its payback phase; the selling is the visible, uninformative half.
On the measurement. He converted exactly 200,000 Class B and sold exactly 200,000 Class A on the same date. The closing Class A balance is zero by construction, not by decision, and it has closed at zero after each of the ten sales on his 2026 record. A “percentage of holdings sold” computed on the Class A line will read −100% every single time, regardless of what he owns. That is a denominator that is not a choice variable.
On the GRAT, and why annuity payback is the better-supported read. Two hypotheses fit a shrinking GRAT line: the trust is paying its annuity back to the grantor in kind, or its term has ended and the remainder has passed to beneficiaries. They make opposite predictions about where the shares went. If the remainder passed out, the shares leave Section 16 reporting and his total falls. If the annuity is being paid in kind, they reappear in his direct line. His direct Class B went from 1,793,102 (1 Apr) to 1,644,004 (12 Oct) while 983,334 shares were converted out of it over that span — so the direct line was replenished by 834,236 shares. The GRAT’s 399,409 accounts for roughly half of that; Inference the balance is most likely PSU vesting, given the 944,250 performance units footnoted in April, but no filing in this set states the source. Direction of travel favours annuity-in-kind.
Why it matters. A GRAT paying its annuity back means the freeze was executed, the appreciation above the hurdle has already moved to the remainder beneficiaries, and the grantor is now receiving the principal he was always going to receive. Hinge Health has gone from $38 to $103 this year. Whatever that appreciation did for the remainder interest, it is done and outside his estate. The $83.8m of visible selling is the part that generates headlines; the 399,409 shares are the part that moved wealth between generations.
A Form 4 or Form 5 showing the GRAT line continuing to fall while his direct Class B falls alongside it. That pattern is a term-end distribution to remainder beneficiaries, not an annuity payback, and my read of the mechanism is wrong.
A proxy or S-1 beneficial-ownership table showing the GRAT and Family Trust aggregated into a single reported holding that has not moved at all. That would mean the line changes I am reading are reclassifications between vehicles, not transfers.
On the measurement claim there is nothing to falsify — the Class A balance closing at zero is arithmetic, not interpretation. What can be wrong is my inference about why the direct line replenished.
Ten reported sales in 2026, roughly 1,250,000 shares for about $83.8m, at 1 Apr ($38.45), 21 Apr ($45.05), 8 May ($55.01), 26 May ($53.79), 1 Jun ($60.22), 5 Jun ($65.53), 1 Jul ($85.05), 11 Sep ($88.97), 14 Sep ($93.22) and 12 Oct ($103.13). Every one under a 10b5-1 plan; every one convert-and-sell; every one ending at zero Class A. The 12 October tranche is the highest price and the second-largest size. Against that cadence, nothing about this filing is unusual — which is precisely why the −100% framing misleads.
Tested for a cleverer story and could not find one that the evidence supports.
Who · Tor Olav Trøim, director, Borr Drilling Ltd | What · Drew Holdings Ltd bought 1,500,000 shares at $4.93 on 13 Oct 2026, $7.40m | Also · Schedule 13D filed 13 Oct 2026 reporting 12.2%
Edition 027 flagged that Trøim’s standing beneficial-ownership filing said 8.8% while the Form 4 trail implied 10.38%. Edition 031 flagged it again at an implied 12.09% and named the falsifier: an amendment landing. It landed. A Schedule 13D was filed on 13 October 2026 by Drew Holdings Ltd reporting 12.2% — about 37.64m shares on a 308.51m count — with Item 4 stating investment purpose, continued review of the position, possible further purchases, possible hedging or distribution to other entities, and no current plans or proposals. The gap is closed and the flag is withdrawn.
I went looking for a second story in it, and tested three. Price support ahead of a financing: nothing in the window shows a Borr equity or debt raise, and a director buying into the market is an expensive and ineffective way to hold a price. Defensive accumulation against a bidder: Item 4 expressly disclaims plans or proposals, there is no competing 13D, and a 12.2% stake is not a blocking position in a Bermuda-incorporated issuer. A cosmetic or offsetting trade: there is no disposal anywhere in the group, and the buying runs through his wholly-owned vehicle at progressively higher prices — $4.2428, then $4.74, then $4.93 — which costs real money and buys no optics.
The one genuinely awkward fact for the ordinary reading is that he is buying strength, not weakness: BORR is up 74% over twelve months and trades at 1.50× a $3.12 book value, and the 13 October tranche is his dearest yet. But averaging up is not evidence of anything other than a view that the stock is going higher still, and a director adding more than 2% of a company over twelve months through a personal vehicle, with a correctly filed 13D and no offsetting sale, is the thing it appears to be. No discriminating evidence exists for a cleverer reading, so the claim is dropped rather than hedged.
One mechanic, three independent filings in nine days.
This is the same measurement failure three times over, from three unrelated issuers, in one window:
The common structure: the person’s wealth sits in Class B, or in a trust, or in a holding vehicle. Class A is created on the morning of a sale and extinguished the same afternoon. Any screen that divides shares sold by Class A held produces a number between −49% and −100% no matter what the person actually owns, and it will produce that number again next week. The denominator is not a choice variable. Three of the four largest “ownership change” figures on this week’s tape are artifacts of it.
Practical consequence for reading the tape: ignore the percentage column entirely. Read the owned-after figures in Table II, by ownership code, and note which lines carry no transaction row — those are where the wealth is.
One private company, one shareholder, Ontario. Most of what is above needs scale, a US domicile, or a liquid public stock. Here is which parts survive the trip down.
Strip the zeros off and the GameStop sequence is simple: the company issued shares, the founder’s percentage fell by 0.94 points, and he spent $97m of after-tax money buying it back in the open market — and is still short. He had no contractual protection against dilution, so he had to repurchase his own position at a 32% markup to his own starting price.
A private Ontario corporation is where that protection is cheap. Pre-emptive rights — the right of first refusal on any new issuance, pro rata to your existing holding — are a clause in the articles or the unanimous shareholder agreement, not a $97m cheque. Anti-dilution and pre-emptive provisions are routine in Canadian shareholder agreements and are vastly easier to insert before an outside investor is at the table than after.
First step this week: pull your articles of incorporation and any shareholder agreement and look for two things — whether the directors can issue shares without offering them to existing holders first, and whether there is any pre-emptive right at all. If the answer is “yes they can” and “no there isn’t,” that is the gap Cohen is paying to close. It is a short conversation with a corporate lawyer and it is far cheaper now than at a financing.
A grantor retained annuity trust is a creature of US estate-tax law, and Canada has no estate tax for it to plan around; it has a deemed disposition at death instead, which taxes accrued capital gains. So the instrument does not transfer. The objective does.
The Canadian analogue is the estate freeze: exchange your common shares in the operating company for fixed-value preferred shares, and let a family trust subscribe for new common shares at nominal value. Your future tax liability is capped at today’s value; the growth from here accrues to the trust. The appreciation Mecklenburg’s remainder beneficiaries captured as Hinge Health went from $38 to $103 is the same appreciation an Ontario freeze is designed to put on the other side of the line — and, like a GRAT, it works best before the growth, not after.
Three things to know before anyone says “just do a freeze.” A family trust faces a 21-year deemed disposition, so it is a structure with a clock, not a permanent one. Trusts now face expanded annual reporting obligations in Canada, so it carries ongoing compliance cost. And a freeze interacts directly with the lifetime capital gains exemption on qualified small business corporation shares — done correctly it can multiply that exemption across family members, done carelessly it can disqualify the shares for it. This is the single highest-value and highest-risk item in this report for someone in your position, and it is exactly the kind of thing that must not be done from a newsletter.
First step this week: establish whether your shares currently qualify as QSBC shares at all. The test is about asset composition and holding period, it is checkable from your own balance sheet, and it determines whether a freeze is a tax win or an expensive reorganisation. That is a question for your accountant, and it is the right question to arrive with.
The 5% threshold. There is no Schedule 13D for a company nobody else owns shares in. Dropping below a disclosure threshold is only valuable if you were above one. The transferable residue is not a structure but a habit: the documents a person is obliged to file are the only ones that explain purpose, and obligations end. When you read about someone’s holdings, check which filing regime they are still inside.
Duffield’s revocable living trust. A US revocable living trust is primarily a probate-avoidance device, and it does not map onto Ontario. A trust you can revoke generally does not remove assets from your estate for Canadian purposes, and the Ontario equivalents for probate planning — alter ego and joint partner trusts — are not available until age 65 and carry their own attribution and deemed-disposition consequences. Multiple wills are the more common Ontario tool for keeping private-company shares out of the estate administration tax calculation. None of this is the same device, and borrowing the American version would be actively wrong.
The Class A transit account. Dual-class conversion mechanics require a public listing and two share classes. You have neither. What does carry over is the reading discipline from “What Repeats”: when someone shows you a percentage change, ask what is in the denominator before you ask what the number means. That applies to a cap table, a customer-churn figure, and a Form 4 equally.
This report is information gathered from public filings and press, not financial, tax, or legal advice. Nothing here is a recommendation to buy, sell, or hold any security. Every structural item above — estate freezes, family trusts, QSBC status, the capital gains exemption, pre-emptive rights, probate planning — depends on facts specific to you and your corporation, and must be confirmed with a licensed Canadian accountant and a corporate or tax lawyer before anything is signed or filed. Percentages marked as inference are the author’s calculations from stated figures, not numbers reported in any filing.