Thirty per cent of “Lefkofsky sold $17.6m” is not his money, and the ten million shares that did not move are the actual disclosure
Eric P. Lefkofsky · Tempus AI · Form 4, 29 October
Confirmed · primary Form 4 XML on sec.gov
Transactions 27 Oct 2026
10b5-1 adopted 8 Mar 2026
Estate-vehicle purpose is inference
- Who
- Eric P. Lefkofsky — chief executive, chairman, director and 10% owner of Tempus AI; co-founder of Groupon, Echo Global Logistics, InnerWorkings and Mediaocean. The Form 4 names him as sole manager of Black Media LLC, Blue Media LLC and Gray Media LLC; a member of and controlling a member of 346 Long LLC; trustee of both the Lefkofsky Family Foundation and the Vas.org Foundation; and records that his spouse is trustee of the Lefkofsky Family 2025 GRAT and the Lefkofsky Family 2026 GRAT.
- What he did
- On 27 October, 250,000 Class A shares were sold across six weighted-average price steps from $67.36 to $72.31 (full range $66.69–$72.84), for about $17,589,637 — a blended $70.3585. The 250,000 was split across four entities and no others.
- The document
- Form 4, period 2026-10-27, filed 2026-10-29, signed by Andrew Polovin as attorney-in-fact. The Rule 10b5-1 box is affirmatively set; footnote F1 reads in full: “This transaction was made pursuant to a Rule 10b5-1 trading plan adopted by the Reporting Person on March 8, 2026.”
Every line on the 27 October Form 4, by selling entity, plus the holdings that reported no transaction. Share counts and residuals are read directly from the filing; dollar splits are computed at the blended $70.3585.
| Entity | Type | Sold | ≈ Proceeds | Held after |
|---|
| Blue Media, LLC | Personal | 135,000 | $9,498,398 | 14,715,469 |
| Gray Media, LLC | Personal | 57,500 | $4,045,614 | 8,487,783 |
| Vas.org Foundation | Charitable | 50,000 | $3,517,925 | 50,000 |
| Lefkofsky Family Fdn | Charitable | 7,500 | $527,689 | 794,631 |
| Lefkofsky Family 2025 GRAT | Transferred | 0 | — | 5,715,203 |
| Lefkofsky Family 2026 GRAT | Transferred | 0 | — | 4,284,797 |
| Held directly | Personal | 0 | — | 2,098,130 |
| 346 Long LLC / Black Media | Personal | 0 | — | 612 |
Consensus read
The founder-chief-executive of a volatile AI-healthcare company sold $17.6m of stock into strength, on a day the shares traded up through a $6 range, and the aggregators are reporting it as one of the largest chief-executive disposals of the week.
Steelmanned: Lefkofsky is a repeat founder who has taken money off the table at every company he has built, Tempus is not yet a settled business, and a nine-figure holder reducing at $70 after an advance is information about what the best-informed seller thinks the shares are worth. The 10b5-1 plan was adopted in March and does not obscure that the plan itself was a decision to sell.
Contrarian read
This filing is not one decision reported once. It is two different decisions reported together, and the headline number adds them up as though they were the same thing.
Of the 250,000 shares, 57,500 belong to two foundations he is trustee of, not to him. Those shares were given away before this sale; the charities liquidating them is an endowment-management act, not a disposal of his wealth. The remaining 192,500 shares — about $13.54m — are the only part that is his. And that part is 0.69% of the 36,146,625 shares the same form discloses. The real content of the filing is the column that did not move.
Evidence that discriminates
- The arithmetic of the headline, which is checkable to the share. $17,589,637 is what the wires report. $13,544,011 is the personal portion. The published figure is 29.9% larger than the amount of his own money involved. That single division separates “chief executive sold $17.6m” from what the document says, and no reading of the price action can absorb it. The foundations are named on the face of the form.
- Four round allocations, not one dollar target. The entities sold 135,000 / 57,500 / 50,000 / 7,500 shares. A seller working to a dollar figure across a pooled position does not produce four round share counts; a plan with per-entity allocations fixed in March does. This is the fact that tells you the sale was composed before it was executed — which is why reading it as a view on the October price is reading the wrong variable.
- The two GRATs hold exactly 10,000,000 shares between them and traded nothing. 5,715,203 plus 4,284,797 is 10,000,000 to the share — a single ten-million-share block split across two annual vehicles, with his spouse as trustee of both. That block is 27.7% of his disclosed position and 40× the quantity sold. A person acting on a price view sells from the largest pocket. He sold from the third and fourth largest and left the block alone. Inference, clearly labelled: the Form 4 states the vehicles are grantor retained annuity trusts and that his spouse is trustee. It does not state their purpose. That a rolling two-year GRAT programme exists to move appreciation out of the taxable estate is standard practice, not a disclosed fact in this document.
Facts that do not discriminate, and which most coverage will lean on: the $67–$72 price range, Tempus’s recent performance, Lefkofsky’s history of selling into his own companies’ rallies, and the sheer size of $17.6m. Every one of those is equally consistent with a price view and with a scheduled, pre-allocated programme. They are the reason the consensus reading feels evidenced when it is not.
What would falsify me
A Form 4 reporting any sale out of either GRAT. One line against the Lefkofsky Family 2025 or 2026 GRAT would show the ten-million-share block is a trading pocket like the others, and the “which pocket” reading collapses into an ordinary diversification story.
A next tranche with ragged per-entity numbers. If November’s filing shows the entities selling, say, 131,447 and 58,219 shares, then the round allocations were an artifact of this month rather than the plan’s structure, and point 2 above is wrong.
Vas.org going below 50,000. If the foundation keeps selling past the round number, then the exact 50,000 landing was coincidence and the anomaly-board entry should not have been written.
Pattern vs his own history
Tempus Form 4s for this reporting person were filed on 31 July, 20 August, 28 August (two), 18 September, 24 September and 29 October 2026 — a roughly monthly cadence running since the plan was adopted on 8 March. This tranche is in cadence. Derived, not filed: summing every holding line gives 36,146,625 shares, which matches the aggregate the screeners report, so the four selling entities plus the untouched block account for the whole disclosed position. There is no fifth pocket hiding.
Berkshire did not sell DaVita. DaVita bought, and the percentage is the proof
Berkshire Hathaway · DaVita Inc. · Form 4, 29 October
Confirmed · Form 4 + 2 independent reports
Transaction 27 Oct 2026
Attribution contested
Agreement text not read directly
- Who
- Berkshire Hathaway Inc. as a 10% owner of DaVita. Coverage is attributing the action personally — one syndicated piece is headlined “Berkshire’s New CEO Just Offloaded DaVita Stocks: Should You Sell Yours Too?”, putting it on Greg Abel. The position was built from the fourth quarter of 2011 under Buffett-era Berkshire; the attribution to a current decision-maker is the thing this entry disputes.
- What happened
- On 27 October, 382,563 shares were disposed at $180.43, about $69,026,033, leaving 28,314,666 shares held indirectly — roughly 45% of DaVita, the same percentage as before the transaction.
- The document
- Form 4 filed 29 October 2026 at 17:21 ET. Contemporaneous reporting describes the counterparty as DaVita itself, under an arrangement established in April 2024 whereby DaVita repurchases shares quarterly from Berkshire so that Berkshire’s stake “remains at 45%”, priced at a volume-weighted average reflecting shares acquired from public investors over the period.
Consensus read
Berkshire is reducing a long-held position. $69m is real money, it follows years of trimming, and it arrives during a leadership handover — so it is being read as the new regime cleaning up an inherited holding, and as a signal that retail holders should reconsider.
Steelmanned: Berkshire has in fact reduced its DaVita share count repeatedly since early 2025, the dialysis industry faces reimbursement and GLP-1-driven volume questions, and a shrinking share count is the normal first evidence of an exit beginning. Nothing about that chain of reasoning is lazy.
Contrarian read
No sell decision was taken on 27 October, by Abel or anyone else. DaVita’s own buyback shrinks the denominator; the April 2024 agreement requires that Berkshire’s stake be brought back to 45% when it drifts above; Berkshire hands back whatever share count the arithmetic produces.
Berkshire is the passive side of DaVita’s capital return. The informative party in this filing is DaVita’s board, which is still retiring stock. Reading Berkshire’s exposure as falling gets the direction exactly backwards: holding 45% of a smaller company while the company buys back stock is how you stay still, not how you leave.
Evidence that discriminates
- The percentage did not change. Before: about 45%. After: about 45%. This is the crux and it is a single field. Every genuine reduction lowers the owner’s percentage — that is what reduction means. A disposal that leaves the percentage untouched is definitionally not a reduction in economic exposure; it is the restoration of a cap. No version of the consensus read survives that one number.
- The counterparty is the issuer, so there is no timing decision to interpret. The shares went back to DaVita at a volume-weighted average set by DaVita’s own market purchases. There was no bid to hit, no price to pick, no moment to choose. The things an observer normally reads off a sale — when, at what level, into what tape — do not exist here.
- The same mechanic, with the same wording, in February 2025. Berkshire disposed of 203,091 shares, reported at the time as cutting the stake to 45%. The identical sentence attaching to a different share count eighteen months apart is the signature of a rule running, not of a judgement being exercised. A base rate of one repetition is thin; a base rate with the cap named each time is not.
Facts that do not discriminate: the $69m figure, the 2011 cost basis, the GLP-1 narrative, and the Abel transition. Each is compatible with both readings, and the $69m is the one doing most of the rhetorical work in the headlines.
What would falsify me
Any quarter in which Berkshire’s DaVita stake prints below 45%. That would be a real decision rather than a cap being restored, and the mechanical reading would be wrong from that filing forward. It is a single observable number, published quarterly.
A Form 4 showing open-market disposals rather than an issuer repurchase. If the counterparty stops being DaVita, the whole argument goes.
The agreement itself giving Berkshire discretion to decline. Honest limit: this scan did not read the April 2024 agreement directly — the mechanics above come from contemporaneous reporting of it. If the text makes the quarterly repurchase optional on Berkshire’s side, then a choice is being made each quarter after all, and “passive counterparty” is too strong a description.
Pattern vs its own history
Fourth consecutive year of the same shape. The position dates from Q4 2011. The share count has been walked down repeatedly since early 2025 while the percentage has been pinned at 45% throughout. The share count is not the decision variable here. The percentage is — and it has not moved.
Charles Schwab sold $18.1m, and the interesting field is not the number — it is the holder column
Charles R. Schwab · The Charles Schwab Corporation · Form 4, 29 October
Confirmed · Form 4 + 2 reports
Transactions 28–29 Oct 2026
10b5-1 status not visible
Partnership purpose is inference
- Who
- Charles R. Schwab, 89, founder and co-chairman, listed on the form as both director and officer. He is the company’s largest shareholder; Forbes put his net worth at about $13.5bn in January 2026.
- What he did
- 28 Oct: 66,750 shares at a weighted $88.3559 (range $88.185–$88.59), leaving 53,482,691. 29 Oct: 136,400 shares at a weighted $89.5005 (range $89.06–$89.92), leaving 53,346,291. Total 203,150 shares, about $18,105,625. Both lines are marked indirect, held by a trust.
- The document
- Form 4 accepted by EDGAR 29 October at 19:23 ET, signed by P. Blake Allen as attorney-in-fact. The form also reports, with no transactions against any of them: 30,021,463 shares held by a limited partnership, 10,624,797 held by his spouse as trustee, and 44,025 held by 188 Corp. The trust is not named in the filing.
All four reported pockets after 29 October. Percentages computed from the filing’s own figures; the total is 94,036,576 shares. Only the first pocket traded.
| Pocket | Shares | Share of stake | Sold 28–29 Oct |
|---|
| Trust (indirect) | 53,346,291 | 56.7% | 203,150 |
| Limited partnership | 30,021,463 | 31.9% | 0 |
| Spouse as trustee | 10,624,797 | 11.3% | 0 |
| 188 Corp | 44,025 | 0.05% | 0 |
Consensus read
An 89-year-old founder is taking $18m off the table. Routine late-life diversification and liquidity, of a piece with a run of similar sales this year ($2.5m, $4.6m, $5.0m, $13.2m in earlier filings). Nothing to read into it.
Steelmanned: this is almost certainly right about motive, and the sales really are routine. Any reading that treats a founder of this age reducing a concentrated position as a market signal is making the standard error, and the consensus is correctly refusing to make it.
Contrarian read
Agreed on motive, and that is not where the information is. The disclosure worth having is which pocket paid. Both sale lines came from the trust. Across the same two days, the 30,021,463-share limited partnership and the 10,624,797 shares his spouse holds as trustee traded nothing at all.
Read as a stock story this is a $18m non-event — 0.216% of a 94,036,576-share position, too small to be diversification in any sense that word normally carries. Read as a structure story it is a map: it shows where the liquid, still-in-the-estate 56.7% sits, and which 43.3% is being left permanently alone.
Evidence that discriminates
- The holder column across two consecutive days, four available pockets, and zero diversity in which one was used. If the objective were cash, any pocket would do and the largest non-trust holding (31.9% of the stake) is the obvious candidate for a block. It was not touched on either day. The consistency is the evidence: a liquidity-driven seller is indifferent to the source; a structurally-driven seller is not.
- The size relative to the position kills the diversification reading on its own. 203,150 of 94,036,576 shares. Selling a fifth of one per cent does not reduce concentration in any measurable way — it is not what diversification looks like, which means something other than diversification is choosing the amount. The number is more consistent with a scheduled drip than with a portfolio decision.
- Two of the untouched pockets are held by someone other than him. The spouse-as-trustee holding and the partnership are, on the face of the form, not his direct personal property in the way the trust shares are. Inference, clearly labelled: at this scale and age, a large family limited partnership and a spouse-trusteed block are, as a matter of ordinary practice, vehicles through which value has already been moved out of the taxable estate at a discounted valuation. Selling from such a vehicle undoes part of what it was built for. The Form 4 says none of this — it names no purpose for any entity. The partnership could as easily be an investment vehicle he controls outright, which is exactly why this is labelled inference and carries a falsification test below.
Facts that do not discriminate: his age, the $18.1m, the $88–$90 price level, and the string of similar sales earlier in 2026. All are equally consistent with ordinary liquidity and with a structured programme, and the age in particular is doing unearned work in most write-ups.
What would falsify me
One Form 4 line reporting a sale by the limited partnership. That is the whole test, it is a single row in a future filing, and it would show the partnership is an ordinary trading pocket rather than a transfer vehicle.
A proxy or Schedule 13D/G disclosing that the partnership is a personally controlled investment vehicle rather than a family transfer structure. That would leave the observation about the holder column intact but strip it of meaning.
Note also the gap in the evidence: the Rule 10b5-1 status is not visible in the extracted filing text, and that matters. If a plan is checked, the pocket selection was fixed months ago and is even more clearly structural. If no plan is checked, these were live decisions on consecutive days, which makes the untouched partnership a sharper fact, not a duller one.
Pattern vs his own history
Repeated 2026 sales reported at roughly $2.5m, $4.6m, $5.0m, $13.2m and now $18.1m, alongside a separate trust sale of 142,900 shares in an earlier filing. The cadence is established and the amounts are stepping up modestly; neither the size nor the timing of this filing breaks it. What has stayed constant across every one of them, so far as this scan can see, is that the partnership never sells.
Why the Accenture purchase is the one item today where the honest answer is that nobody can tell yet
Julie Spellman Sweet · Accenture plc · Form 4, 28 October
Confirmed · Form 4 + 3 reports
Transaction 26 Oct 2026
Open market, code P
Motive genuinely unresolved
- Who
- Julie Spellman Sweet, chair and chief executive of Accenture, reported as both director and officer.
- What she did
- On 26 October, two purchase lines, code P, both direct: 2,100 shares at a weighted $245.1063 ($244.59–$245.58) and 10,100 at a weighted $246.1701 ($245.59–$246.54). Total 12,200 shares, $3,001,041. Direct holding 36,710 → 48,910.
- The document
- Form 4 filed 28 October 2026, signed by Danika Haueisen as attorney-in-fact. No Rule 10b5-1 plan is reported. No indirect holdings reported. For comparison, her two prior acquisitions are code A, not P: 205 shares at $189.095 on 5 September and 197 shares at $196.755 on 5 October, each footnoted as a purchase from Accenture “pursuant to the Accenture Voluntary Equity Investment Program”.
Consensus read
A chief executive put $3m of her own after-tax money into her own company on the open market, with no plan to hide behind, after a brutal year for IT services. She thinks the market has the AI-disruption-of-consulting story wrong. The market agreed: the shares rose about 3.5–4.4% on 28 October when the filing hit, alongside news of a $151m government contract.
Steelmanned, and this is strong: the academic literature is lopsided here. Insider purchases carry abnormal returns; insider sales largely do not. A non-plan, open-market buy that lifts the buyer’s own holding by a third is close to the cleanest version of the signal that exists.
Contrarian read
It is probably not a valuation call, because the price path is the wrong shape for one. She acquired shares at $189.095 on 5 September and $196.755 on 5 October, then bought 12,200 at $245–246 on 26 October. She bought size after a roughly 25–30% advance from where she had most recently been transacting.
A person acting on cheapness buys more when it is cheaper. The variable that changed between early October and 26 October was not price — it was the information set and, plausibly, what she was permitted and willing to do with it. That makes this a bet on a known forward number, or a deliberate disclosure act, rather than a bargain being taken.
Evidence that discriminates — and where it runs out
- The price sequence is the one hard fact, and it cuts against pure cheapness. $189.095 → $196.755 → $245–246, with the large order last. That ordering is checkable on three Form 4s and is not compatible with a straightforward “the shares are too cheap” story.
- But the September and October acquisitions do not prove she declined to buy. This is the honest limit, and it has to be stated: both were code A purchases from the company under a standing Voluntary Equity Investment Program election, not discretionary orders. They show her standing commitment was about 200 shares a month. They do not show she looked at $189 and passed. Treating them as a revealed preference would be exactly the overreach this report exists to avoid.
- The leverage of the act is measurable, and it is extreme. $3.0m is roughly 94% of all insider buying at Accenture over three months ($3.20m bought against $3.76m sold). It was disclosed two days later and the shares moved several per cent. Whatever she intended, the purchase was a very cheap way to move several billion dollars of market capitalisation — and it landed days after the chief accounting officer sold on 23 October and the chief operating officer sold on 14 October, into a week when the stock had fallen about 5.7%.
Three live hypotheses, and what would separate them. This is a case where the true reason is not knowable from the documents, so no claim is being made.
- Valuation. She believes the shares are worth materially more. Distinguishing test: she buys again if the price falls back below roughly $230. A cheapness buyer adds on weakness.
- Signalling / stabilisation. The purchase is primarily a disclosure, timed to answer a week of insider selling and a falling share price. Distinguishing test: no further purchases at any price, and no repeat once the narrative settles.
- Guideline compliance. She was below an internal share-ownership requirement and topped up. Distinguishing test: the next proxy statement’s stock-ownership guidelines table, which states the multiple required and whether she met it. This one is directly checkable in a public document in a few months.
What would falsify me
The contrarian claim here is narrow — that this is not best explained as a cheapness call. A second open-market purchase at a materially lower price would falsify it outright. That is the clean test and it requires no interpretation.
Equally: if the next proxy shows her comfortably above her ownership guideline both before and after, hypothesis 3 dies and the field narrows to two, which strengthens rather than weakens the consensus case.
Pattern vs her own history
In her recent filing record, acquisitions run at 197–205 shares a month through a company programme. This is 12,200 shares on the open market — roughly 62× the 5 October line — under a different transaction code, with no plan reported. Derived, not filed: her direct holding was 17,196 on 5 October and 36,710 immediately before the 26 October purchase, so about 19,514 shares arrived in between with no reported purchase — almost certainly award settlement or vesting. Worth noting because it means her personal cost base in a large slice of the position is nil, which makes a $3m cash outlay a genuinely different act from holding what she was given.
Attack the measurement: every “who is buying the midterms” table this week is missing its most expensive 41 hours
Elon Musk · America PAC · and the FEC filing system
Single source · Reuters, carried by Jerusalem Post
Musk figures via Forbes reading FEC
Outage 26–28 Oct 2026
Measurement critique
- Who and what
- Elon Musk made three direct contributions to America PAC (FEC committee C00879510) totalling about $74.8m between 6 and 14 October, plus about $1.98m through United States of America Inc. The filings became public around 22 October. Cumulative cycle giving is reported at about $443m, against $288m across the whole 2024 presidential cycle. America PAC spent about $42.7m in the first two weeks of October and held $46.5m in cash at 14 October.
- The event in window
- The FEC’s electronic filing system went down just after 22:00 ET on Monday 26 October after an attempted breach, and was restored on the night of Wednesday 28 October — about 41 hours. During the outage, committees could not file mandatory 24- and 48-hour independent-expenditure reports. Outside groups had been spending an average of more than $50m a day in the preceding week; more than $250m was reported on 20 October alone. Election day is 3 November.
Consensus read
Two stories, both straightforward. One: Musk is buying the midterms, outspending his own record presidential effort by more than half again. Two: the FEC had an unfortunate IT incident, contained quickly, and the filings were caught up afterwards — regrettable but not consequential, since nothing stopped voters voting.
Steelmanned: $443m from one individual is a genuinely extraordinary number and deserves the coverage it gets; and the agency did take the system down deliberately, which is the correct trade of a known outage against an unknown risk.
Contrarian read
The second story invalidates the first. The 41-hour hole sits precisely where the marginal, decisive money lives — the last week — and it is the 24- and 48-hour reports, the only near-real-time disclosure that exists, which went unfiled. At the prior week’s run-rate the gap covers on the order of $85m of outside spending that was invisible while it was being spent.
Separately, the $443m statistic has a sample problem independent of the outage: it counts disclosed contributions to disclosing vehicles. Spending routed through 501(c)(4) organisations never enters it. The number is described as measuring influence and is in fact measuring the visible subset of it.
Evidence that discriminates
- The outage window versus the spending curve. This is the crux. A 41-hour outage in June would be a footnote. A 41-hour outage starting 26 October, eight days before the election, lands on the steepest part of the spending curve — $250m reported on a single day the previous week, $50m-plus daily averages. The timing is what converts an IT incident into a measurement failure, and timing is a fact, not an interpretation.
- Which reports were blocked, specifically. Not the quarterlies — the 24- and 48-hour independent-expenditure reports. Those exist for exactly one purpose: to make late money visible while it can still be reported on. Blocking only the near-real-time layer is the maximally damaging failure mode for anyone computing a current league table, and it is the layer the week’s stories were built on.
- America PAC’s own balance sheet separates contribution totals from spending. $46.5m cash at 14 October against a $443m cumulative headline shows how much of that headline is historical rather than prospective. The commonly quoted figure is a cumulative inflow to a vehicle, not a measure of money in the field in the final week — a denominator that is not the choice variable the stories describe.
What does not discriminate: the size of Musk’s giving relative to 2024, the fact that he leads the donor table, and the polling in contested races. None of those speak to whether the measurement is sound, and all of them are being used as though they do.
What would falsify me
The catch-up filings themselves. After restoration on 28 October, the delayed 24- and 48-hour reports had to be filed. If those reports show that spending during 26–28 October was small relative to the $50m-a-day run-rate, then the hole was immaterial, the league tables stand, and this entry is wrong. That is a directly observable count in the FEC’s own database within days.
And on the second point: a credible estimate showing 501(c)(4) spending is a small fraction of disclosed super PAC spending this cycle would make the sample objection academic.
Pattern vs his own history
$288m across the entire 2024 presidential cycle, of which $239m went to America PAC; about $443m this midterm cycle, roughly 1.54× the presidential effort, in a non-presidential year. The escalation is real and is not what is being disputed here. Recorded as single-source: the contribution figures come from Forbes’s reading of FEC filings and the outage account from a Reuters report carried by the Jerusalem Post. Neither was verified against the FEC database directly from this scan, and the FEC figure for cumulative giving blends a New York Times tally through September with the October filings.