Q4 2025 13F signal summary — 3,713 moves, 27 managers, the 9 patterns that mattered

The Q4 2025 13F filing window closed on February 17, 2026. It is now sixty-six days later — the filings are cold enough that sell-side research has already reacted, retail newsletters have already summarized the Buffett delta, and the prices those filings reported at (December 31, 2025) are three months in the past. That is precisely the right distance from which to compute honest cross-investor synthesis: late enough that noise has settled, early enough that the patterns still inform position-sizing going into the May 15 Q1 2026 deadline.

Across the 27 superinvestors who filed for Q4 2025, HoldLens normalized 3,713 discrete position changes. Every figure below is computed at build time from data/edgar-moves.json in the HoldLens repo — the same raw SEC 13F-HR XML the filers submitted, parsed by scripts/fetch-edgar-13f.ts, cross-referenced against their prior-quarter holdings to produce action-typed transitions (add / trim / new / exit).

The headline distribution

• 2,186 buys (adds + new positions)
• 1,527 sells (trims + full exits)
• 353 first-time positions — manager bought a ticker they held none of last quarter
• 296 complete exits — manager sold the entire position in a ticker they had held

Buy-to-sell ratio 1.43-to-1 is the strongest net-buying quarter in the HoldLens lookback. The textbook reading: superinvestors were deploying capital into Q4 2025's market rather than distributing — not panic, not euphoria, just positioning for what filers apparently believed was a constructive 2026.

One caveat that matters for the rest of this commentary: Joel Greenblatt alone accounted for 2,625 of the 3,713 moves. Greenblatt's Gotham Capital runs a high-turnover quantitative strategy (published hundreds of positions, quarterly rebalance). Filtering Greenblatt out, the remaining 26 managers made 1,088 moves — an average of 42 per filer, consistent with the concentrated-discretionary approach most HoldLens-tracked superinvestors run. Patterns below are cross-validated against the non-Greenblatt subset so a single high-turnover filer doesn't fabricate consensus.

Pattern 1 — Amazon split the room 6-to-7

Amazon (AMZN) was the most-disagreed mega-cap of the quarter. Thirteen of the 27 tracked filers had an AMZN transaction:

• 6 bought, including 3 entirely new positions — Andreas Halvorsen (Viking Global, +1.9% portfolio weight), David Rolfe (Wedgewood, +2.6%), Seth Klarman (Baupost, +9.3% — a materially large new conviction), plus size-increases from Bill Ackman (+65% to 14.3% weight, now his largest position), Stanley Druckenmiller (+92%), and Joel Greenblatt (+27%).
• 7 trimmed or exited — Warren Buffett (−77%, essentially closing the position), Chase Coleman (Tiger Global, −9%), David Tepper (−13%), Lee Ainslie (Maverick, −23%), Polen Capital (−23%), Stephen Mandel (Lone Pine, −14%), and Tom Slater (Baillie Gifford, −7%).

The ConvictionScore v5 composite model registers AMZN as a high-intensity divergence — the 6×7 buyers×sellers product of 42 is the largest divergence intensity of any mega-cap this quarter. When 13 of the world's most disciplined capital allocators cannot agree on a single ticker, the honest read is that AMZN at its December 31, 2025 price was trading close enough to fair value that bear and bull theses are both defensible — and correspondingly, that concentrated conviction in either direction deserves smaller position sizes than a unanimous signal would warrant.

Full per-investor view: /stock/AMZN/.

Pattern 2 — Buffett cut AMZN 77%, added CVX and Chubb, entered NYT

Berkshire Hathaway's Q4 filing showed 12 distinct moves. The three that matter most for reading Buffett's attention:

• AMZN −77%. Buffett initiated AMZN in Q1 2019 at ~$1,700/share (split-adjusted $85), held quietly through the post-COVID run, and has now effectively closed it to a residual 0.2% portfolio stub. This matches Berkshire's historical pattern of exiting positions gradually once the thesis no longer justifies the opportunity cost of the capital.
• CVX +7% and CB (Chubb) +9%. Both are consistent with Berkshire's durable preference for cash-generative, capital-disciplined insurance and energy operators. CB is now a 3.9% position — meaningful but far below the top five (AAPL 22.6%, AXP 20.5%, BAC 10.4%, KO 10.2%, CVX 7.2%).
• New positions: NYT and LLYVK. NYT (New York Times) is a 0.1% stake — small enough to be either a manager pick from Todd Combs or Ted Weschler, or a deliberate toe-in from Buffett himself. LLYVK (Liberty Live Group) similarly opens a small media/entertainment cluster.

What the filing does NOT show: any change to Berkshire's top-5 concentration. AAPL, AXP, BAC, KO, CVX together remain 70.9% of the $274B portfolio. The durable-compounder thesis is intact. Full breakdown: /investor/warren-buffett/.

Pattern 3 — Ackman's all-mega-cap quarter

Bill Ackman's Pershing Square filing showed a pronounced consolidation into mega-cap tech: AMZN +65% (now 14.3% weight, his largest position), NEW META (first-ever Pershing Square position in Meta), BN +50% (Brookfield), GOOGL −86% (effectively exited after riding the 2023-2025 AI-infrastructure narrative), CMG −100% (full Chipotle exit).

The Ackman-vs-Buffett mirror on AMZN is the most striking data point of the quarter: Buffett exiting Amazon at the same time Ackman accumulates. Two concentrated-conviction allocators, opposite directions, same ticker, same quarter. Neither is wrong — they are simply operating on different time horizons and different underwriting models. It is the purest expression of why 13F consensus matters more than any single filer's signal. Full per-investor view: /investor/bill-ackman/.

Pattern 4 — The Magnificent Seven are fragmenting

Beyond AMZN, three other Mag-7 names posted net-bearish filer balances in Q4:

• META — 4 buyers vs 7 sellers. Ackman's new entry is the largest single conviction; the seven trimmers collectively reduced exposure.
• GOOGL — 4 buyers vs 6 sellers. Ackman's −86% is the headline exit.
• TSM (Taiwan Semiconductor) — 3 buyers vs 6 sellers. The sharpest net-sell of the quarter among names this concentrated in superinvestor portfolios.

The pattern is not that mega-cap tech is collapsing — it is that the AI-infrastructure trade that dominated 2023-2024 is fragmenting into differentiated company-level theses. Filers who were long the whole basket through 2024 are now picking winners and rotating losers.

Pattern 5 — Three managers independently started MEDLINE

Medline Industries (MEDLINE), the medical-supply distributor that completed its long-awaited IPO in late 2025, showed up as a new position for three distinct filers: Andreas Halvorsen (Viking Global, 0.8% weight), Stephen Mandel (Lone Pine, 3.6% — the largest new conviction of the three), and Tom Slater (Baillie Gifford, 0.4%).

Three independent first-time positions in a recently-IPO'd name is a textbook consensus-buy signal at the cleanest signal-to-noise — none of these filers could have held MEDLINE pre-IPO, so the decision to size meaningfully is unambiguously a post-IPO thesis about the business, not a legacy position inherited from Mandel's or Slater's earlier vintage.

Pattern 6 — The insurance and payments consensus

Four tickers drew unanimous buying (≥3 buyers, zero sellers) and are worth noting because consensus at this density is uncommon:

• CB (Chubb) — 4 buyers, 2 new. Buffett added 9%, Rolfe and Halvorsen both opened new positions, Greenblatt added 52%. Insurance-book compounders re-rating after disciplined-underwriting years.
• PGR (Progressive) — 3 buyers, 2 new. Same thesis, different operator.
• TW (Tradeweb) — 3 buyers, total delta +426%. Financial-infrastructure pick.
• DRAFTKINGS — 3 buyers, 1 new. The lone high-beta name in the consensus set.

Pattern 7 — Intuit is the quietest consensus sell

INTU (Intuit) drew five different trimmers with zero offsetting buys: Halvorsen (full exit), Dev Kantesaria (−15%, a 4.2% portfolio reduction), Polen (−23%), Greenblatt (−29%), and Tom Slater (−6%). The ConvictionScore v5 composite reads this as a high-signal consensus sell — the opposite of MEDLINE's consensus buy, but with equal signal density.

Other notable consensus sells: APP (AppLovin) — 4 sellers post-2024-run; MCO (Moody's) — 4 sellers after years as a Buffett-class compounder; WDAY, CVNA, AMD, GM, POOL — each 3 sellers.

The software-and-rating-agency rotation is the clearest cluster within the consensus-sell set. INTU, WDAY, MCO, and APP together represent a filer-level vote that late-2025 valuations for enterprise-software and information-services compounders had run ahead of what the next twelve months of earnings will likely justify.

Pattern 8 — The quiet Coca-Cola exits

Three filers reduced KO this quarter: Bill Nygren (Oakmark, full exit), Polen Capital (full exit), Joel Greenblatt (−10%). Two full exits in a single quarter from a name as historically-held as Coca-Cola is unusual — KO has been a multi-decade Buffett anchor (Berkshire still holds 10.2% of its portfolio in KO and made no Q4 change).

This is one of the cleanest examples of why a single filer's signal is insufficient. Nygren and Polen exit. Buffett holds. Greenblatt trims. The ConvictionScore v5 composite reads this as a mild net-bearish signal — weighted by the quiet Buffett hold (non-action is signal), not overwhelmed by the two exits. See /stock/KO/ for the full per-filer view.

Pattern 9 — Where the new-position capital went

353 first-time positions across 27 filers is the richest source of forward-looking signal in the quarter. The top clusters:

• AMZN — 3 new (Halvorsen, Rolfe, Klarman). Counter to Buffett's exit.
• MEDLINE — 3 new (Halvorsen, Mandel, Slater). Post-IPO healthcare distribution thesis.
• TMO (Thermo Fisher) — 2 new. Life-sciences tools compounder.
• CB — 2 new. Part of the insurance-compounder rotation (see Pattern 6).
• RIOT (Riot Platforms) — 2 new. Cryptocurrency-infrastructure exposure; first time any HoldLens-tracked filer has taken direct Bitcoin-miner equity risk at this consensus density.
• IONQ — 2 new. Pure-play quantum-computing exposure; a high-risk, small-weight bet from two different managers independently.
• PGR — 2 new.

How the ConvictionScore v5 composite reads this quarter

Q4 2025 is one of the clearer quarters HoldLens has processed since launch. The composite model shipped on April 24, 2026 synthesizes nine signal layers — 13F positioning, 13F moves, insider Form 4 activity, 8-K material events, bankruptcy filings, enforcement actions, activist campaigns, proxy contests, and analyst divergence — into a signed −100…+100 score per ticker. For Q4 2025, the top of the ConvictionScore leaderboard is dominated by the names above: CB and Chubb-class insurance compounders at the high-positive end; INTU and the enterprise-software rotation at the high-negative end; AMZN in the middle with compressed net score but elevated divergence intensity.

See the live leaderboard at /today/, the per-ticker 9-layer drill-down at /signal/, and the full cross-investor divergence feed at /divergence/ (currently 238 divergence events across all tracked quarters, 36 in Q4 2025 alone).

What this commentary is NOT

This is a description of what the Q4 2025 filings contained, not a forecast of what the Q1 2026 filings will contain. The 45-day 13F-HR publication lag means every move described above was executed at prices from October through December 2025 — now nearly four months in the past. A manager who bought AMZN at $220 in November has a materially different position by late April than the filing implies.

Filers also only disclose long US-listed equity positions above the 13F-HR threshold. Short positions, options, international holdings, cash balances, and fixed-income allocations are entirely invisible. The filings describe a slice of each manager's book — the slice that moves markets most visibly — but never the full book.

Cross-checking this commentary against the next Q1 2026 filings (due by May 15, 2026) will reveal which of the patterns above were quarterly noise and which were the beginning of durable rotations. That check is what the Q1 2026 pre-wave primer was designed for.

Methodology + verification

Every figure in this report is computed at build time from data/edgar-moves.json and data/edgar-holdings.json, fetched via scripts/fetch-edgar-13f.ts directly from the SEC EDGAR full-text 13F-HR endpoint. No third-party aggregator. Every per-investor move links back to /investor/<slug>/ where the underlying filing accession number is surfaced and resolvable on sec.gov/edgar.

See /learn/conviction-score-explained/ for the full ConvictionScore v5 nine-layer formula, /methodology/ for the broader data pipeline, and /disclaimer/ for the filing-lag + not-investment-advice framing. Machine-readable twins of this report's underlying data at /api/v1/moves.json and /api/v1/holdings.json.