Consensus Sheet

13f.darren-zhu.com — what 34 well-known investment managers own, and what they changed last quarter, from their SEC filings.

Questions it answers:

• Which stocks did several managers buy in the same quarter? The Patterns page ranks them, and every row names the managers behind it.
• Who owns NVIDIA, and did they add or trim? Any stock page lists its holders with each one's weight and share change.
• Has an activist just taken a stake in a company? The Ownership page tracks Schedule 13D and 13G filings as they land.
• Are officers or directors buying their own stock? The Insiders page tracks Form 4 transactions for issuers a tracked manager holds.

Everything is derived from public filings and recomputed from scratch on every run. How each number is defined, and what it does not cover, is in docs/METHODOLOGY.md — read that before trusting a figure.

This is not investment advice.

Finding your way around

Four things the site is built to do:

1. Find a crowded stock. Patterns → Consensus Buys. Open a row's Buyers to see which managers, and its Score to see the arithmetic behind the ranking.
2. Inspect who holds it. Click the symbol. The stock page lists every tracked holder, its weight, and whether it added or trimmed. It also shows 13D/13G stakes and insider trades. A timeline puts insider trades next to the quarter-end price the 13F filings imply. A "Sellers appearing" strip shows whether managers, big holders, or insiders are selling.
3. Compare what changed. A manager's page shows its whole reported book, quarter over quarter, with sector exposure and its most similar managers. A position the manager also holds puts or calls on gets a badge.
4. Verify a filing. Every manager-quarter links its source filings on EDGAR by accession, so any number here can be traced back to the document it came from.

Stock and manager links keep the selected quarter. You can pick an older quarter on a stock page. If a quarter has no data, the page says so.

Every research table can be exported to CSV. The export has all the table's rows. Each row also carries the period, the source filings, how many managers filed, and the methodology version. Ranked tables keep their top-N cut.

The Managers page also ranks each manager by how easy its 13F is to copy. A manager that trades rarely and holds a few large positions is easier to copy. This says nothing about returns.

Filter the research universe

On Patterns, open Research universe to choose a style, individual managers, and the minimum manager count. Every table recomputes for that selection; the URL preserves your selection and exports include it. Stock and manager links still show their full reported holdings. Default views read precomputed results; custom subsets aggregate the published manager-quarter data locally.

Watchlist

Use "Watch this name" on a stock or manager page, then open Watchlist. Saving establishes a baseline; later visits show new quarterly reports, revisions, and methodology recalculations with filing links. Use "Check latest data" to reload the newest dataset. The list and digest live only in this browser; clearing browser storage removes them. If storage is blocked, the list works for the current session.

What this is

This site shows what big investors own. The data comes from SEC Form 13F filings. It also finds patterns across managers, like which stocks many of them are buying at the same time.

Coverage and limits

• 34 managers, listed below, chosen by hand for being well known and running concentrated books. Adding or removing one changes every count and average on the site. "Consensus" always means consensus among this list, not the market.
• 12 quarters of history. Older filings reported values in thousands rather than dollars, which would need separate handling.
• 13F holdings update monthly, on the 16th. A quarter's filings are not due until 45 days after it ends.
• 13D/13G ownership updates daily, from 2024-12-18 onward. That is when the SEC's structured format became mandatory.
• Form 4 insider trades update daily, from 2025-09-01 onward. They cover only companies a tracked manager holds.
• Long US-listed equity positions only. No shorts, cash, bonds, foreign listings, or private holdings. For an endowment, most of the real portfolio is invisible here.
• A missing filing is shown as missing, never as a manager holding nothing.

Full definitions and the current open limitations: docs/METHODOLOGY.md.

What 13F data is (and is not)

A 13F is a report. Large US investment managers must file it with the SEC every quarter.

• It shows long positions in US stocks. A long position means the manager owns the stock.
• It does not show short positions. A short position is a bet that a stock will fall.
• It does not show cash, bonds, or most foreign stocks.
• It can be filed up to 45 days after the quarter ends. So the data is always a little old.
• It includes options. A put option is not the same as a short, and can be a hedge. The site labels them "Reported Put Exposure" and keeps them out of every weight, because an option row reports the value of the underlying shares, not what was paid.
• Values are in US dollars.
• This site is not investment advice.

Managers tracked

Manager	Person	Style label
Berkshire Hathaway	Warren Buffett	Value / Quality
Pershing Square	Bill Ackman	Activist / Quality
TCI Fund Management	Chris Hohn	Quality / Compounders
Baupost Group	Seth Klarman	Value / Event Driven
Appaloosa	David Tepper	Macro / Value
Duquesne Family Office	Stanley Druckenmiller	Macro
Coatue Management	Philippe Laffont	Tech / Growth
Tiger Global	Chase Coleman	Tech / Growth
Viking Global	Andreas Halvorsen	Tech / Growth
Lone Pine Capital	Steve Mandel	Tech / Growth
Third Point	Dan Loeb	Event Driven / Activist
Markel	Tom Gayner	Value / Quality
Fairfax Financial	Prem Watsa	Value / Quality
Ruane, Cunniff & Goldfarb (Sequoia Fund)	—	Value / Quality
Davis Selected Advisers	Christopher Davis	Value / Quality
Elliott Investment Management	Paul Singer	Activist
Carl Icahn	Carl Icahn	Activist
Starboard Value	Jeffrey Smith	Activist
Trian Fund Management	Nelson Peltz	Activist
ValueAct	Mason Morfit	Activist
JANA Partners	Barry Rosenstein	Activist
Bridgewater Associates	Ray Dalio	Macro
Soros Fund Management	George Soros	Macro
Altimeter Capital	Brad Gerstner	Tech / Growth
Whale Rock Capital	Alex Sacerdote	Tech / Growth
D1 Capital Partners	Dan Sundheim	Tech / Growth
Situational Awareness	Leopold Aschenbrenner	Tech / Growth
Paulson & Co	John Paulson	Event Driven
Farallon Capital	—	Event Driven
Davidson Kempner	—	Event Driven
Akre Capital Management	Chuck Akre	Quality / Compounders
Polen Capital	Dan Davidowitz	Quality / Compounders
Harvard Management Co	—	Endowment
MIT	—	Endowment

13F filings from endowments only cover their sliver of US public equities. Most of an endowment's assets sit in private equity, hedge funds, and other holdings a 13F never reports, so these rows are a much smaller slice of the real portfolio than a fund like Berkshire's.

Yale was tracked until September 2026 and has been dropped. A manager only has to file a 13F if it holds more than $100 million of US-listed stock. Yale's directly held stock is worth about $2 million, so it stopped filing after the third quarter of 2025 and there is nothing left to show.

The style labels are set by hand. You can change them in ingest/funds.json.

Ownership filings (13D and 13G)

The site also tracks two other SEC filings: Schedule 13D and Schedule 13G.

Either one is filed by an investor who owns more than 5% of a class of a company's shares.

Which form you file depends on whether you are eligible for the shorter one.

A 13G is the short form. Three kinds of investor may use it. Large institutions buying in the ordinary course of business. Passive investors who hold under 20% and do not seek control. And investors exempt from the rules for other reasons. All three must lack any intent to change or influence control of the company.

A 13D is the long form. It is what you file when you are not eligible for a 13G. That includes investors who do intend to influence the company. It also includes investors who simply hold too much, or who lost their 13G eligibility for a technical reason.

So a 13D is worth reading, but it is not proof of an activist campaign. And a 13G filer is not always a quiet small holder. The largest index funds file 13Gs on hundreds of companies.

The deadlines differ too. An initial 13D is due within 5 business days of crossing the threshold. A 13D amendment is due within 2 business days of a material change. A 13G deadline depends on the filer's category and on what triggered the filing. A passive investor's first 13G is due in 5 business days. A large institution's is due 45 days after the end of the quarter. A 13F, by contrast, can take up to 45 days after the quarter ends.

The SEC's own summary of these rules is in its beneficial ownership fact sheet.

Ownership events start on December 18, 2024. Older filings are not structured data, so we skip them.

Each new filing becomes one of these events:

• New. A first filing on a stake above 5%.
• Increased. The stake grew by a meaningful amount.
• Decreased. The stake shrank by a meaningful amount.
• Below 5%. The stake fell under the 5% reporting threshold. This is not the same as selling out. The investor may still hold 4.9%, which no longer has to be reported here.
• Switched. The investor moved from a 13G to a 13D, or the other way.
• Updated. Something else changed, like the filing's stated purpose.

Each event also shows how many tracked managers already held the stock. That number comes from the last 13F quarter, so it is always older than the filing next to it. Zero means none of them held it. A dash means the 13F side has not run yet.

Insider transactions (Form 4)

The Insiders page tracks Form 4 filings — the SEC form officers, directors, and 10%+ owners file when they buy or sell their own company's stock. It only covers issuers a tracked manager holds, since the SEC gets around 500 of these a day and fetching every one of them for every public company is not something this site's budget covers. Every transaction is classified by its SEC code, never by edgartools' own label: a stock award or option exercise is never shown as a purchase, and tax withholding or a gift is never shown as a sale. Sales split into Planned (a pre-scheduled Rule 10b5-1 trade) and Discretionary, with "not stated" kept as its own honest answer when a filing does not say. See docs/METHODOLOGY.md for the full code table.

The signals

The site computes these signals once per quarter.

1. Manager Conviction. How big each stock is inside a manager's reported equity holdings, and how that changed since last quarter. Option, note and warrant rows are excluded from every weight.
2. Stock Consensus. How many managers own a stock, and how much of their portfolio it is.
3. Consensus Buys. Stocks that three or more managers bought or added in the same quarter.
4. Consensus Exits. Stocks that three or more managers sold or trimmed in the same quarter.
5. High-Conviction Overlap. Stocks that three or more managers each hold at 3% or more of their reported equity holdings.
6. Conviction Score. A score from 0 to 100. It rewards stocks that a few managers hold in big size and just bought. It is relative within one quarter — 100 is that quarter's top score, not a rating — so scores are not comparable across quarters.
7. Sector Exposure. How much of each manager's portfolio is in each sector, and how that changed.
8. Sector Rotation. Which sectors managers are moving into or out of, as a group.
9. Manager Similarity. How alike two managers' portfolios are, from 0 to 1.
10. Manager Clusters. Managers grouped by style, like "Tech / Growth".
11. Ownership Change. How the number of managers holding a stock changed over time.
12. Position-Weight Trend. How the average portfolio weight of a stock changed over time.
13. Put / Call Exposure. Which managers report puts or calls on a stock. This is kept separate from stock holdings.
14. Copyability. How well a copy of a manager's last 13F can track it. It uses turnover, how concentrated the book is, and how many new positions are still held a year later. It is not a measure of returns.
15. Implied Quarter-End Price. The price the filings imply: each holder's reported value divided by its shares, then the median. A 13F carries no other price.

How it works

GitHub Actions (once a month, or by hand)
  └─ ingest/ingest.py (Python)
       ├─ fetch   : SEC EDGAR → last 12 quarters of 13F filings per manager
       ├─ enrich  : CUSIP → FIGI (OpenFIGI) → ticker, issuer, sector (security-master)
       ├─ derive  : every signal listed above
       └─ store   : Google Cloud Storage (files) + Firestore (documents the site reads)

GitHub Actions (once a day, or by hand)
  └─ ingest/ownership.py (Python)
       ├─ fetch   : SEC EDGAR → new Schedule 13D/13G filings
       ├─ derive  : new / increased / decreased / exited / switched / updated events
       └─ store   : Google Cloud Storage (archive) + Firestore (documents the site reads)

GitHub Actions (once a day, or by hand)
  └─ ingest/insider.py (Python)
       ├─ fetch   : SEC EDGAR → new Form 4/4-A filings, issuers a tracked manager holds
       ├─ derive  : buy / sell / award / exercise / tax / gift / conversion, priority, clusters
       └─ store   : Google Cloud Storage (archive) + Firestore (documents the site reads)

GitHub Actions (on every push to main)
  └─ build the site → Firebase Hosting → your domain

A script runs once a month. It downloads the latest filings and computes every signal. It writes the results to Firestore as a new snapshot. The site switches to it only after every document is written. The newest 3 snapshots are kept, so a page left open during an update still works. Older ones are deleted.

The website only reads and displays the results. Default views need no live computation. Custom manager and style filters recompute their selection in the browser.

Each page reads a few whole documents. It never runs a query or an aggregation. Every page reads meta/latest. On top of that:

• Patterns, Ownership, and Insiders read one more document each.
• A manager page reads three more: the manager, its quarter, and its 13D/13G filings.
• A stock page reads five more: the stock, its quarter, its 13D/13G filings, its insider trades, and the short list of stocks with cluster buying.
• The search box loads its symbol list once, the first time you focus it.

A second script runs once a day. It checks for new 13D and 13G filings and turns each one into an event. You can see them on the Ownership page. They also show up on a stock's own page. Every investor gets their own page too — a tracked manager's page, or /investor/:cik for everyone else.

A third script also runs once a day. It checks for new Form 4 filings on issuers a tracked manager holds and classifies each transaction line. You can see them on the Insiders page, on a stock's own page, and every insider gets their own page at /insider/:cik.

Each ingest run also saves a small file, data/last_ingest.json, into the repo. It shows when the data was last updated. It also keeps the schedule alive. GitHub turns off schedules in repos with no activity for 60 days.

For diagrams of the full system and the data pipeline, see docs/ARCHITECTURE.md.

Set up your own copy

You need a Google account and a GitHub account. Some values are secret. Never put a secret in the code.

1. Go to the Firebase console. Create a project. Turn on Firestore (Native mode) and Hosting.
2. In Project settings, add a Web app. Copy the API key, project ID, and app ID. These three values are public. Put them in web/.env and in your GitHub repository Variables.
3. Upgrade the project to the Blaze plan. This turns on billing. Normal use stays inside the free tier. Set a budget alert at $5 to be safe. Create a Cloud Storage bucket. Put its name in the GCS_BUCKET variable.
4. In Google Cloud IAM, create a service account. Give it the Firebase Admin and Storage Object Admin roles. Create a JSON key. This key is secret. Paste the whole file into a GitHub secret named FIREBASE_SERVICE_ACCOUNT. Keep a copy outside the repo for local runs.
5. Get a free API key from openfigi.com. This is secret. Save it as the GitHub secret OPENFIGI_API_KEY.
6. The SEC asks for your name and email on every request. Save Your Name your@email.com as the GitHub secret EDGAR_IDENTITY. This is secret because it is your email.
7. Create a public GitHub repository. Push the code. Add the secrets and variables under Settings → Secrets and variables → Actions.
8. In the Firebase console, open Hosting. Add your custom domain. Add the DNS records it shows you at your domain registrar.
9. Run npx firebase-tools login once. Then run npx firebase-tools deploy --only firestore:rules. This makes the data readable by the site.

Run locally

Requires Python 3.12 and Node 24.22.1 or newer. Every command below says which directory to run it in.

1. Preview the site against live data

Only needs the three public VITE_FIREBASE_* values, so this is the quickest way to see the app.

cd web
npm install
cp .env.example .env        # PowerShell: Copy-Item .env.example .env
# fill in .env, then:
npm run dev

2. Run the 13F ingest

Needs EDGAR_IDENTITY, OPENFIGI_API_KEY, and a service-account key outside the repo, pointed at by GOOGLE_APPLICATION_CREDENTIALS.

cd ingest
python3.12 -m venv .venv    # PowerShell: py -3.12 -m venv .venv
source .venv/bin/activate   # PowerShell: .\.venv\Scripts\Activate.ps1
pip install -r requirements.txt
cp .env.example .env        # PowerShell: Copy-Item .env.example .env
# fill in .env, then:
python ingest.py --dry-run

--dry-run downloads and computes everything, prints a summary, and writes nothing: no Firestore documents, no securities/ cache entries, no GCS files. Drop the flag to write for real. A real run rewrites every quarter in the window, so it is safe to repeat.

Useful flags: --fund CIK --dry-run to preview one manager, --quarters N to shorten the window, --refresh all to re-resolve cached tickers and sectors after security-master changes.

3. Backfill 13D/13G ownership

GCS_BUCKET is required here. This pipeline keeps its state in the bucket and has no local fallback. Each run picks up a few days before the newest filing it already has. The first run has nothing yet, so it starts at start_date in signals_config.json (2024-12-18).

cd ingest
python ownership.py --dry-run --since 2025-01-01 --until 2025-03-31

Drop --dry-run to write. Backfill a long window in slices, a quarter at a time, not in one run. Use --rebuild only when you need every issuer and investor document rewritten.

4. Backfill Form 4 insider trades

This also needs GCS_BUCKET. It needs a published 13F run too, because it only covers companies a tracked manager holds. Run step 2 for real first.

cd ingest
python insider.py --dry-run --since 2025-09-01 --until 2025-09-30

The first run starts at start_date in the insider block of signals_config.json (2025-09-01). The same slicing and --rebuild advice applies.

5. Validate before pushing

From the repository root:

(cd ingest && pytest && ruff format . && ruff check .)
(cd web && npm run test && npm run build && npm run lint)

6. Deploy

Pushing to main builds the site and deploys it to Firebase Hosting. Firestore rules are separate and deploy from the repo root:

npx firebase-tools deploy --only firestore:rules

There is no sample-data preview: the app reads Firestore directly, so seeing it with data means pointing it at a project that has some. Step 1 against the live project is the closest thing.

Add a manager

1. Find the manager's CIK number on the SEC EDGAR site. The CIK is the SEC's ID for a filer.
2. Add a line to ingest/funds.json with the CIK, name, short name, and style label.
3. Run the ingest workflow from the GitHub Actions tab.

Tuning the signals

Edit ingest/signals_config.json. Then run the ingest workflow.

• quarters — how many quarters to load. Default 12 (three years). Do not go past 12: filings for quarters before 2023 report dollar values in thousands, so they would read 1000 times too small.
• consensus_min_managers — how many managers make a "consensus". Default 3. With 34 managers tracked, 2 matched most of the market and the tables stopped meaning anything.
• high_conviction_min_weight — the share of reported equity holdings that counts as high conviction. Default 0.03 (3%).
• high_conviction_min_managers — how many managers make a high-conviction overlap. Default 3.
• sector_move_threshold — the sector weight change that counts as a move. Default 0.005 (0.5 points).
• top_n — how many rows each ranked table keeps. Default 25.
• score — the constants inside the Conviction Score formula. See docs/METHODOLOGY.md for the formula.
• ownership and insider — the settings for the two daily pipelines, like start dates and event thresholds. After changing a threshold, run that pipeline once with --rebuild so every page is rewritten.

Sector data

13F filings do not include a ticker or a sector. Both come from security-master, an open dataset of US-listed stocks and funds. A filing names each holding by its CUSIP. OpenFIGI turns the CUSIP into a FIGI, another ID for the same security, and security-master looks the FIGI up.

The sector comes from the company's SEC industry code (SIC). security-master maps that code to a sector with its own table. SIC is a filing code, not a finance one, so some companies land in an odd sector. Tickers are in the SEC's form, so Berkshire's B shares are BRK-B.

Funds are the exception. ETFs and closed-end funds are labelled "ETF / Fund". A fund's own industry code says "investment offices", which would put an S&P 500 fund in the Financials sector.

security-master starts on 2026-10-19. A stock that was bought out or delisted before then is not in it. Those holdings keep the ticker and sector they were first given.

Lookups are cached per stock and never redone on their own. After security-master changes how it sorts stocks, run python ingest.py --refresh all once. It updates every stock security-master knows and leaves the older ones as they are. --refresh unknown is the cheaper version: it only redoes the stocks whose sector came out "Unknown".

Where the docs live

Document	For
This README	using the site, and running your own copy
docs/METHODOLOGY.md	what every number means, and what it does not cover
docs/ARCHITECTURE.md	diagrams of the system and the data pipeline
docs/PLAN.md	implementation history and outstanding work

License

MIT. See LICENSE.