How every number is computed.
Every number on the site traces back to a filing. Each rule sits next to a real example from the current season, so you can check the rule against the product itself.
The source
Everything comes from SEC EDGAR: the quarterly Form 13F holdings report (13F-HR), its amendments (13F-HR/A) and notices (13F-NT). A restatement replaces the original. A new-holdings amendment extends it. A notice says the filer reported nothing itself. Every document page links the filings it was built from.
The SEC publishes its own structured data set months after each quarter. We reconcile our read against it when it lands.
What a 13F shows, and hides
A 13F is a photo of US-listed long positions as of a quarter end, filed up to 45 days later. It does not show shorts, cash, non-US listings, private holdings or the price paid. Options appear at the notional value of the underlying.
We call what it shows the US stock: the US-listed long stock the filing reports. Never "the portfolio", because the filing cannot see the whole portfolio. Both dates are always visible: as of, and filed.
the whole of a 13F, by instrument
the rest of a manager's book
The four objects
A manager is the firm, the legal author of the filings. People appear as dated roles on the firm (portfolio manager, CEO, founder), edited by hand. The face on a document is the role current at that document's as-of date.
A filing is one manager's 13F for one quarter, with its effective amendments merged. A season is every filing for one quarter across the managers. A stock is one security across managers, identified by its share class, since tickers and CUSIPs change. When a company changes its CUSIP (Honeywell in 2026), the position under the new CUSIP is the same position: not an exit and a new buy.
Picture
Coverage is our own measure, not a regulatory or academic standard. It is the US-listed long stock we can read as a share of everything the filing discloses (stock, options notional and notes). "US-listed" means listed on a US exchange as of the filing's period, so a Swiss company listed on the NYSE counts, and a security we cannot identify does not. Bands: fragment below 40%, partial 40 to 70%, mostly 70 to 90%, full from 90%.
The band is a rule of the interface. Every document and the directory say the band in words; where a table shows one manager's points next to another's, a partial picture is marked partial. Those points describe only the disclosed part of a book, so they do not compare with a full one. A fragment is not a bad manager.
Two real filings, Q2 2026: the fullest picture of the season, AKO Capital, and the thinnest, Check Capital.
Picture
Picture
Weights
A position's weight is its value divided by the US stock total of that filing. Options are weighted against their own notional total, and notes (convertibles, debt) against theirs. Neither is ever summed into the stock. Older filings that reported values in thousands are normalized to dollars, detected per filing by the implied share price.
Moves
A position's weight can change while the manager does nothing, because prices move. We only call something a move when the share count changed, and share counts are split-adjusted first: a manager flat across a 50:1 split shows no move. The four verbs are New, Added, Trimmed and Exited, and they are the only ones we use. Points ("+2.6 pp") are the change in that manager's US stock: the weight change for added and trimmed, the position's weight for new, minus it for exited.
A quarter-over-quarter change is the net of everything between two photos. "Added 45%" may have been ten trades, or a buy and a partial sell. We say moves, never trades.
A real filing, Q2 2026: First Eagle.
Moves
Concentration and turnover
Concentration is the effective number of positions (1 divided by the sum of squared weights). A book with one 60% position and nineteen small ones behaves like about two. The bands are ours, anchored on two public points: Morningstar calls a fund with fewer than 30 holdings concentrated, and the S&P 500 behaved like about 46 equal-sized stocks at the end of 2024, its lowest in 55 years. Very concentrated below 10, concentrated 10 to 20, focused 20 to 40, diversified from 40.
Turnover follows the churn-rate method of Gaspar, Massa and Matos (2005): the smaller of buys and sells between two filings, divided by the US stock value, scaled to a year. It is not the mutual-fund turnover ratio, which counts a full year of trades. Morningstar reports 50 to 70% a year as typical for active US stock funds. Low below 25%, moderate 25 to 75%, high from 75%.
Two real filings, Q2 2026.
The book
- Very concentrated
- Concentrated
- Focused
- Diversified
- Low
- Moderate
- High
The book
- Very concentrated
- Concentrated
- Focused
- Diversified
- Low
- Moderate
- High
Top 5 positions
The five largest positions by weight. In the academic record, managers' highest-conviction positions have beaten the rest of their own books by roughly 3 to 4% a year (Cohen, Polk and Silli 2010). It is the one single-filing read with documented edge, which is why it leads the document. A hedge, a legacy holding or an index fund can rank there too. Weight is size, not proof of conviction.
Tenure and streaks
Tenure is the number of consecutive quarters a name has been in a book, counted back from a filing over calendar quarters. A quarter with no filing breaks it, because we cannot see what the book held that quarter; a quarter the name is absent breaks it too. A streak is a tenure still running; it ends in the quarter the name leaves. On a document the count is per share class; on a stock a manager's tenure unions the classes, so a swap from one class to another of the same company is not a break: a stock is one issuer and one number, whatever its classes (ADR-013, ADR-014).
A + after a tenure, as in 53+ quarters, means the name was already held when our record begins, so it has been held at least that long. The record has three depths: a manager's filings we have read, a stock's quarters since any manager on the roster first held it, and the season's quarters since the XML filings of mid-2013. Our window starts there; many of these names were held for years before it. Tenure is a label like Picture, Concentration and Turnover: a measure of the record, not a rating, and never a ranking of managers.
The season
The season counts the filings in. k of N is always the denominator, and nothing is hidden while it is partial. The managers that file last tend to be the concentrated ones, so shared counts move late. A season is complete when every manager has filed or has no filing for the quarter, with the reason beside it (deregistered, before its first 13F, below the threshold, or a notice). Never a silent gap. Once complete and past the deadline it stays a record. Later amendments are listed against it, never applied quietly.
Shared moves groups moves by stock and direction. A stock is shown when at least two managers, or 5% of the filers, moved it the same way. That is a display threshold, not a signal. Where the managers disagree is the same stock moved in opposite directions, counting a manager only when its move changed the weight by at least 0.05 pp. Stock names everywhere come from our security master, one spelling for every filer. The name as filed stays in the EDGAR filing each document links. Across the managers lists stocks no manager held last quarter that at least one holds now, and the reverse. Most widely held is crowding, not a signal: the research finds widely held stocks do not outperform (Chen, Jegadeesh and Wermers 2000). Options never enter any of these counts.
Amendments
When a quarter has amendments, the document says what each one did. A new-holdings amendment reveals positions, often after confidential treatment. A restatement corrects rows and moves the total. History carries revealed positions back to the quarter they belong to, so they are never shown as a new buy in the quarter they surfaced. We show the change, never the intent.
Sectors
Sectors come from the SEC's industry code of each issuer, grouped by a rule we wrote and publish (for example, code 28 splits into industrial gases, pharmaceuticals and consumer products by its third digit). It is the only grouping we do. No themes, no narratives.
The managers we read
Admission is a written rule, not taste: a 13F history of eight or more quarters, US stock we can read (coverage at least partial), and a discretionary manager. No index funds, custodians or quant shops. Each manager's page says when it was admitted. A firm that files under several SEC ids rolls up into one manager (Berkshire files through several). Duplicate co-filings are detected and flagged, never summed twice.
Data quality
Every pipeline run publishes its own defect log. Right now: 4964 filings published across 98 managers, 5000 quarantined items. A filing that disagrees with its own cover-page totals is quarantined, visible on the season and the manager, never silently skipped. A row we cannot identify is left out of its document and listed on the document's record. Parsed totals are reconciled against the SEC's own structured data set when it is published. Of the US stock value in the latest filings, 100% sits in stocks we identified. The rest render as "unresolved", never as a blank.
Our own restatements
We ask filers to show what an amendment changed. We hold ourselves to the same rule. When our method changes a number that was already published, the change is dated here, with what moved and why.
Why no returns
We do not reconstruct a return. A 13F cannot produce an honest one: it omits shorts, cash, non-US and private holdings, reports options at notional rather than at cost, lands up to 45 days late, and hides every trade inside the quarter. We show what the documents show.




