GetCoattail republishes what large investors, corporate insiders and members of Congress report in public filings, turns those filings into screens and ranked lists, and — the part most sites skip — publishes how each signal actually performed afterwards. Nothing on this site is a recommendation.
13F holdings: SEC EDGAR Form 13F filings for the 149 tracked funds (147 of them with a filing for the latest quarter), collected from the SEC’s official data. Insider trades: SEC Form 4. Politician trades: U.S. House Clerk STOCK Act periodic transaction reports (official PDFs) and U.S. Senate eFD reports (electronic filings, added Oct 2026). Financial screens and red flags: SEC company financial statements (companyfacts) and Form 8-K filings. Macro regime: FRED. Prices and charts: Yahoo Finance daily bars. Raw filings keep their accession numbers / document IDs in the pipeline; derived pages cite the filing and its dates.
One menu, defined once. The site’s navigation — Find, Timing, Stocks, Funds, Politicians, Research, Track Record, and the Tools group (Overlap Checker, Game, Practice timing, Follow a Fund, 13F season tracker) — is defined in a single shared menu that the home app, the Find page and every static page read at build time. Home is reached through the logo. Methodology lives in the footer, not the menu. Before any build ships, an automated gate compares the menu on six sample pages (home, Find, one stock page, one fund page, the research index and one research article) and fails the build if they differ.
13F filings for the quarter ended September 30, 2026 are due Monday, November 16, 2026, by 5:30 p.m. Eastern. A fund that has filed is marked “Q3 filed Nov 6” — the filing date, month and day — under its fund-page title and on the matching row of a stock page’s holdings table, linked to the 13F season tracker (/13f-q3-2026). The home page shows the season strip directly below the search box: how many of the 149 tracked funds have filed, how many days remain until the deadline, and the three most recent filers. The strip removes itself the day after the deadline. Filing status comes from the pipeline’s Q3 watch (EDGAR), not from the holdings files, so a fund can be marked filed before its holdings are merged into the pages.
The counts on the home page are computed by the build from the pipeline’s own data files, each with its own as-of date, and are regenerated on every build — they are never typed into the template. Tracked funds: 149, with the count filed for the latest complete quarter shown beside it. Form 4 filings parsed: the corpus count with its as-of month. Schedule 13D/G filings: the count of unique filings (one per accession number) with its as-of date. Senate STOCK Act documents and trades: from the Senate eFD collection, with the latest disclosure date. 8-K events: the count and the issuer count over the stated window. Form 144 sale notices: the enumerated count over its window. Company fundamentals records: the universe count with its generation date. Research articles: the count of published articles, split into Filing Tape audits and features and studies, counted at build time. House STOCK Act trades are not shown: no single canonical file reproduces the House totals, so the number is withheld until a reproducible source exists.
A fund’s SEC filer identity can change: Pershing Square filed under CIK 0001336528 through 2026 Q1 and under CIK 0002026053 from 2026 Q2. When a filer changes, the change column and the “opened / added” counts on fund and stock pages are measured against the combined holdings of the previous and the new filer — the old CIK’s last filing is the previous quarter, not an empty portfolio. A position that existed under the old CIK and grew under the new one is shown as ADDED, not NEW; only positions that did not exist under either filer count as NEW. The same combined history is used for the fund’s clone performance (Follow a Fund / manager profile): the performance period reaches back through the previous filer’s history. The mapping is recorded in the pipeline (previousCiks) and the change is noted on the fund page and in Corrections. A sweep runs over the tracked funds for the same break — a quarter in which at least 80% of holdings are NEW while the previous quarter has no filing — and any fund that matches is checked the same way before its pages are published.
Each of the 33 lenses has its own page at /find/{lens-slug}, where the slug is the lens name in lowercase with hyphens. The page states the lens standard in one line, then lists the passing stocks — at most 100 rows, the rest behind an expander — with the lens metrics, each row linking to the stock page. The title count is the number of tickers actually shown: tickers without a company name in the registry are left out, so the title matches the table and the live app. Chart lenses have no metrics, so they list ticker links instead of a metric table. A Track Record summary appears only where the lens has been scored (today: Insider Cluster Buying, scored on a 14-day formation window — the lens itself lists the last 90 days, so the two windows differ and the page says so). Other lenses link to the Track Record page without a summary. Red Flags pages carry the note "A screen for what to avoid — not a buy signal." Only a lens with zero passing stocks is excluded from indexing; all 33 lenses currently have passing stocks. The Find page links every lens row to its page, and the home lens cards link to the matching family anchor on /find.
Every stock page opens with a six-cell summary panel: tracked funds holding, insider buys vs sales over the last 90 days, members of Congress reporting trades, revenue for the latest fiscal year, red-flag screens flagged, and shares short as a share of shares outstanding. Each cell links to its section; a cell hides when its data is absent (zero is a measured value and is shown). A sticky table of contents sits below the panel with the same sections. The holdings table shows the 10 largest holders first with a "Show all" expander and New / Added / Trimmed / Exited filters; exits are rows in the same table. The price chart is two years of closing prices as an inline SVG built at build time — raw (unadjusted) closes, so dividend dates show small steps — with dashed lines at 13F filing deadlines. It is hidden when fewer than 130 trading days are on file. The holder-count chart shows tracked funds holding the stock by quarter, from the quarterly BEST lists; a stock appears there only if it ever made a quarterly BEST list, and the count is tracked funds only, not all 13F filers (the "All institutions" table reports the full-filer count, which is a different number).
The "5% holders" section lists the most recent Schedule 13D/G filings for the stock: filer, form (13D or 13G, including amendments), shares and percentage owned as filed, and the event date, with a link to the filing on SEC EDGAR. One filing is one row; a filing may list several reporters and the first is shown. Figures are as filed, including amendments that report zero shares after a transfer. The parsed corpus is 57,443 filings to October 2, 2026; the home-page count is an accession count from a later collection and is larger — the two counts cover different sets and are not the same number.
The "Planned sales" section lists the most recent Form 144 notices for the stock: seller, role, shares to be sold, aggregate market value, and the approximate sale date, with a link to the filing on SEC EDGAR. A Form 144 notice is a planned sale, not a completed sale. The corpus is 63,013 notices from April 13, 2023 to October 9, 2026. A stock with no notices shows no section.
Each quarter, stocks are ranked by how many tracked funds hold them, then by how many opened or added (13F BEST); by how many insiders bought on the open market (Insider BEST); and by how many House members bought (Politician BEST). Lists show the Top 50.
The home page list "What super investors bought this quarter" ranks stocks by how much of the company the tracked funds bought in the latest complete quarter — net buying as a share of market value — not by how many funds bought. Numerator. Across the tracked funds, the change in shares held from the previous quarter-end to this quarter-end, multiplied by the quarter-end close: sum((Q2 shares − Q1 shares) × Q2-end price). Share counts are split-adjusted (see below), and the quarter-end price is used so that price moves inside the quarter do not count as buying. Denominator. Market capitalisation at the same quarter-end: quarter-end price × shares outstanding from the most recent SEC filing. Ranking. Numerator divided by denominator, highest first, top 8. Each row also shows the dollar net buying and how many tracked funds added shares. Qualification. A stock qualifies only with market capitalisation of at least $1 billion, net buying of at least $100 million, and at least 3 funds adding shares. Funds absent in a quarter. A fund that did not file in Q1 contributes zero Q1 shares for every stock; a fund that did not file in Q2 contributes zero Q2 shares. A fund that held a stock in Q1 and sold it entirely in Q2 counts as a full sale. New listings. A stock that first listed during the quarter has zero Q1 shares, so its full Q2 position counts as net buying; IPO allocations and open-market purchases are not separated. Spin-offs. Shares received in a spin-off distribution are not buying. FedEx Freight (FDXF) began trading June 1, 2026 as a spin-off of FedEx, distributed at one FDXF share per two FedEx shares: for each fund, half its Q1 FedEx shares are subtracted from its Q2 FDXF shares before summing. Splits. Q1 share counts are adjusted to the post-split basis for every split in the quarter, verified against company announcements: Carvana 5-for-1 (May 7, 2026), Booking Holdings 25-for-1 (Apr 6), KLA 10-for-1 (Jun 12), Vanguard Information Technology ETF 8-for-1 (Apr 21), Vanguard Growth ETF 6-for-1 (Apr 21), iShares Russell 1000 Growth ETF 4-for-1 (Apr 29), Powell Industries 3-for-1 (Apr 6), Cardlytics 1-for-10 reverse (Jun 5), DuPont 1-for-3 reverse (Jun 24), Mueller Industries 2-for-1 (Jun 30). Shares outstanding are adjusted the same way when the filing date behind the count predates the split. Exclusions. Stocks without shares-outstanding coverage are excluded, as are stocks whose tracked Q2 share count exceeds the reported shares outstanding (an impossible denominator, which flags a bad share count). Where the reported share count is a weighted average rather than a quarter-end count, the percentage is approximate. Staleness. The list is a snapshot of one quarter and is recomputed only when the next quarter’s filings are complete; it does not move with daily prices. Nothing on this page is a recommendation.
Entry and exit. The Top 30 of each quarter’s list, equal-weighted, are scored over the following quarter. Two entry bases are reported side by side: quarter-end (bought at the quarter-end close — measurable but not tradeable, because the filings are not public yet) and filed-entry (bought at the filing deadline, 45 days after quarter-end — the first day the full list was public; if the deadline falls on a weekend, the next business day). Exit is at the following quarter-end close for both bases. The two bases therefore cover different holding periods: the quarter-end basis is held for exactly one quarter, while the filed-entry basis — bought about 45 days after quarter-end — is held only until the following quarter-end close, about six to seven weeks (43–47 calendar days across the 40 scored quarters).
Benchmark. SPY over the identical window, using dividend-adjusted closes for both the stocks and SPY. Costs. Trading costs, slippage and taxes are not reflected. Sample sizes. Every score reports n — how many of the 30 names could actually be measured.
The most-held (BEST) track record now spans 40 quarters, 2016Q3–2026Q2, extended in November 2026. The 2016Q3 “adds” counts every position as new, because there is no prior quarter in the panel, so it is not comparable to later quarters. The added quarters (2016Q3–2021Q2) are built from the panel of original 13F filings, while the 2021Q3 and later quarters keep the earlier live-cache values; the two sources differ by about one add per affected ticker in 2021Q3, where the live cache picked partial amendments.
Form 13F is due within 45 days after quarter-end and covers long U.S. positions only. The holdings are as of quarter-end, so the lag between quarter-end and publication varies by when each manager actually files. Any "copy the fund" return measured from quarter-end contains look-ahead: you could not have known the list. That is why the filed-entry basis exists and why both bases are always shown together. The same lag applies to House STOCK Act reports.
Rule. Every stock page and every politician page shows the Congress as-of date. When Congress data is more than 14 days older than the Form 4 as-of date in the same build, the page also shows a notice reading "Congress data last updated [date]". The reference is the Form 4 as-of, not the calendar, so a rebuilt page and its notice always agree. The 14-day limit is a display rule, not a data rule: stale pages keep showing the last verified numbers and never substitute an estimate. As of the latest build, Congress data runs to Nov 16, 2026 and Form 4 data to Nov 19, 2026, inside the 14-day limit, so no stale-data notice is shown.
Rule. On a stock page, “Related stocks” ranks other stocks by lift: for a candidate stock T, lift is the share of the funds holding the page stock S that also hold T, divided by the share of all tracked funds that hold T. A candidate needs at least 3 funds holding both S and T. The 20 most widely held stocks in the whole universe are excluded, so the list is not just the mega-caps every fund holds. The top 8 by lift are shown. If fewer than 4 stocks meet the rule, the list is filled to 8 with stocks in the same industry (same SEC SIC code first, then the same 2-digit SIC division), ranked by holder count — and the page says how many of the 8 are such fillers. The denominator is the tracked funds with a filing for the latest quarter. As of the latest build, that is 147 funds. Lift is a co-holding pattern, not a recommendation and not evidence that the stocks move together.
Rule. The Latest filings page carries three lists of five rows each. Insiders: open-market purchases (Form 4 transaction code P) of $100,000 or more, newest filing date first. 13F: the five most recent filings by tracked funds for the quarter ended September 30, 2026, newest filing date first. Congress: the five most recent member-level disclosures, newest disclosure date first. Each list is shown only while its newest item is no more than 14 days old at build time; an older list is hidden and the page says why. “Newest item” is the newest filing or disclosure actually in the list — not the source file’s as-of label, which can run ahead of its data. Two limits are stated plainly. The Form 4 corpus carries no Rule 10b5-1 plan flag, so plan trades are not removed from the insider list. House reports arrive as PDFs without transaction-level rows in the current corpus, so the Congress list shows per-member activity, not individual trades.
Rule. A stock page shows committee jurisdiction only from the stock’s SEC SIC industry code and the current (119th Congress) committee rosters. Each industry group maps to the standing committees whose remit names it: finance (SIC 6000–6999) to House Financial Services and Senate Banking; energy (SIC 1300–1389, 2900–2999) to House Energy and Commerce and Senate Energy and Natural Resources; health (SIC 2830–2839, 3840–3849, 8000–8099) to Senate Health, Education, Labor, and Pensions and House Energy and Commerce; education (SIC 8200–8299) to Senate Health, Education, Labor, and Pensions and House Education and Workforce; defense (SIC 3480–3489, 3720–3729, 3730–3739, 3760–3769, 3810–3819) to House and Senate Armed Services; agriculture (SIC 0100–0999, 2000–2099) to House and Senate Agriculture; transport (SIC 4000–4799) to House Transportation and Infrastructure and Senate Commerce, Science, and Transportation; telecom (SIC 4800–4899) to House Energy and Commerce and Senate Commerce, Science, and Transportation; mining (SIC 1000–1299, 1400–1499) to House Natural Resources and Senate Energy and Natural Resources. Committees whose reach is economy-wide or non-industrial (Appropriations, Judiciary, Ethics, Intelligence and the rest) map to nothing, and subcommittee seats do not count — only the standing committee itself. The page also names the tracked members who traded the stock in the current window and sit on one of its jurisdiction committees; if none do, it says so. Assignments are the current Congress only: a trade made in an earlier Congress may have been made under a different assignment, and the page says that too. An industry with no mapped committee shows that plainly instead of a forced match. The same mapping was used in our committee-jurisdiction study and its returns follow-up.
Rule. A stock page shows a “Government relations” section only when the stock has lobbying or federal-contract data; otherwise the section is hidden, not filled with an estimate. Lobbying: Lobbying Disclosure Act filings naming the company as client, matched by company core name, 1999–2026 — 154713 distinct filings matched to 1328 tickers. The page shows the filing count, the year range and the most common issues; it does not show a dollar total per ticker, because the income and expense figures are reported per filing and per registrant. A filing can match more than one ticker under core-name matching, so per-ticker counts do not sum to the distinct-filing total. Federal contracts: USAspending obligated amounts (money committed, not money spent), fiscal years 2021–2026, for 666 tickers, matched by recipient name — contracts under a subsidiary’s name can be missed. The page shows the fiscal-year totals and the largest awards collected. The collection records no per-ticker award count, so none is shown. Lobbying and contracts are shown as disclosure facts; our lobbying-to-contracts study found no link between lobbying increases and next-year contract growth.
Rule. The Research hub lists every published article — article URLs never change. The front page shows the 6 newest articles, then the newest 20 non-audit articles per page. Each article has exactly one category — Studies, Explainers, Filing season or Stock audits — shown on its own category page (Studies, Explainers, Filing season, Stock audits), and can carry several tags (13F, Insiders, Congress, Short interest, 8-K) drawn from its subject. Filing Tape audits are per-ticker articles; the hub’s ticker search filters the visible cards in the browser and searches titles and tickers, not the article bodies. A new article appears in the hub on the next build; an article with no category yet is listed under Studies and logged in the build output until it is classified.
Three filing signals are scored across 2021 Q3 – 2026 Q1: crowded accumulation (3+ funds increasing the same stock’s portfolio weight in a quarter — the position’s value as a share of the fund’s total reported 13F value, reconstructed from reported share counts and quarter-end prices), tripled positions (a fund at least tripling its share count, positions of $5M+), and crowded exits (3+ funds closing a position entirely). Entry is at the close on the first trading day after the filing deadline, and exit one quarter later on the same rule, on dividend-adjusted closes. The deadline-day close itself is not used: an audit of EDGAR acceptance timestamps found that about 31% of the filings behind these signals were accepted only after that close, so it was not always a price at which the full signal was public information. Under the previous entry — the deadline day’s own close — the same signals’ excess returns were crowded accumulation -0.4%, tripled positions -1.3% and crowded exits -4.8%; the change of basis moves no conclusion. Of 3,749 signal events, 3,710 could be measured with price data. Crowded exits are the thinnest sample: 151 of 190 events had usable prices — the rest were largely delisted or unmapped names, so that result in particular should be read with survivorship bias in mind. Events that cannot be measured are excluded from averages, never estimated.
The headline table reports each 13F signal\u2019s average excess return against the S&P 500. An average over 19 quarters is an estimate, so we publish a 95% confidence interval next to it. The interval answers one question: if the same signal rule had been run over a different 19 quarters drawn from the same history, how far could the average move? Method — quarter-block bootstrap. We resample quarters, not individual signals. Signals from the same quarter share that quarter\u2019s market move, so treating them as independent would understate the uncertainty. Each bootstrap replicate draws 19 quarters with replacement from 2021Q3–2026Q1; a drawn quarter contributes all of its measured signals. The replicate\u2019s excess return is the same event-weighted mean the headline uses. We repeat this 100,000 times (fixed seed, so the result reproduces exactly) and take the 2.5th and 97.5th percentiles as the 95% interval. The quarterly excess returns behind the calculation are published on the Track Record page, so anyone can see which quarters drive the average.
| Signal | Signals (n) | Excess | 95% CI (excess) |
|---|---|---|---|
| Heavy accumulation — 3+ tracked funds increased together | 2,936 | -0.5% | -1.8% to +1.1% |
| Share count tripled (3x) | 623 | -1.5% | -3.5% to +0.6% |
| Heavy exit — 3+ tracked funds sold out together | 151 | -5.4% | -10.3% to -1.3% |
The S1 and S2 intervals include zero. The average excess return is statistically indistinguishable from zero — a null result, not evidence of losing. The S3 interval lies entirely below zero: the stocks that three or more funds exited together went on to underperform the S&P 500 by 5.4 points on average, and that gap is distinguishable from zero at 95%. Read it with the sample in mind: only 151 signals, and two quarters dominate the lower tail (2025Q3 -30.4% on 5 signals; 2025Q4 -29.9% on 9). This is a property of these 19 quarters, not a rule that exits predict losses.
Two reading notes. First, the interval is computed on the raw mean excess return. The headline point value next to it is the display value (displayed average minus displayed SPY, one decimal): S1 shows −0.5% where the raw mean is −0.4%, and S3 shows −5.4% where the raw mean is −5.3%. The 0.1-point gap is display rounding, not a different estimate; the interval bounds are rounded the same way. Second, the interval is about the average, not about any single signal. Most individual signals beat or trail the market by far more than the interval is wide — the interval says how precisely we know the average, not how predictable any one stock is.
The Track Record page also splits each 13F signal by the economic backdrop in force on its entry date — the close on the first trading day after the filing deadline — using the same regime rule as the home page\u2019s FRED label. Growth: the NBER recession indicator (FRED USREC) at or above 0.5 means recession, otherwise expansion. Inflation: CPI year-on-year (FRED CPIAUCSL) at or above 3% means high inflation. Rates: the effective federal funds rate (FRED FEDFUNDS) over three months — rising if up 0.1 percentage points or more, falling if down 0.1 or more, otherwise on hold. Inversion flag: the 10Y–2Y Treasury spread (FRED T10Y2Y) below zero appends "yield curve inverted" to the label. Within each regime cell, signals are aggregated exactly as the headline aggregates them, and a cell with fewer than 30 signals is marked "small sample". Every cell is published, including cells where the signals trailed SPY.
Point in time. The regime is judged using only FRED observations that had actually been published by the entry close. FRED\u2019s files carry observation dates, not release dates, so we apply the publishers\u2019 standard release lags, conservatively: CPI for month M is treated as published on the 15th of M+1, the funds rate and the unemployment rate on the 8th of M+1, and the spread uses the last observation dated before the entry day. These lags are assumptions about release timing, not FRED vintage data, and they are stated here so the classification can be reproduced and challenged. In-sample note: the only NBER recession months in the fetched history are March and April 2020, so there was no recession at any signal entry date from 2021Q3 to 2026Q1 — the growth axis is expansion for every quarter in the sample and the scorecard effectively splits on inflation and rate direction. Insider clusters are not split by regime: the pipeline saves only their aggregate statistics, not per-formation dates and returns, so a regime split cannot be computed from existing pipeline output.
Survivorship and missing data — the details. The manager universe for these statistics is the 83 funds that made up the tracked roster when these statistics were computed (81 with SEC CIK numbers); the roster has since expanded to 149 tracked funds, but the figures in this section stay on the original roster, so the sample described below is the sample that was actually measured. Funds that closed, merged or left the tracked list before that snapshot do not appear in any quarter, including quarters when they were active — the historical roster is not available to us, so this bias cannot be measured, only disclosed. Funds that disappear are disproportionately poor performers, so a survivor-only roster tends to flatter the signals built from it. Within the sample itself the roster is stable: every tracked fund that filed in 2021 was still filing in 2026, the effective manager count rose from 77 to 81 as four filers were added (in at least one case because the fund’s SEC filer entity changed), and 76 of 81 funds filed in all 21 quarters.
A stock can only enter a signal if its position’s CUSIP maps to a ticker and its quarter-end prices exist for both quarters being compared — a stock that delists mid-quarter disappears at this stage, before any return is measured. Every signal that was generated was then measured wherever prices exist: 3,710 of 3,749 events across the three signals, and 2,936 of 2,936 for crowded accumulation. The exception is crowded exits: only 151 of 190 events could be measured. The missing 39 are concentrated in names that were acquired, merged, renamed or delisted — exactly the outcomes a price series cannot follow — so that signal’s average should be read together with its 79% measurement rate.
The biases we can identify do not all point the same way. Survivorship — in the roster above and in the names that stop trading mid-measurement — pushes the published figures up, not down: the headline result is that crowded accumulation did not beat the S&P 500 (−0.4 percentage points of excess return), and correcting for survivorship would, if anything, make that figure worse. The other hole cannot be signed: positions whose CUSIP never maps to a U.S. ticker or price history are missing from the signals entirely, and those names plausibly include both future delistings, which would have dragged returns down, and obscure winners, which would have lifted them — so for that omission we report counts and measurement rates rather than a direction. We publish the sample size and measurement rate next to every average so the size of each hole is visible, and we never estimate or backfill a missing return.
The fund counts on a stock page aggregate the tracked roster by ticker: every share class whose CUSIP maps to that ticker counts toward it — Alphabet’s legacy CUSIP, still used by some filers, is mapped to GOOGL, for example — and funds added to the tracked roster after an article was published appear here as soon as their filings are collected. Research headlines instead cite the filer count from the direct EDGAR collection as it stood on the publication date, for the share class named in the headline. The two bases can therefore differ by a holder or so on a given stock; the filings behind both are the same, and neither figure is a restatement of the other.
Insider screens use open-market purchases from Form 4 over the last 90 days. The "opportunistic buys" screen additionally removes routine buying: an insider whose purchases repeat in the same quarter in at least two of the previous three years is treated as routine (for example employee stock-plan purchases) and excluded, and issuers without a verifiable filing history before 2025 are excluded as well.
A cluster is three or more distinct insiders of the same company buying on the open market — Form 4 transaction code P, non-derivative shares — whose transaction dates fall inside a rolling 14-calendar-day window. Buys are first collapsed into insider-days (one insider, one stock, one transaction date), so repeated buys by the same insider on the same day count once. Amended filings (Form 4/A) replace the filings they amend; in the 2025–2026 corpus only original Form 4 filings are used. A cluster forms when the third distinct insider’s buy joins the window, and the formation date used for scoring is that buy’s filing date — the day the cluster became public — never a transaction date, so the scoreboard contains no look-ahead of the kind that trade-date timing would create.
Scoring. Entry is the close on the first trading day after the formation date; exit is 60 trading days later, on dividend-adjusted closes; the excess return is measured against SPY over the identical window. Only the first formation per stock per calendar quarter is scored; later formations in the same quarter are counted in the published data file as a sensitivity. Clusters whose 60-day window has not yet elapsed, and clusters with no usable price history (mostly delisted or unmapped names), are excluded from the averages and counted where the results appear — never estimated or backfilled. Because a name that disappears cannot be measured, the measured averages are biased upward by exactly the size of that hole.
10b5-1 plan buys. The headline series removes Rule 10b5-1 plan buys before clusters are formed: an insider-day is removed when every filing behind it carries the aff10b5One flag — the filer’s own disclosure that the trade was made under a pre-arranged plan. The flag is read from the filing text on EDGAR, and it exists only in filings received from April 2023, when the SEC’s disclosure rule took effect; buys filed earlier cannot carry it and are included in both series. A second series keeps plan buys in, so the size of the adjustment is visible. In the current build, 1,296 insider buying days were removed this way, out of 134,746 insider buying days in the full record.
This scoreboard is not the same measurement as our earlier study of insider clusters: that study used a 90-day window, suppressed repeat events for 90 days and compared clusters against isolated single-insider buys. Different window, different question — the figures are not expected to match, and neither revises the other.
The Track Record page also runs one paper portfolio per member of Congress, built from STOCK Act periodic transaction reports (House Clerk PDFs and Senate eFD) filed between Jan 1, 2021 and Sep 30, 2026. Entry. A disclosed purchase buys one unit at the close on the first trading day after the report was filed — the first close at which a reader of the filing could have acted; trade-date timing is never used, so the portfolios contain no look-ahead. Exit. When the same member discloses a sale of the same stock, every open unit of that stock is sold at the close on the first trading day after that filing. A stock with no disclosed sale is held to Sep 30, 2026 and valued at that day’s close; such still-open positions are counted separately and excluded from the closed-position win rate. Weighting. Every disclosed buy is one equal-weight unit regardless of size: each day, the portfolio’s return is the average of its open units’ returns, compounded from the member’s first entry; days with nothing held are cash (0%). SPY is measured over the identical span for each member. Trading costs, slippage and taxes are not reflected.
Stocks only. Senate rows are limited to asset type "Stock" (stock options, bonds and other asset classes in the same reports are excluded — 578 option rows in the window, among others). House rows carry an asset-class tag in the filing; rows tagged as options ([OP]) or as non-equity classes (government or corporate securities, mutual funds and similar) are excluded on the same basis. Amounts. STOCK Act reports disclose ranges, not values, so the headline portfolios ignore amounts entirely. As a sensitivity, each buy can instead be weighted by the midpoint of its disclosed range — with the open-ended "Over $50,000,000" range taken at its $50,000,000 lower bound. That series is published next to the headline figures and is an assumption built on ranges, not a measurement. Deduplication. Amended and refiled reports are collapsed on (member, ticker, transaction type, trade date, amount range), the same key as our earlier congressional studies.
Coverage limits. House filings that exist only as scanned images cannot be machine-read and are excluded, never estimated — 476 of the 2021–2026 House filings in the current cache; the Senate eFD record is electronic. Buys whose stock has no usable price history (mostly delisted, renamed or unmapped names) cannot be scored and are excluded and counted where the results appear. A buy is also excluded when the first traded price after its filing is more than 15 days later — the signature of a delisted or reused symbol whose current price history does not cover the disclosure — and the same 15-day test decides whether a disclosed sale can be priced as an exit. Members need at least 10 scored buys to be listed, so the scoreboard covers frequent traders only. A portfolio is not a signal average: our Senate-vs-House study scores every buy independently over the 60 trading days after disclosure, while a portfolio holds overlapping positions until a sale is disclosed and compounds them — the two measurements answer different questions and are not expected to match.
On member pages, one reported trade is counted once per account: rows that share the same account, date, and amount range are combined into a single trade, while rows from different accounts are counted separately. An amended filing replaces the original row, and a deleted filing removes it, so a corrected or withdrawn disclosure does not double-count. Option rows tagged [OP] are included in the purchase and sale totals and are flagged on the ticker row; they are not shown as a separate headline total. Trades in tickers outside the tracked stock universe are counted separately and excluded from the member's table, so the table shows only trades in tracked stocks.
Three published accounting models, used as warning screens only: the Beneish M-Score (eight-variable model; flagged above −1.78), the Altman Z″-Score (flagged below 1.10) and Sloan accruals (flagged when the accruals ratio exceeds 25% in absolute value). As of the latest build, the M-Score could be computed for 430 companies (39 flagged), Altman Z″ for 1130 (321 flagged) and Sloan accruals for 1693 (160 flagged). These models produce false positives — very fast-growing companies routinely trip the M-Score — so a flag is a prompt to read the filing, not a conclusion.
Stock pages are indexed only when at least 2 of 4 data signals hold for the stock: held by 2 or more tracked funds in the latest 13F quarter, Form 4 insider activity in the last 90 days, a congressional STOCK Act trade, or a red-flag screen hit. A page with fewer than 2 signals is marked noindex and left out of the sitemap, so a page with almost nothing to show does not present as content. Filings-audit pages are indexed only when the stock has a red-flag finding; audits without one are noindex and left out of the sitemap. Pagination pages are left out of the sitemap.
The corpora behind these measurements — the full Form 144 collection, the SEC fails-to-deliver and FINRA short-interest panels, the canonical House and Senate transaction tables and the tracked-roster 13F aggregate — are published as free CSV downloads on our Open Data page, with coverage, known gaps and citation terms stated for each file. If you build on them, the citation we ask for is a credit and a link back to that page.
Fund pages with no 13F filing in the registry are marked noindex and left out of the sitemap, since they show a status notice rather than holdings. Q3 season child pages are all marked noindex and left out of the sitemap: they duplicate the fund page holdings, and the Q3 season hub is the canonical season page.