check_citations
ChatGPTCheck every number in a draft against the filing characters it cites, and link them. Write your draft with a citation right after each figure, using the ref that list_figures or verify_figure gave for it: revenue of $68.1 billion [ref:0001045810-26-000019/q4fy26pr.htm@112+4] For a number you calculated, cite its inputs: up 73% [calc: ref:<current>, ref:<prior>] A ref names the exact characters in the document: <accession>/<filename>@<start>+<len>. Returns per citation: ok / rounded (your number is the cited figure times a power of ten, rounded), mismatch (the cited characters are a different number), bad_ref (no number at that position), not_derivable (no sum, difference, ratio or growth rate of the inputs gives your number). unsourced_numbers lists figures with no citation. Fix every failure before presenting the text, then use linked_text and footnotes: each footnote links to the highlighted figure in the filing, so a reader can check it. Args: text: The draft, with [ref:…] / [calc: …] citations (max ~200 KB)
get_company
ChatGPTLook up a company by CIK to get its name(s), ticker(s) and industry. When known, the result also has the SEC's SIC classification: sic (4-digit code, e.g. "6200"), sic_description (e.g. "SECURITY & COMMODITY BROKERS, DEALERS, EXCHANGES & SERVICES"), sic_url (a page listing the other companies in the same industry), fiscal_year_end (MMDD, e.g. "1231" for 31 December) and state_of_incorporation. SIC is the SEC's own scheme and is not GICS; individuals and most funds have none. succession_filing, when present, is an 8-K12B or 8-K12G3 from the last five years in which this company says it succeeded an earlier registrant; its older history may be under that predecessor's CIK (see search_companies). Args: cik: Central Index Key (e.g. "0000320193" or "320193")
get_document
ChatGPTFetch a filing document's content from storage. For one section of a 10-K, 10-Q or 8-K (MD&A, Risk Factors, Legal Proceedings, Item 2.02…), use get_filing_item instead: it returns just that Item, a fraction of the document, with a link highlighting it. Returns the document as markdown (default) or raw HTML. Use this to read the actual text of a whole 10-K, 8-K, exhibit, or any other filing document. The first lines give its source: the sec.gov URL to cite, the filing's EDGAR index page, and a beatandraise.com viewer link. Args: cik: CIK of the filer (e.g. "0000320193") accession_number: SEC accession number filename: Document filename from the content_object (e.g. "ef20060722_8k.htm") get_markdown: Return markdown version (default True). Set False for raw HTML.
get_filing_detail
ChatGPTGet full filing metadata including its content_object. content_object lists every document in the filing with filename, doc_type, description, sequence and whether a markdown version exists; read one with get_document(cik, accession_number, filename). Cite sec_base_url + filename; the beatandraise.com viewer is beatandraise_base_url + filename. The R*.htm files are XBRL financial-table pages; list_filing_tables says which table each one is. Args: accession_number: SEC accession number (e.g. "0001140361-26-006577")
get_filing_item
ChatGPTRead one Item of a 10-K, 10-Q or 8-K, e.g. Risk Factors or MD&A, instead of the whole document. Much smaller than get_document: a 10-K's Item 1A is typically 5-15% of the filing. The first lines say which item it is, give the sec.gov source to cite, and a highlight_url marking the whole section in the filing. If the item's heading was not found, you get the text between the neighbouring items instead, and the first line says so ("APPROXIMATE"). That span contains the item if the filing has it: search within it. Args: accession_number: SEC accession number of a 10-K, 10-Q or 8-K item: The item: a number ("1A", "7", "7A", "2.02") or a name ("Risk Factors", "MD&A", "Legal Proceedings"). For a 10-Q, "Part II Item 1A" also works. part: 10-Q only: "I" or "II" when the number exists in both parts ("1", "2", "3", "4"); a bare number means Part I. get_markdown: Markdown (default). Set False for the raw HTML of the section.
get_filings
ChatGPTQuery SEC filings by form type, date range, and/or CIK. Returns deduplicated filing metadata sorted by filing date (newest first). Each filing's content_object lists its documents (filename, doc_type, description, sequence; sequence 1 is the primary document); read any of them with get_document(cik, accession_number, filename). A document's sec.gov URL, to cite, is sec_base_url + filename, and its beatandraise.com viewer page is beatandraise_base_url + filename; the filing's own source gives the primary document's links and a ready-made citation. The R*.htm files are XBRL financial-table pages; list_filing_tables says which table each one is. Args: form: SEC form type (e.g. "10-K", "10-Q", "8-K", "4", "13F-HR", "S-1") cik: Company CIK (e.g. "0000320193" or "320193") from_date: Start date inclusive (YYYY-MM-DD). Without cik, defaults to 30 days before to_date (or today); pass it to reach older filings. to_date: End date inclusive (YYYY-MM-DD) limit: Max results (1-1000, default 100) offset: Pagination offset
get_financial_history
ChatGPTA company's financial statements from its last several 10-Ks or 10-Qs, in one call. For trends (revenue, margins, cash flow over years or quarters) instead of calling get_financial_statements filing by filing. Returns each filing, newest first, with its statements as that filing presents them (the same text get_financial_statements gives). They are not combined into one table: a 10-K shows two or three years, so periods repeat across filings (the newest filing's figure is current; it may restate), and labels, line items and scale can differ between filings. Read the figures from each filing yourself. Args: cik: Company CIK (use search_companies to find it) period: "annual" (10-K, or a foreign filer's 20-F / 40-F; default 5 filings, max 10) or "quarterly" (10-Q, default 8, max 20; foreign filers report quarters on 6-K, not covered) statements: Optional subset: "income_statement", "balance_sheet", "cash_flow", "equity", "comprehensive_income". Ask for fewer to get more filings in one answer. limit: Number of filings to read before: Only filings filed on or before this date (YYYY-MM-DD), for history as of a past date fiscal_year: Only filings for this fiscal year, as the company states it on the filing (fiscal, not calendar: Medtronic's fiscal 2018 runs May 2017 to April 2018) fiscal_period: With or without fiscal_year: "Q1", "Q2", "Q3" (a 10-Q) or "FY" (the annual report). There is no Q4 filing: the fourth quarter is in the annual report, as the year. E.g. fiscal_year=2018, fiscal_period="Q1" returns that one 10-Q; fiscal_year=2018 with period="quarterly" returns its Q1-Q3. Each returned filing carries its fiscal_year and fiscal_period.
get_financial_statements
ChatGPTExtract structured financial statements from a 10-K or 10-Q filing. Returns balance sheet, income statement, cash flow statement, equity statement, and comprehensive income — each as markdown table content extracted from the filing's XBRL R-files. To use this tool, first find a 10-K or 10-Q accession number using get_filings, then pass it here. Args: cik: Company CIK (e.g. "0000320193") accession_number: Accession number of a 10-K or 10-Q filing
get_holdings
ChatGPTGet parsed 13F-HR institutional holdings for a filer. Returns individual stock positions with issuer name, CUSIP, FIGI, market value, share count, investment discretion, and voting authority. Data is extracted from the 13F information table XML. Each holding also carries issuer_cik, sic and sic_description (the SEC's SIC industry classification of the issuer, not GICS) when the CUSIP maps to a company with a SIC. In a typical quarter that is about 87% of the value held; funds, ETFs, options and unmapped CUSIPs have none. "with_sic" in the result says how many holdings on this page were classified, so say how much of a portfolio an industry breakdown covers. Use sic_url from get_company to see peers in an industry. Args: cik: CIK of the 13F filer / institutional investor (e.g. "1535950") period: Get holdings as of this date (YYYY-MM-DD). Returns the most recent 13F filed on or before this date. limit: Max holdings per page (1-5000, default 500) offset: Pagination offset
get_insider_transactions
ChatGPTGet parsed Form 4 insider transactions for a company. Returns insider buy/sell transactions with owner details (name, title, relationship), transaction details (date, code, shares, price), and post-transaction holdings. Data is extracted from the Form 4 XML. Transaction codes: P=Purchase, S=Sale, A=Award, M=Exercise, G=Gift, F=Tax Args: cik: Issuer CIK — the company whose insiders are trading (e.g. "320193" for Apple) from_date: Start date (YYYY-MM-DD) to_date: End date (YYYY-MM-DD) limit: Max filings to process (1-100, default 20) offset: Pagination offset
get_institutional_holders
ChatGPTWho holds a company, from 13F filings: how much of each share class institutions hold, and how that changed quarter by quarter. The company-side view of 13F; get_holdings is the manager side (what one institution owns). Use it for questions like: - "What % of AMD do institutions hold, and how has it changed over the last 4 quarters?" -> quarters=4; read quarters[].share_classes[].pct_of_class. - "Is Baupost an investor in AMD?" -> holder="Baupost" (a name fragment or a filer CIK). holder.matched_filers says which 13F filers the name matched; holder.positions gives each quarter's position; holder.not_held_in lists quarters with none. If matched_filers is empty, no 13F filer has that name. With holder the response is lean by default (about 5 KB: no top_holders/buyers/sellers lists); pass detail="full" if you also need those. - "Who are the largest holders / who bought or sold most last quarter?" -> top_holders, biggest_buyers, biggest_sellers, largest_new_positions, largest_exits. - "Who owns more than 5%?" -> large_holders_5pct (latest 13D/13G per holder). Returns, newest complete quarter first (the quarter still being filed is in in_progress_quarter, not in quarters): quarters[]: period, holders, new_holders, exited_holders, shares_held, value_usd, share_classes[] (one per class on the company's cover page: shares_outstanding as of the 10-Q/10-K after the quarter, listed/symbol, held_by_13f, pct_of_class, holders, cusips), share_count_source (the filing the count comes from), holders_url. source.beatandraise_url: the holders page, the human view of the same data. Report each share class separately; never add classes together (Meta's Class B is unlisted, so no institution can hold it). State the caveats in notes when giving numbers: 13F is as filed and not deduplicated, options are excluded, holdings can exceed 100% for heavily shorted stocks, and a "new position" can be a manager whose previous 13F was late or a placeholder. Cite share_count_source.sec_url for share counts and holders_url for holdings. Args: company: Ticker (e.g. "AMD") or CIK of the company being held (not of the institution) quarters: Number of complete quarters, newest first (1-12, default 4) holder: Optional 13F filer to follow across the quarters: a name fragment ("Baupost", "Berkshire") or a CIK ("1061768") limit: Number of top holders to return for the latest quarter (1-100, default 25) detail: "lean" or "full". Default: lean when holder is given (just the holder's positions, share classes and sources), full otherwise (adds top holders, buyers, sellers, new positions, exits and 5%+ holders)
get_private_offering
ChatGPTGet one Form D filing in full, by accession number (from search_private_offerings). Everything search returns, plus: the issuer's full address, phone and previous names; co-issuers; related persons (executive officers, directors, promoters) with their roles and addresses as filed; sales compensation recipients (placement agents and brokers) with FINRA CRD numbers, their broker-dealer and the states they solicited in; notes the filer added to amounts, fees and use of proceeds; the signatures; and offering_history: every filing of the same offering (original notice and amendments, oldest first) with its cumulative amount sold and investor count, which shows how the raise filled over time. Args: accession_number: Form D accession number, e.g. "0001104659-26-114962"
get_shares_outstanding
ChatGPTGet a company's shares outstanding over time, as stated on the cover page of each 10-K, 10-Q, 20-F and 40-F (dei:EntityCommonStockSharesOutstanding). One row per share class and as-of date, oldest first, each with the filing it comes from (source). Counts are as filed and never split-adjusted, so a stock split appears as a jump in ratio_to_previous (the count divided by the same class's previous count; Apple 2020: 3.98 for its 4-for-1, the gap being buybacks). splits lists every stock split and reverse split found, each with its ratio ("4-for-1", "3-for-2", "1-for-10"), the dates it falls between (happened_after, by) and the 10-K/10-Q that first restated the company's earlier share counts by that ratio, which accounting rules require for every split (source: cite it). restated_figures is how many share counts were restated: 1-2 is thin evidence. Two splits between the same two filings show as one with the combined ratio. share_count_tagging_corrections are jumps of 1000x or 1,000,000x in the reported data that were the company fixing the scale of its XBRL share-count tag, not splits. Covers 10-K, 10-Q, 20-F and 40-F filers (a 20-F/40-F filer reports once a year, so its window is about a year wide). A split is listed once a filing has restated the share counts, so a ratio_to_previous far from 1 with no split listed is worth checking in the filing. Use this for share-count history, dilution, buybacks and splits instead of reading filings one by one. Multi-class companies (Alphabet, Meta) have one series per class, named by the XBRL member (CommonClassAMember, CapitalClassCMember); "listed" maps classes to tickers from the newest filing. Covers filings from coverage_from (2009) on, for companies that tagged the count in XBRL. Filer errors (a zero, a mistyped date) are returned as filed. Args: cik: Company CIK (e.g. "320193" for Apple) share_class: Only this class, as returned in share_class (optional) from_date: Earliest as-of date (YYYY-MM-DD) to_date: Latest as-of date (YYYY-MM-DD)
list_companies_by_sic
ChatGPTList the companies in an industry, using the SEC's Standard Industrial Classification (SIC). Pass a SIC code (3-4 digits, e.g. "6200" for securities brokers and exchanges, "3571" for electronic computers, "7372" for prepackaged software) to get its companies: CIK, name, ticker, state of incorporation and fiscal year end, 100 per page, companies with a ticker first. Get a company's own code from get_company or search_companies (sic). If you do not know the code, pass a word from the industry name instead ("software", "pharmaceutical", "bank", "oil") and you get the matching SIC codes with how many companies each has; then call again with the code you want. list_sic_codes shows the whole classification. SIC is the SEC's classification of the filer, assigned when it registers. It is not GICS or NAICS, individuals and most funds have none, and some codes are broad (e.g. 6770 "blank checks"). Args: industry: a SIC code like "6200", or a word to look up codes by ("software") query: with a code: only companies whose name or ticker contains this page: with a code: page number, 100 companies per page (default 1) listed_only: with a code: only companies that have an exchange ticker (default True). Set False to include the many dormant, delisted and asset-backed entities that carry the same code.
list_figures
ChatGPTList the numbers a filing document actually states, each with its label. Each entry is a number with its label (the table row's label, or the words leading up to it in prose) and its digits with formatting removed, e.g. {"digits": "416161", "text": "416,161", "label": "Total net sales"}. Any figure you state from this document must be one of these digit strings times a power of ten (possibly rounded): 416161 can be cited as $416.2 billion if the table is in millions, never as $416.9 billion. Work out the scale from the label and the table heading. Tables usually list columns in order (current period first); confirm the period from the document if it matters. Use near to filter by label (e.g. "gross margin", "net income"). A 10-K holds about 2,000 figures, so filter unless the document is short (press releases run ~400). Link an entry by appending "?hl=" + its hl + "&src=mcp#br-hl-0" to highlight_base. Args: cik: CIK of the filer (e.g. "0000320193") accession_number: SEC accession number filename: Document filename (e.g. "q4fy26cfocommentary.htm") near: Keep only figures whose label contains this text (case-insensitive) limit: Max figures returned (1-2000, default 300)
list_filing_items
ChatGPTList the Items of a 10-K, 10-Q or 8-K: which sections the filing has and where. Each entry gives the item ("1A", "7", "2.02"), its part ("I".."IV"; 10-Q items repeat across Part I and Part II), the SEC's title, the heading as the filer wrote it, its length in characters, and a highlight_url that opens the filing with that whole section marked. Read one with get_filing_item. Items are found by their headings in the document, not by searching for the words "Item 1A", so a table of contents or a cross-reference is never mistaken for the section. not_located lists the form's items that were not found: the filing may not have them (Item 1C before 2023, a 10-K/A carrying only Part III, the many 8-K items a given 8-K doesn't report), or the heading was missed. get_filing_item still answers for those, with the text between the neighbouring items. Args: accession_number: SEC accession number of a 10-K, 10-Q or 8-K (e.g. "0000320193-25-000079")
list_filing_tables
ChatGPTList every R*.htm financial table in a filing (10-K, 10-Q, S-1, etc.). Returns one entry per <Report> in the filing's FilingSummary.xml — typically the Cover page, the core financial statements (income, balance sheet, cash flow, comprehensive income, equity), notes, and detail breakdowns. Each entry includes the filename you'd pass to get_document, plus a canonical type (xbrl_role_normalized) so you can find e.g. the income statement across different filers regardless of how they ordered their R-files. Common xbrl_role_normalized values: 'income_statement', 'balance_sheet', 'cash_flow', 'comprehensive_income', 'equity', 'segment', 'cover'. Empty list means the filing has no FilingSummary.xml (most 8-Ks, pre-2009 filings, short-form prospectuses) or it hasn't been indexed yet. Args: accession_number: SEC accession number (e.g. "0000320193-25-000079")
list_sic_codes
ChatGPTList the SEC's SIC industry codes, so you know which code to use with list_companies_by_sic. With a query (a word from the industry title: "software", "bank", "oil", "pharmaceutical") you get the matching codes, most companies first, at most 50, each with the code (sic), the SEC's industry title (description), the SEC office that reviews filers in it (office), and how many companies we hold under it (companies, of which listed have an exchange ticker). Then call list_companies_by_sic with a code, or get_company for one company's own code. With no query you get every code (about 440) as one short line each, "code title (companies)", sorted by code. That is a long list, so prefer a query. SIC is the SEC's classification of the filer, assigned when it registers. It is not GICS or NAICS. Args: query: a word to look for in the industry titles (optional)
search_companies
ChatGPTSearch for companies by name or ticker symbol. Returns matching entities with CIK, all known names, and ticker symbols, plus the SEC's SIC industry code and name (sic, sic_description) for companies that have one. Use this to find a company's CIK before using other tools. A result with succession_filing filed an 8-K12B or 8-K12G3 in the last five years: it became the successor to an earlier registrant (a new holding company, a change of domicile, a merger), and the ticker usually moved with it. Its filings before that date, such as past 10-Ks, are often under the predecessor's CIK. To get the predecessor's name, search that filing (search_document) for "successor" or "12g-3" -- filers write "successor issuer", "successor registrant" or "successor to" -- then search for it. Args: query: Company name or ticker (e.g. "apple", "AAPL", "Tesla") limit: Max results (1-50, default 10)
search_document
ChatGPTSearch within a filing's documents for specific terms. Returns matching lines with context. If filename is omitted, searches every readable document in the filing (the main document and its exhibits, e.g. an 8-K's press release AND its CFO commentary), up to 12 files. If filename is provided, searches only that file. Each file in the result carries its own source (sec_url) — cite the file the match came from. Workflow: 1. Use get_filings to find the accession number for the filing you want 2. Call this tool with just cik, accession_number, and query — it will search all its documents 3. Optionally use get_filing_detail to see all files, then search a specific one by filename Supports multiple search terms separated by commas — lines matching any term are returned. A comma inside a number is kept: "3,007" or "464,391" is searched as one figure. Args: cik: CIK of the filer (e.g. "0000320193") accession_number: SEC accession number (e.g. "0001045810-25-000023") query: Search terms, comma-separated (e.g. "singapore,revenue" or "risk factors") filename: Optional document filename. If omitted, searches all readable documents. context_lines: Lines of context around each match (0-50, default 10)
search_filings
ChatGPTFull-text search across all SEC filing content. Returns matching filings with relevance-ranked results, highlighted snippets, and facets by form type and year. Use this to find filings mentioning specific topics like "revenue growth", "cybersecurity risk", "stock buyback", etc. Args: query: Search query (e.g. "revenue growth", "cybersecurity risk factors") form: Filter by SEC form type (e.g. "10-K", "10-Q", "8-K") cik: Filter by company CIK number year: Filter by filing year (e.g. "2025") limit: Results per page (1-100, default 20) page: Page number (0-indexed)
search_private_offerings
ChatGPTSearch Form D filings: private placements and private funds (exempt offerings under Regulation D, Section 4(a)(5) and the Investment Company Act 3(c) exclusions), since 2009. Each result is one filing: issuer (CIK, name, city, state, entity type, year of incorporation), industry group and fund type, revenue or net-asset-value range, exemptions claimed (code and meaning), securities offered, date of first sale, total offering amount (null with offering_amount_indefinite=true when the filer wrote "Indefinite"), total amount sold so far, remaining, minimum investment, number of investors and non-accredited investors, sales commissions, finders' fees, proceeds paid to executives/directors/promoters, and counts of related persons and sales recipients. totals sums the WHOLE match, not just the page, and totals.by_category breaks it down by industry group and fund type. Categories are SEC's own (Form D Item 4), as the filer chose them; filter with exactly these (case does not matter, several comma-separated). industry: Banking & financial: Commercial Banking, Insurance, Investing, Investment Banking, Pooled Investment Fund, Other Banking and Financial Services Real estate: Commercial, Construction, REITS and Finance, Residential, Other Real Estate Technology: Computers, Telecommunications, Other Technology Health care: Biotechnology, Health Insurance, Hospitals and Physicians, Pharmaceuticals, Other Health Care Energy: Coal Mining, Electric Utilities, Energy Conservation, Environmental Services, Oil and Gas, Other Energy Travel: Airlines and Airports, Lodging and Conventions, Tourism and Travel Services, Other Travel Single: Agriculture, Business Services, Manufacturing, Retailing, Restaurants, Other fund_type (given only when industry is Pooled Investment Fund): Hedge Fund, Private Equity Fund, Venture Capital Fund, Other Investment Fund. Issuer size (Form D Item 5), also SEC's own values: revenue_range (operating companies, and VC/PE funds): No Revenues, $1 - $1,000,000, $1,000,001 - $5,000,000, $5,000,001 - $25,000,000, $25,000,001 - $100,000,000, Over $100,000,000, Decline to Disclose, Not Applicable nav_range (hedge funds and other investment funds): No Aggregate Net Asset Value, $1 - $5,000,000, $5,000,001 - $25,000,000, $25,000,001 - $50,000,000, $50,000,001 - $100,000,000, Over $100,000,000, Decline to Disclose, Not Applicable About 82% of filers choose Decline to Disclose for revenue (nearly all funds for NAV), so a size filter finds only those that disclosed; for "smaller raises" filter on amounts instead (max_sold, max_offering) and say which you used. Examples from real filings: Founders Fund Growth IV, LP -> Pooled Investment Fund / Venture Capital Fund Blackstone Private Equity Strategies Fund L.P. -> Pooled Investment Fund / Private Equity Fund Renaissance Institutional Diversified Global Equities Onshore Fund -> Pooled Investment Fund / Hedge Fund BlackRock Private Equity Co-Investments 2025 (Lux) SCSp -> Pooled Investment Fund / Other Investment Fund Alexander Series Trust (a cash fund) -> Pooled Investment Fund / Other Investment Fund New York Life Insurance & Annuity Corp -> Insurance Blackstone Real Estate Income Trust -> REITS and Finance Blue Owl Real Estate Exchange VI DST (a real estate syndication) -> Other Real Estate Space Exploration Technologies (SpaceX), X.AI Holdings, Databricks -> Other Technology Koios Medical -> Biotechnology The filer picks the category, so similar vehicles can land in different ones (a single-deal SPV may be Pooled Investment Fund or Other; some real estate syndications tick "pooled investment fund interests" as the security while choosing a real estate group). The securities field (pooled_fund, equity, debt...) is a second, independent signal. How to read the numbers: - A Form D/A restates the whole form, and total_amount_sold is cumulative. amendments="latest" (default) returns each offering once, at its newest filing, so sums never double count; "new" returns only original notices filed in the date range (new raises); "all" every filing. - For a Pooled Investment Fund, abo…
verify_figure
ChatGPTCheck that a number you are about to state appears in a filing, however it is written. Every number is compared as its digits alone (separators, decimal point, sign, $ and % dropped, zeros trimmed): "416,161", "416.161" and "416,161,000" are all 416161. Your figure must be one of the filing's numbers times a power of ten, allowing for rounding to the digits you give: "$416.2 billion" matches 416,161 (in millions), "75%" matches "75.0 %". Searches every document in the filing (main document and exhibits) unless filename is given. Candidates are ranked by exactness and by your label's words appearing in the number's row label or lead-in text. Returns found, match_count, and up to 10 matches, each with its document, label, context and a highlight_url that marks that exact number in the beatandraise.com viewer. Matching ignores magnitude, so confirm each candidate's label, period and scale from its context, then cite its highlight_url. found=false means no number in the filing has those digits: do not attribute the figure to the filing. If you calculated it (a growth rate, a margin), say so and cite the inputs instead. Args: cik: CIK of the filer (e.g. "0000320193") accession_number: SEC accession number value: The figure as you would state it (e.g. "$416.2 billion", "75.0%", "-1,234"). Needs at least two significant digits, except percentages. label: What the figure is (e.g. "total net sales", "GAAP gross margin"). Used for ranking. filename: Optional single document to check instead of the whole filing
verify_quote
ChatGPTCheck that a figure or sentence really appears in a filing document, and get a link that highlights exactly where. Use this before stating any number or quotation from a filing. Analysts need to see the figure in the source; a link to the whole document is not enough. Quote the text as it appears in the document: a sentence, or a table row's label and figure ("Total net sales 416,161", "Gross margin 75.0%"). Words and numbers must match exactly. Spacing, table-cell breaks, "$" and curly quotes may differ. Returns found (true/false) and, for each occurrence, its surrounding context, a highlight_url (beatandraise.com viewer, marks that exact span, including across table cells), and where possible a sec_fragment_url (sec.gov, scrolls to the text in Chrome/Edge/Safari). If there are several occurrences, pick the one whose context fits the claim, or quote more of the row to pin one. If found is false, do not present the text as a quotation from this document: numbers_elsewhere shows where its figures do appear, which usually reveals a misremembered figure or the wrong document. Args: cik: CIK of the filer (e.g. "0000320193") accession_number: SEC accession number (e.g. "0000320193-25-000079") filename: The document the text came from (e.g. "aapl-20250927.htm", or an exhibit such as "q4fy26cfocommentary.htm"), as listed in content_object or in a search result quote: The exact text to find (max 1000 characters)
