> ## Documentation Index > Fetch the complete documentation index at: https://docs.focusalpha.ai/llms.txt > Use this file to discover all available pages before exploring further. # Data quality and sources > How FocusAlpha keeps financial data correct: figures as the company filed them, continuous cleaning across the full history, and a link from each figure back to the filing it came from. Data quality comes first at FocusAlpha. A figure you get from the API should be the figure the company published, and you should be able to check it against the document it came from. ## Figures as the company filed them * **Nothing is converted or normalized away.** Statement rows carry the company's own reporting `currency` and `accounting_standard`; amounts are never converted to USD. * **A null means not reported, not zero.** When a company does not disclose a line, the field is null. We do not fill it from a substitute quantity or an estimate. * **Withheld rather than guessed.** When the filings disagree in a way the data cannot settle, such as a derived fourth quarter that the annual report contradicts, the value is withheld instead of served. * **Company speech stays separate from filed figures.** Management guidance, analyst consensus and outside research each have their own datasets. None of them is mixed into a statement figure. ## Continuous cleaning Financial statement data is cleaned continuously, and each round of cleaning is re-run across the full history, not just the newest filings. * **The statements must agree with themselves.** Assets equal liabilities plus equity, and gross profit, operating income and net income bridge line by line. Cash movement reconciles to the change in cash, and per-share figures match the share count they sit beside. A row that breaks an identity is investigated and corrected, not silently dropped. * **One basis within a series.** Share counts and per-share figures are restated onto the company's current share basis across splits, so years compare directly. A quarter is derived only from figures on the same basis as the year it belongs to. * **Restatements and amendments are applied.** When a company restates, or files an amendment, you get the corrected figure by default. * **Every defect becomes a check.** When we find an error, we correct it across every company it affects. The case also becomes a permanent check, so the same error cannot come back. ## Every figure traces to its source Where the market publishes its filings as documents, each row names the filing it was read from and links to it: | Market | What each row carries | | - | - | | United States (SEC) | `filing_url` and `accession_number` of the 10-K, 10-Q, 8-K or 20-F | | Japan | `filing_url` of the EDINET filing or the TDnet 決算短信, with `doc_id` | | Korea | `filing_url` of the original DART document | | Hong Kong | `filing_url` and `page_no` of the results announcement, for `source=filing` rows (2026 onward) | | Mainland China, Taiwan | The period and report it belongs to; mainland China rows also carry `announcement_id`, the filing that made the report public | The `source` field on regional rows says what kind of evidence stands behind a figure. `filing` means read from the published document, and `aggregated` means standardized statement data without a document reference. Filing-level evidence is being extended to more markets and more history. The same applies beyond statements. Guidance, earnings-release figures and 8-K events carry the quote or the document they were read from, and [semantic retrieval](/transcripts/semantic-retrieval) returns the passage it matched. ## Related pages What each dataset covers and how often it refreshes. Income, balance sheet and cash flow, as reported. This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.