> For the complete documentation index, see [llms.txt](https://docs.filed.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.filed.com/reference/products/ai-review/understanding-flags.md). # Understanding flags ## Where a flag's rule comes from Every finding AI Review surfaces traces back to a specific check, not a generic pass over the return. These checks are the same protocol mechanism documented in the [Playbook](/reference/playbook/creating-protocols.md): firm-wide and personal rules that can be reviewed, edited in plain language, and turned on or off. AI Review-scoped protocols are what decide which checks run on a given review. This means a flag is never a black box. If a finding looks wrong or unexpected, the protocol behind it can be opened in the Playbook to see exactly what it checks for and adjusted or disabled if it does not fit the firm's judgment. ## Turning a check on or off Protocols scoped to the AI Review task type can be disabled from the Playbook without affecting AI Tax Prep, Chat, or other task types - see [Task types](/reference/playbook/creating-protocols.md#choosing-the-right-task-type). Disabling a protocol means it is skipped on the next run; it does not retroactively remove findings already on an existing set of leadsheets. Re-run AI Review after disabling a check to produce a clean set of leadsheets without it. ## Example: wages vs. Medicare wages One of AI Review's built-in checks compares Box 1 (wages) against Box 5 (Medicare wages) on a W-2. These two boxes are usually close but not identical - pre-tax deductions like a 401(k) contribution reduce Box 1 but not Box 5 - so the check looks for a difference that falls outside the normal range for those deductions rather than flagging any difference at all. This is useful because tax software does not perform this cross-check on its own: it accepts whatever values are entered in each box independently. A transposed digit or a data-entry error on either box will pass through undetected by the software, but AI Review catches it because it is comparing the return against the source W-2 evidence. The same evidence-based approach is why AI Review can catch situations involving a corrected W-2: if a corrected form supersedes an original, comparing the return only against the most recent version - rather than trusting whichever copy the return happens to be entered from - is what surfaces a mismatch. If a finding references box amounts that don't match your expectation, check whether more than one version of that document (original and corrected) exists in the binder. ## Reading a flag card Each finding card in the Leads tab shows: * The severity level (Critical, High, Medium, or Low) - see [Issue severity levels](/reference/products/ai-review.md#issue-severity-levels) * The form and line the finding is anchored to * A description of the discrepancy * The evidence the AI used to reach that conclusion * A comment thread for discussing the finding with colleagues See [Working through findings in the Leads tab](/reference/products/ai-review.md#working-through-findings-in-the-leads-tab) for the full row and issue views. ## When a flag doesn't apply Not every flag indicates an error. Severity reflects the AI's confidence and the significance of the pattern it found, not a final verdict - see [What do the severity levels mean?](/reference/products/ai-review/frequently-asked-questions.md#what-do-the-severity-levels-mean). Use the comment thread to record why a finding was determined to be a non-issue, then sign off the row to clear it. See [Signing off on leadsheets](/reference/products/binder/leadsheets.md#signing-off-on-leadsheets). --- # Agent Instructions This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com. ## Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question. Perform an HTTP GET request on the current page URL with the `ask` query parameter, and the optional `goal` query parameter: ``` GET https://docs.filed.com/reference/products/ai-review/understanding-flags.md?ask=&goal= ``` `ask` is the immediate question: it should be specific, self-contained, and written in natural language. `goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.