> ## Knowledge Base Index
> Fetch the complete knowledge base index at: https://help.lido.app/sitemap.xml
> Use this file to discover available pages before exploring further.
> Pure-Markdown content can be obtained by appending a '.md' suffix to the content URLs listed in the sitemap (without the trailing slash).

# Troubleshooting extraction: missing rows, blank results, wrong values

Match your symptom below. After any fix, remember: **re-extracting the same document within 24 hours is free.**

## Only one row extracted (or only the first page)

Turn ON **"Extract multiple rows per document"** in the Data Extractor. This is the single most common fix. Without it, Lido returns one summary row per document instead of one row per line item / person / transaction.

## Extraction "succeeds" but the sheet is blank

- **Check your column headers.** Columns are your instructions - they must exist before extracting, and names referenced in Extra Instructions must match exactly. A single typo (for example Employee_Nam instead of Employee_Name) can produce fully blank rows.
- Your instructions may be too strict: if you filter by a page title, the match must be exact.
- Scroll down - data sometimes lands below empty rows.
- If the file previews fine but always extracts blank, the PDF may be corrupted: re-download or re-save it and upload again.

## Wrong values

- **Numbers reformatted** (leading zeros dropped, suffix letters changed, 00123 becomes 123): set the column format to **Plain** - click the **123 dropdown** on the column header and choose Plain.
- **Negative signs lost** (bank statement debits showing positive): add to Extra Instructions: "Preserve the original sign of all amounts. Debits are negative, credits positive. Do not convert signs." and set the column to Plain.
- **Dates changed** (year replaced with the current year, inconsistent formats): add: "Extract dates exactly as printed on the document. Never change the year. Output all dates as MM/DD/YYYY."
- **Wrong field picked** (employer name instead of employee name, etc.): name the column precisely and describe where the right value appears: "Employee_Name is the person listed at the top of each pay section, not the company in the header."
- **Sub-rows mixed together** (tables with R / O / VCX style sub-lines): describe exactly which rows to output: "For each employee output one row per earning type. Include rows labeled R, O, and VCX. Ignore NET rows."

## Checkboxes and scanned forms

- Add **[at]ocr_mode: vision** on its own line at the bottom of Extra Instructions (use vision_only if no text in the PDF can be selected).
- To stop the AI guessing unticked boxes: "Only output an option if there is a visible mark in its checkbox. If no option is marked, leave the field blank."

## Old or deleted data reappearing in exports

Empty-looking rows at the top of the sheet can hide old data. Click into the sheet, press **Ctrl+A then Delete** (Cmd+A on Mac) to clear all data - **headers are preserved** unless you select only row 1. Do this between batches or monthly runs.

## Information spans page breaks / large files

See "Extracting large documents (150+ and 1,000+ pages)" for [at]parallel directives and the Split File workflow.

## Still stuck?

1. Ask the **AI assistant** (lightning bolt icon, top middle of your sheet) - it can rewrite your instructions against your actual document.
2. Contact support in chat with: the **URL of your sheet**, a screenshot of the wrong output, and which document/page shows the problem.