Turn a messy Excel sheet into a reviewed cleaning run

For XLSX uploads, the product reads the workbook structure and asks you to select one sheet for the project. It does not assume that the first tab contains the right table, nor does it combine sheets with different schemas. That decision becomes part of the project context shown in the real workspace.

The selected sheet is then profiled using the same deterministic process as a CSV. Column counts, row counts, inferred types, missing values, duplicate rows, and warnings establish a factual baseline before plan generation.

Spreadsheet cleanup is easier to audit when a change is expressed as a rule that applies to a column or row condition. The plan can normalize inconsistent text, parse dates and numbers, cast types, rename fields, remove empty rows or columns, drop exact duplicates, and flag outliers using only supported operations.

You can inspect all suggestions together or move through them step by step. The views do not create different plan types; they are two ways to review the same frozen rule set. High-risk operations remain blocked until they are explicitly acknowledged.

Execution always writes a new result. The original workbook is not overwritten, and failures do not replace it with a partial file. The report records what the engine actually applied, including warnings and row or cell impact where the Worker reports it.

The current beta supports one selected output format per run plus PDF and JSON audit downloads. Cross-sheet recipes, persistent dictionaries, and automatic schema mapping remain deferred until the supporting backend exists.