CSV cleaner

A CSV cleaner is useful when the structure of a table is sound but selected values need consistent treatment. A product name might have spaces at both ends, a category might alternate between upper and lower case, or a source system might write N/A to indicate a missing description. These are different decisions. TableWorkbench lets you choose the affected columns and put explicit cleaning rules in an ordered sequence. Importing a file alone does not make any of those decisions for you.

Start by choosing a CSV, TSV, or delimited text file. You can also paste cells copied from a spreadsheet. In the parsing dialog, check the delimiter, quote character, header setting, and any initial records you need to skip. Parse and preview the complete input, then inspect both the column names and representative values before confirming import. If a quoted field is malformed or a record has extra fields, the import stops instead of silently discarding data. Short records can be padded only through the visible confirmation option.

The column checkboxes define the scope for your cleaning sequence. Begin with a small set of fields whose meaning you understand. An identifier column should usually stay outside a rule intended for descriptions. Add a cleaning step, choose its action, and repeat when more than one action is needed. The up and down controls change order; disabling a step leaves its settings available without running it. Run and review performs the selected sequence over all records, while the grid presents manageable pages of the result.

Order matters. If an exact missing marker is written as a space, N/A, and another space, replacing N/A before trimming will not match it. Trimming first can expose the marker for a later exact replacement. Conversely, uppercasing before a case-sensitive literal replacement can change whether a word matches. Review the sequence as a short set of instructions rather than treating all options as interchangeable cleanup switches. The changes report lists the affected column, previous value, resulting value, and original record location.

Trim edges removes surrounding ordinary spaces, tabs, carriage returns, line feeds, nonbreaking spaces, and full-width spaces. It does not remove every invisible Unicode character. Collapse consecutive ASCII spaces reduces runs of the ordinary space character inside a cell; it does not collapse every kind of Unicode spacing or every line break. Remove U+200B addresses that specific zero-width space. It does not promise to remove joiners, directional controls, or other invisible characters that may be meaningful in written language.

Lowercase and uppercase use the browser's Unicode casing behavior and change the stored text in selected cells. Literal replacement looks for exactly the text you supply. Missing-marker replacement is an exact cell-value rule, so an empty string, the word NULL, the word false, and the character zero remain distinct until you explicitly choose otherwise. Missing cells created by confirmed padding are represented separately inside the workbench, although ordinary CSV output cannot preserve every internal distinction between a missing cell and an empty field.

Imagine a file with SKU, Product, and Note columns. SKU 001 has the product text “ BLUE PEN ” and the note “ keep this spacing ”. Select Product only, then add Trim edges, Collapse consecutive ASCII spaces, and Lowercase in that order. The product becomes “blue pen”. SKU stays 001 and the Note column retains its original spacing. This narrow scope is useful when descriptions need consistency but notes are evidence that should not be rewritten.

Add another record whose product is N/A. That text does not disappear merely because it resembles a missing-value marker. To clear it, add an explicit missing-marker step and leave the replacement empty. If you also select Remove entirely empty records, the tool checks the whole record: an empty product does not make a row with a populated SKU empty. Removed records remain available as a separate report. Empty-column removal considers only selected columns and reports how many were dropped from the result.

The original view provides the input used for this run. The result view shows the transformed table, and the cell-change report explains modifications. A grid search helps locate examples but does not define what gets exported. After checking the complete counts and enough representative changes, apply the result to the workflow. Subsequent tools receive that applied table without requiring a download and re-import. Undo recomputes the prior workflow state; redo restores an undone step when it remains available.

Changing a rule or input marks an existing result stale. A stale result cannot be downloaded as though it reflected the new settings. Run again, inspect the result, and then open export. Choose the report or result explicitly, set CSV or TSV, decide whether to include headers and a UTF-8 BOM, and review spreadsheet protection. The export verification step rereads the generated text and compares it with the intended output. Download creates a new copy; it never opens or overwrites your source file.

This tool handles text data, not an Excel workbook with formulas, styles, charts, or macros. Export a relevant sheet to UTF-8 CSV first when your source is a workbook. The workbench does not infer dates, calculate formulas, repair a corrupted encoding by guessing, or reconstruct leading zeros that were already lost elsewhere. Its per-file, record, column, cell, field-length, and task-time limits are visible safeguards. A phone may need much smaller data than a desktop even below those limits.

If the preview shows one giant column, check the delimiter. If the first product appears as a column name, revisit the header setting. If a cleaning step reports no changes, inspect the actual cell text, the selected columns, and the order of rules. A character that looks like a space may not be covered by the chosen action. The cell-detail view helps inspect long values without widening the entire page. Do not keep broadening the cleanup scope merely to make the modification count increase.