Clean up your tables.Make every change clear.

A file that opens as a table is not necessarily ready for its next destination. Supplier headers can disagree, descriptions can contain accidental spaces, identifiers can repeat, and a new export can quietly omit records present last month. TableWorkbench brings these preparation tasks into one browser session. Each tool answers a distinct question, while a shared workbench preserves the original inputs, makes rules visible, and produces new output copies that you can review before using elsewhere.

Begin with the task rather than the file extension. Cleaning changes selected cell text. Deduplication groups records by identity and applies a retention policy. Merging appends rows, while joining adds fields through matching keys. Comparison describes differences between snapshots. Mapping reshapes columns, filtering selects records, sorting changes their order, transposition exchanges matrix coordinates, and splitting produces separate output files. A list comparison handles membership of single-value items. Literal replacement provides a narrow way to edit known text.

The import step accepts CSV, TSV, delimited text, and cells pasted from spreadsheet software. Each file receives its own confirmation for delimiter, quoting, headers, and skipped introductory records. The supported encoding is UTF-8, with or without its byte-order mark. Other encodings should be converted in the source application before import. A filename extension alone does not prove that the contents form a valid table, so the parser checks the actual data and reports malformed records.

Every value begins as text. Product codes such as 001, long identifiers, and ambiguous dates are not converted because they happen to resemble numbers. Missing cells, empty strings, whitespace, and literal markers such as NULL have distinct meanings inside a preparation workflow. When output cannot preserve all those distinctions, the export explanation says so. This is a practical boundary of delimited text, not a reason to guess silently about what the source meant.

Import three supplier files and confirm their structures individually. A field called Product ID might correspond to SKU, but that relationship needs your decision. In the merge tool, map the relevant names and select a column strategy. Same columns enforces a common mapped schema; All columns retains different fields and introduces empty values where a source lacks one; Shared columns deliberately discards unshared fields after confirmation. The merged record count should equal the sum of the selected inputs.

Apply the merged table, then inspect duplicate SKUs. A repeated code with different prices is a conflict to review, not permission to average prices or choose a random row. Use Mark only first if you do not yet know which occurrence should survive. Once you accept a specific retention policy, apply it and compare the resulting catalog with the previous month's snapshot. The comparison separates additions, removals, modifications, and unchanged records, with old and new values retained in the modification report.

For a lightweight membership task, paste one item per line into the two list boxes. Choose whether case and surrounding whitespace should affect identity and whether empty items should count. The result provides common items, items unique to either side, and a distinct union. Repeated occurrences remain visible through original text and source locations, while membership counts treat a repeated matching value as one item. This is different from counting how many times each item occurred.

List order does not determine whether an item is common. An identifier at the beginning of one list can match an identifier at the end of the other. If you need to compare several fields of corresponding records, choose the CSV comparison tool instead. If you need to compare cell positions rather than identity, select that explicit mode. These distinctions matter because an accurate result for one task can still answer the wrong question for another.

A system-import workflow often starts by cleaning a limited set of descriptions or category fields. Review the cell changes and keep identifier columns outside transformations that do not apply to them. Use the column mapper to rename and arrange fields in the destination order, omit unnecessary data, and add intentional constant columns. Required-column checks verify the presence of specified headers; they do not certify that every row meets all of a destination's business rules.

Next, split the result by a maximum number of data records per file or by a selected category value. Each part can include a header, and a manifest records its name and size in records. Group values used in filenames are sanitized, with continuous numbering to prevent collisions. Inspect empty groups and confirm that every source record belongs to exactly one output part. Download a ZIP when the task produces multiple files, keeping the manifest with the data so the package is understandable later.

Running a tool computes a result for inspection. Applying that result makes it the current input for the next step and records the operation in the workflow. Undo and redo let you examine earlier accepted states. Changing rules makes an old result stale until a new run finishes. These boundaries prevent a preview based on yesterday's settings from being treated as the output of today's rule. The displayed record range is separate from the total processed count.

Export asks for the intended scope instead of assuming the current grid search defines it. You can choose the complete result or a supporting report, set CSV or TSV, control headers and UTF-8 BOM, and review formula protection. Protected output changes formula-like prefixes and reports how many cells were affected. Raw output preserves text but can be interpreted by a receiving spreadsheet. Generated text is reread to verify records and values before download. Your original file is never overwritten.

The application processes file content in the browser and has no table-upload backend. Opening pages and loading static resources still use the network, and the hosting provider can retain ordinary access logs. The workbench code includes no advertising, session replay, or analytics that receives file content. It does not execute imported formulas, macros, or scripts. A local-processing design is a concrete statement about the data path, not a claim of absolute privacy or zero network activity.

The session is held in page memory. Compatible tools share the general table session; specialised workspaces may require a separate handoff. Refreshing or closing the page clears in-memory work. Export useful results before leaving. This version has limits on bytes, records, columns, total cells, field length, output parts, and task duration. Those are protective ceilings, not promises that every phone or browser will handle the maximum comfortably. Dedicated tools support selected XLSX ranges, calculations, and date or number conversion. Check each tool’s supported inputs. Automatic persistent projects, cloud synchronization, and guaranteed offline use are not provided.