One workflow · Browser-local checks · Your own approved AI tool
AI Workflow Checker
Repeat product-catalogue categorization with a named, versioned set of rules. Choose the minimum columns to share, review the exact CSV export, take it to your own tool manually, then verify the returned file against your original. This site makes no model calls and uses no website AI credits.
The checker can flag missing, extra, duplicated or unknown IDs, schema changes, altered exported prices and protected fields, and category labels outside your list. It cannot decide whether an allowed category is semantically correct or whether the original catalogue is true. It does not overwrite your original or automatically publish a catalogue.
Files and run evidence stay in browser memory unless you explicitly download them. Only the “Save rules in this browser” button persists a rule set; it does not save CSV data. Column minimization is not anonymization or a compliance guarantee: free text can still contain sensitive information. The surrounding site may request assets; browser extensions and external services are outside this tool’s control.
New to this workflow? Read the worked product-catalogue guide, with downloadable synthetic inputs and explained outcomes.
1. Load the original and choose a job
Use a UTF-8 comma-separated CSV with a header, at least one product, unique nonblank IDs, a price column and an existing category column (blank categories are fine). Add a blank category column in your own editor if needed. Limits: 1 MiB, 5,000 logical records including the header, 100 columns and 100,000 cells. Column names must be at most 120 characters.
No original loaded.
Reset clears working files, previews, results and previous-run evidence. Rules explicitly saved in this browser remain until you choose “Delete saved rules”; downloaded files remain on your device.
2. Name, version and configure the rules
4. Reimport and verify
No returned CSV loaded.
The synthetic return belongs to the worked example only. A file or rule change clears results; source and rule changes also require a new export preview.
5. Compare another run of this job
No previous run retained.
Checks apply only to the selected job rules and exact CSV strings. Passing does not certify category meaning, original truth, model quality, privacy or compliance. No AI is called. Column minimization does not guarantee privacy; free text may contain sensitive data. Run evidence contains both CSVs; imported results are recomputed, not authenticated. This tool never reconstructs or overwrites a full production catalogue. Reset clears this page’s working memory; only an explicit save stores rules. Site assets may make ordinary requests; extensions and downloaded files are outside this tool’s control.
What repeat-run comparison means
Keep a verified run in this session or explicitly import its local evidence JSON. The tool recomputes the previous checks from its CSVs and rules. A comparable run requires identical original file text and identical full rule settings, not merely the same job name or revision. Changed inputs or rules are labeled as different contexts, and failure counts are never described as model-quality improvement.
Exact strings, bounded files, explicit limits
IDs, prices, currencies, dates and formula-like cells remain strings. There is no numeric conversion, date interpretation, fuzzy deduplication or formula execution. Original IDs must be unique and nonblank before export. Returned headers must match the selected export names and order. Fields kept local cannot be compared in the returned file and are reported as not checked.
Use AI Work Check for general before-and-after CSV comparison, or CSV Cleanup & Check for conservative cleanup. This workflow adds repeatable rules and a deliberate manual handoff; it is not a live AI integration, privacy detector or certification.