I m building CSV Repair & Deduper and want to understand what proof matters before someone imports cleaned data into a CRM, store, or accounting system.
Which would give you the most confidence: a before/after row count, a list of duplicate groups, a change summary, column validation, or something else?
The tool runs locally in the browser, but local processing alone does not prove the output is correct. I d rather prioritize the audit evidence people actually use. Please don t share real customer data just the checks that matter in your workflow.
Hi Product Hunt 👋
You can use part of this right now without paying or signing up: drop a CSV on the product page and it will tell you what's actually wrong with it — exact duplicate rows, near-duplicates that exact matching misses, malformed rows, broken quoting, encoding damage, and columns whose leading zeros Excel is about to destroy.
The part I care most about: your file never leaves your browser tab. Don't take my word for it — open DevTools → Network before you pick the file and watch nothing carrying your data go out. That's the whole reason I built this. Plenty of CSV cleaners want you to upload a customer list to someone else's server, which is a nonstarter for lead lists or contact exports. There are good client-side ones too — what I wanted specifically was duplicate review: seeing the near-duplicates that exact matching misses and deciding on each one myself, rather than a tool silently dropping rows.
The free check reports problems. The $19 download is the thing that fixes them: review each suggested removal, edit rows inline, undo/redo, and export the cleaned CSV. Same offline single-HTML-file architecture, no subscription, 14-day money-back guarantee.
Honest limitations, also on the page: it's a browser tool, so very large files get slow and the free in-page check caps at 20 MB. Near-duplicate matching is similarity scoring, not judgement — you review every suggested removal, nothing is deleted for you.
One thing I'd genuinely like feedback on: the near-duplicate threshold. My first version reported ~200 "near-duplicates" in a completely clean file, because clean1@x.io and clean2@x.io are 92% similar as strings. Sequential identifiers are now explicitly excluded. If you run it on a real export and it flags something it shouldn't, tell me — a dedupe tool that cries wolf is worse than no tool.