Job near-duplicate-files-found-with-a-dry-run-list-and-nothing-deleted · revised 12 October 2026
Find near-duplicate files in a folder, with a list to review and nothing deleted
A script lists groups of files that are the same or nearly the same, and says how they differ. You review the list. The script has no way to delete, move or rename anything.
You might be seeing
- Several files with nearly the same name and content
- Nobody dares delete a file because it might be the latest version
No passwords, keys, card details or admin invites needed to start.
What usually happened
Exact copies are easy to find, but near-copies are not: a contract with one clause changed, an export re-run with new rows, a document saved in two formats. Two files from the same template with different names or amounts look alike and must not be treated as copies. A tool that deletes by itself will remove one that mattered.
Who it’s for: Office manager or operations lead with a shared drive full of copies, such as "final", "final v2" and re-exported spreadsheets, who is afraid to delete the wrong one.
Usually starts when: The drive is full of near-copies, nobody knows which version is current, and an earlier clean-up deleted something people needed.
The result: A script you run on a copy of a folder lists groups of exact and near-duplicate files with the method used, a similarity figure, what differs and which file is newest. Files from the same template that differ in names, amounts or dates are listed as "similar, differ in" and never as duplicates. The script has no deleting, moving or renaming code, and your person decides everything.
Check whether this job fits
Answer these without sending files, tickets, code or logins. Nothing is submitted unless you choose to contact us.
What you get
- The groups list with differences
- The look-alike list
- The review file with a decision column
- The script and the command to re-run it
- A note on what the figures mean
Included
- One folder tree of up to 5,000 files: documents (PDF, Word, text) and exports (CSV, spreadsheets)
- Exact copies by file fingerprint, near-copies of documents by text overlap and near-copies of exports by row and column overlap
- Groups with the method, the similarity figure, the newest and largest file, and the lines or columns that differ
- A separate list of look-alike files from one template that differ in key fields
- A dry-run review file with a column for your decision, and nothing applied
Not included
- Deleting, moving, renaming, archiving or merging any file
- Deciding which file is the right one to keep
- Reading scanned images or handwriting
- Files held in a live drive: you run the script on a copy
- Personal or special-category files, unless you run the script on your own machine and send only the lists
- Legal records holds or retention decisions
How we know it’s done
Agreed with you before work starts. Each check produces evidence you keep.
A script search of the delivered code finds no call that deletes, moves, renames or writes a file outside its own output folder, and a fingerprint of the folder before and after a run is identical.
Evidence: The search output and the two fingerprints.
At least 36 of the 40 planted duplicate groups appear in the list, including at least 18 of the 20 you keep from us until the run. One wrong classification of a held-back look-alike as duplicate fails the whole test; these planted cases are a smoke test, not a general detection-rate claim.
Evidence: The planted-group table.
None of the 10 look-alike pairs (same template, different names, amounts or dates) is listed as a duplicate, and each is listed as "similar, differ in" with the differing fields. One look-alike listed as a duplicate fails the whole test.
Evidence: The look-alike table.
After the matching rules are frozen, you randomly select and label 50 output groups, keeping your labels from us until acceptance; at least 46 must be true duplicates or near-copies under your written rule (observed 92%; exact one-sided 95% lower bound, rounded down: 82%). One group violating a never-duplicate rule fails the whole test.
Evidence: Your marks beside each group.
A second run gives identical groups, and your team re-runs the command you are given and gets the same lists.
Evidence: Both outputs compared.
Sign-off. You re-run the acceptance checks with the steps we hand over, read the results and sign off in writing. Payment follows sign-off.
If it fails. If an agreed check fails, you do not pay for this scope. We hand over what we found and agree whether to stop or re-quote; no surprise work.
When it fits, and when we stop
It fits when
- A copy of the folder, or permission for you to run our script on it yourself
- A named person who reviews every group and decides what to do
- A rule from you on which types of file never count as duplicates
- 40 duplicate groups in a copy of 500 files, with 20 groups kept from us, plus 10 look-alike pairs with 5 kept from us until the test run
We stop and tell you if
- You want files deleted or moved automatically
- The folder holds legal-hold records you cannot copy
- The files are mostly scans of paper with no text
- You will not review the list yourself
What could go wrong
Nothing is changed, so there is nothing to undo. The script has no code that deletes, moves or renames, and our copies of your files are deleted when the job ends.
Scroll the table sideways to read it all.
| Risk | How we handle it |
|---|---|
| Two different files are listed as copies and one is deleted. | The script cannot delete. Look-alikes from one template are listed apart, and any look-alike listed as a duplicate fails the whole test. |
| A near-copy is missed. | The report states how many planted groups were found and says a miss can happen. |
| Files hold personal data. | You can run the script yourself and send only the lists. By default no file content goes to a model. |
| The newest file is not the right one. | "Newest" is shown as a fact about dates only. Which file is right is your decision. |
A separate automated reviewer checks the work and the test results before hand-over. Your named person reviews every group and decides what to keep, archive or delete, outside this job. No human supervisor from us is included unless your agreement names one, and at launch the work is largely automated.
How we deliver
We arrange the work and independent review, then show you the result against the agreed checks. You keep authority over your systems.
- Agree the folder, file types, never-duplicate rules and reviewer
- Receive a copy, or the script is run by you; plant test files
- Fingerprint every file and compare text and row overlap
- Separate look-alikes that differ in key fields from true near-copies
- Check the script has no delete code; independent automated review
- Send the lists, review file, script and note
This is a one-off job, not emergency cover or a subscription. We confirm eligibility, price, start window and delivery date in writing before you accept, and work starts only after the agreed inputs are in place. An enquiry creates no charge or booking.
Need to keep it working?
Discuss a repeat run in six months, after you have acted on the first list.
Ongoing work is separately scoped and quoted: no monitoring, response-time guarantee or automatic subscription is included in this job.
Mention the responsibility you need in your enquiry.
What you can check
This is a new service. We have not delivered this job for a client yet.
Other ways to get this done
- Run a free duplicate-file finder on a copy of the folder and review its list yourself.
- Move older versions into an archive folder by hand, with a dated name.
Questions
Does it delete duplicates?
No. It has no code that deletes, moves or renames, and we show you the search that proves it.
What about two invoices from the same template?
They are listed as "similar, differ in" with the fields that differ. They are never listed as duplicates.
Does my content go to an AI?
Not by default. Matching uses fingerprints and text overlap. An optional note on how two files differ can be switched on, and then only the differing lines go to the provider named in the terms.
Does a pass mean it will be right next month?
No. A pass describes items like the test sample on the day of the test. It is not a promise about future items, and it does not replace your own checking.
Send an enquiry
Send us
- Roughly how many files and which types
- Where they are stored
- What a wrong delete would cost you
- No logins, keys, customer data or payment details in the first enquiry
Later, once you agree
- A copy of the folder tree, or your confirmation that you will run our script on it
- Your never-duplicate rules
- The planted test files you add to the copy
You keep your drive and every login. We work on a copy you send through a secure route, or you run our script yourself so no file leaves your machine and you send us only the lists. The script cannot change any file.
Email fallback: open your mail app
If website submission is unavailable, review and send the fallback email yourself. An email fallback is not a website receipt. Or write to hello@syntheticindustry.ai with “near-duplicate-files-found-with-a-dry-run-list-and-nothing-deleted” as the subject.