Repeat copies are caught as they arrive. Nothing is deleted automatically, so the signed original is never swept up with the clutter.
The risk in cleaning up a repository is not missing a duplicate. It is deleting the one signed copy.
Every cleanup exercise runs into the same problem: the file that looks redundant is sometimes the executed original, and the one that looks authoritative is sometimes a draft somebody renamed. Detection is worth having only if it can tell those apart.
Repeat copies of a document are detected as it arrives rather than accumulating quietly across folders for someone to find later.
Matching works on document contents, including text read from scanned and photographed pages, so a file renamed on the way between two folders is still recognised as the same document.
Every document carries a draft, active or superseded status. A superseded version is part of the history rather than clutter, and is not treated as a redundant copy of the one that replaced it.
A detected match is surfaced with the context needed to judge it — owner, source, status and where each copy came from. What happens next is chosen by someone with the rights to choose it.
The action taken on a group goes into the same tamper-evident history as everything else, so a cleanup can be explained afterwards rather than merely trusted.
Four groups the system would surface. The two it declines to call duplicates are the ones worth reading — they are what separates a cleanup you can defend from one you cannot.
The same document, saved twice under different names by two departments that did not know the other held it.
What happens
Both are shown as one group with their owners and sources. Someone decides which is the record to keep; neither disappears on its own.
One document that arrived four times because each recipient filed their own copy of it.
What happens
The repeat arrivals are flagged as they come in rather than after the folder has four of them.
Two versions of one agreement, and the older one is the record of what was in force at the time.
What happens
Left alone. A superseded document is history, not clutter, and status is what keeps it from being read as a redundant copy.
Nearly identical text, but only one of them is evidence of anything.
What happens
Left alone. This is the case that makes automatic deletion the wrong default no matter how confident the match looks.
Nobody sets out to store the same document eleven times. In each of these it happened as a side effect of people doing something reasonable.
An old server, a stack of external drives and a cloud folder all hold overlapping copies of the same archive. Detecting the repeats as material comes in stops the new repository inheriting the old one’s mess.
One head-office document ends up filed independently in a dozen branch folders, and none of them knows the others exist. Catching the repeats on arrival keeps one record authoritative instead of twelve equally plausible ones.
An attachment gets saved by everyone on the thread, each into their own folder and under whatever name their machine gave it. Comparing contents rather than filenames is the only thing that recognises those as one document.
Cleanup is easy to do and hard to defend once someone asks what happened to a particular file. Because resolutions are recorded rather than silent, the answer is retrievable instead of remembered.
No. Nothing is deleted, merged or overwritten automatically, no matter how confident a match looks. The system detects and presents; a person with the rights to act decides. That division is deliberate — the cost of a wrong automatic deletion is unbounded, and the cost of a duplicate sitting there one more day is not.
The safeguard is not clever matching, it is that removal is never automatic. On top of that, documents carry an explicit draft, active or superseded status, so a version that is part of the record is distinguishable from a redundant copy before anyone is asked to decide anything.
No. A superseded document is the record of what was in force at the time, which is exactly what you need when someone asks what the rule was two years ago. Status keeps it in the history rather than in the cleanup pile.
Yes. Scanned and photographed pages are read on arrival, so matching works on what the page says rather than on the filename. Two photographs of the same register page, saved under different names, are recognisable as the same document.
Yes. Comparison works on document contents rather than filenames, which is the point — a file that travelled between folders picking up a new name at each stop is the single most common way one document becomes four.
No. A group only contains records that reader can already open. Documents stay private to whoever added them until deliberately shared, so a match cannot be used to work out what is stored elsewhere in the organisation.