Data quality
Duplicates rarely announce themselves. The same pump gets registered twice under two spellings, two teams report different totals for the same fleet, and nobody can say which number is the real one. Uniqueness means every real thing exists once in your data, and everyone is looking at that one.
What it means
Uniqueness asks a simple question — does each real thing, a pump, a vessel, a meter, a work order, appear exactly once in your data? When the answer is no, every other dimension gets harder to trust, because accuracy and completeness are being measured against a record that has a twin somewhere else.
Duplicates are almost never created on purpose. They arrive when the same asset is registered by two teams, when a supplier sends a list that overlaps one you already have, or when a name is typed with a space in a different place. The data looks fine on the surface, and then the totals do not add up.
What it costs
A duplicate never breaks anything loudly. It quietly inflates a count, splits a maintenance history in two, and lets two people leave the same meeting with different numbers for the same operation.
Finance reports one asset count, operations reports another, and the difference turns out to be records that exist twice under slightly different names. The meeting is then spent reconciling spreadsheets instead of deciding anything.
When a pump exists twice, half its work orders attach to one record and half to the other, so neither shows the full picture. A failure pattern that would be obvious across the whole history stays invisible in two partial ones.
Duplicate records generate duplicate tasks — two inspections raised for one valve, or two orders for one spare part. The mirror image is worse: a duplicate gets closed, everyone assumes the job is done, and the real record is still open.
Our approach
Uniqueness is not solved by cleaning up afterwards. It is solved by giving every real thing one identity in one shared model, so there is nowhere for a second copy to live.
Every asset, location and measurement point is described once in a shared model of your operation, and everything that refers to it points at that same description. A second version of the same pump has no separate place to exist.
Because things are connected — this sensor belongs to this pump, which sits on this platform — a near-duplicate gives itself away by where it sits rather than by how it is spelled. Two records claiming the same position on the same asset are visibly the same thing, even when their names disagree.
When two records turn out to be one, resolving them should join their histories rather than force a choice about which years to keep. Everything that was true of both records stays attached to the single record that survives.
The other dimensions
Quality is not a single number. Each dimension fails in its own way, and each one is worth understanding on its own terms.