When the site matters more than the sample
We spent two days in Lesotho recently, deploying Labstudio for WASCO, the Water and Sewerage Company. Lovely country, and a
deployment that turned out to be less about software configuration than about agreeing on what
the lab is actually measuring.
Three labs in one
The WASCO lab does three kinds of work at the same time:
- Quality assurance on treated water.
- Process control across the treatment and reticulation network.
- Ad-hoc tests for external clients.
Those are not variations of the same workflow. They disagree about what the primary record is.
Sample first, or site first
In a client facing lab, the sample is the centre of the universe. It arrives with a traceable
number, that number follows it through registration, testing, review and reporting, and the
certificate is about that one sample. Traceability of the individual sample is the product.
QA and process control invert this. Nobody at a treatment works cares about sample number
24-08-0417 on its own. They care whether the clear well has been drifting for three weeks, or
whether a reservoir has been out of spec twice this month. The site is the thing being monitored.
The sample is just one observation of it.
In process control the sample is evidence. The site is the subject.
A system that only knows how to be sample first will happily store every result and still fail to
answer the question the plant operator is actually asking.
The naming convention
To make the site first view work, we needed a naming convention that identifies sites and sample
points consistently, and that holds up in two very different places:
- Machine to machine interfaces, where the identifier has to be stable, parseable and free of the
kind of drift that creeps into hand typed text. - Manual data capture, where a technician under time pressure has to recognise the point on a
sheet or a screen without looking anything up.
Those two requirements pull in opposite directions. Computer friendly wants short, rigid and
structured. Human friendly wants the name people already say out loud in the plant.
The afternoon it took
On paper this is a trivial piece of design. In practice it took a full afternoon on site with
everyone in the room, because a naming convention is never only about naming. It surfaces:
- Legacy names that are already printed on tags, logbooks and historical records.
- Local names that differ from the names on the official schematics.
- Disagreement about where a site ends and a sample point begins.
- The question of what happens to years of existing data if the naming changes.
Getting all the parties to agree took longer than any of the technical work, and it was the right
place to spend the time. Every interface, report and trend built afterwards depends on that
agreement holding.
What we took away
If you run a mixed lab, decide early which record is primary for each stream of work. A LIMS can
carry both a sample first and a site first view, but only if the site structure is designed
deliberately rather than inherited by accident from the sample register.
And get the naming convention agreed in a room with the people who will type it, not in a
specification document written afterwards.