The Explainer Every mechanism, in plain language.

One system for the whole company

Assembly Floor covers everything you do with your own equipment and your own people — from a part arriving at the door to a device retired years later.

• What do we have, where is it, and is it any good?
• Can we take a job three months out without stranding another?
• Which devices are alive in the ground — and which went quiet?
• When something fails: what failed, why, has it failed before?
• Who is doing what today, across every site?

Your customer-facing systems stay customer-facing. Everything internal belongs here.

Nine stages, one record

Every device follows the same path. The job is to keep one unbroken record of it the whole way round.

The loop is the point. Without it, a device goes in the ground and you hear nothing until a customer calls. Close the loop and the device tells you it stopped, someone goes out, and what they find changes how you build the next one.

What this covers

Six areas of the business. The whole thing on one page.

Warehouse first

Where two groups disagree, the warehouse wins. It uses more of this system than anyone else.

The Ingester

Every packing slip goes through one scanner. From that single act, everyone who needs to know, knows.

Parts arrive for jobs and for research work, and the person who asked for them has had no way of knowing they're in the building. The Ingester reads the paperwork, matches it to an order, files it against the job — and tells whoever requested it, automatically.

Seven seats, seven jobs

Nobody uses all of this. Everybody uses their part — and each of them stops doing something they do today.

Red tag, green tag

Stock runs out. Often the answer is already sitting on the repair shelf.

If it can't be saved, the same path ends in a decommission — recorded with what was tried, by whom, and why.

From the floor to an order

The person who knows you're running out is the one using the last of it.

Nothing is ever ordered automatically. The work of preparing the order is automated. The decision to spend is not.

Packed by two, signed by both

Packing a box is a two-person job: one packs, one checks. That check moves off paper and onto a tablet.

A kit cannot ship until the checklist is complete and signed. It stays readable — from the kit, every device in it, and the job — years later.

Calibration and the sales pack

An instrument is only trusted if it can prove it measures correctly.

Calibration is captured at the bench and becomes part of the device's permanent record rather than a certificate in a drawer. A sales kit must carry its certificate of acceptability and its calibration certificate — and a device outside its calibration period cannot go into one. Those documents are generated from the records, and sales can pull the pack for any job without asking the warehouse to compile it.

One serial, one life

A device isn't a row in a stock list. It's an object with a history — and the history is what makes everything else possible.

Built on a named bench in a named batch. Shipped in a named kit, packed and signed by two people. Installed by a named technician at a named site. Transmitting, then quiet. Repaired, returned, redeployed. Eventually retired, with the reason recorded.

Read along that line and you can answer: how long it ran, which batch it came from, who touched it, and what the failure cost.

What broke, and why

“Antenna damage” in a box is a note. The same failure against a named part, counted across the fleet, is a diagnosis.

Two components cause most of the cost. You cannot see that from free text.

Is it the device, or the hole?

Some manholes eat equipment. Replace the device and it fails again, because the device was never the problem.

This only works because repair history hangs off the site, not the device — so it survives every swap.

When a device goes quiet

Without the watcher, a device dies in the ground and you find out when the customer rings.

The warehouse hears at the same moment as everyone else — so the replacement is built and tested while the visit is still being scheduled. Devices switched off on purpose never alert: a disposition gate separates outage from maintenance from dormancy, so an alarm means something every time.

Battery history

Battery is the most common failure recorded — and currently the least traceable.

Everything is signed

Not surveillance — the only way to tell a people problem from a process problem from a product problem.

A deliberate limit. A run of failures might be the sites, the season, or the batch. The system shows the question with its context. It never produces a league table.

Analytics, money and the awkward cases

Once every step is recorded, a set of questions that used to need a phone call answer themselves.

• Job endedEquipment back in the shop, or a rental finished — accounting told without chasing.
• WarrantiesWhat you owe the customer, and what the supplier owes you.
• Shipping historyEverything ever sent to a customer, with tracking.
• Test historyProof of what left the building, and in what condition.
• Certificates per jobCalibration and acceptability, assembled per shipment.
• Consumables per jobDesiccant, ties, hardware — or job cost is understated.

Customer-owned equipment is marked as theirs from the moment it arrives — it never enters your stock, your asset counts or your retirements, even though you service it with the same detail. Add carrier data and maintenance history together and cost per project stops being an estimate: parts, shipping, labour, travel and rework, per job.

The Foreman

Every number in the system, in plain language, without knowing where it's kept.

Charged on what it's used for, not per person. No licences to buy, and you can switch AI provider whenever you like — swapping is a config change, not a rebuild.

The memory it runs on

The Foreman isn't a chatbot bolted on. It runs on a memory layer every future agent will share.

• Append-onlyNothing is ever overwritten.
• SourcesAnswers carry sources — anything can be checked.
• Permissions firstRestricted data is never fetched.
• Behaviour is dataChange it with an edit, not a release.

It can live wherever you want it — including inside your existing Microsoft estate. That's a hosting decision, not a rebuild.

How it gets built

Warehouse first. Each stage is useful on its own.

• 1 WarehouseEvery site as one, kits, red tags, the shipping record.
• 2 The floorTask lists, the Ingester, low stock to purchase order.
• 3 CommitmentSales can see and hold what's free months out.
• 4 FieldOffline capture, scanned IDs, swap detection.
• 5 ShippingLabels, tracking, air cargo — visible to sales and accounting.
• 6 WatchingLive device data and the gone-quiet alert.
• 7 ReliabilityPassports, component failures, site diagnosis, cost.
• 8 IntelligenceThe Foreman across all of it.

Where this comes from

This explainer describes the finished system. It was written inside a working wastewater instrumentation manufacturer, and the platform is being built in that order of work — warehouse first, each stage useful on its own.

We publish this because you'd find out anyway — and because the fastest way to make it wrong is to show it to people who run floors. The current state always lives on the transparency page.