The engine

Guardian Engine

The always-on intelligence layer for operations — designed to watch everything, learn continuously, and act before issues become problems.

In development· three checks run today, shown below. The wider design on this page — continuous observation, learning from every action — is not built yet.
What it says today

A finding, not an alert

An alert tells you a threshold was crossed. Your stock screen already does that — and it is why 28 units against a reorder level of 26 reads as healthy. Three checks run today. Here is what each of them says, and why a threshold could not have said it.

Critical Inventory

Supplier 9 — BAK-00484, 4 days of cover

Bakery item 484 has 4 days of cover. A replacement order takes about 10 days, so ordering today still leaves 6 days with nothing on the shelf.

Recommended: Raise an order to Supplier 9 for 184 units of BAK-00484.
Units shipped630
Days observed90
Leaving per day7
Available28
On hand46
Reserved18
Reorder level26
Days to zero4

Lead time is assumed at 10 days — no per-supplier lead time is recorded yet, and every finding that relies on it says so.

One system held every number needed for this, and none of them divided one by another.

Critical Warehouse Fleet

Calgary Depot loses a third of its vehicles on Thursday

Two of the site’s six vehicles are booked into the workshop on the same Thursday morning, while dock use there has moved from 88% to 94%. The two bookings were made weeks apart, by different people, in different systems. Neither system was wrong. Neither could see the other.

Recommended: move one of the two service bookings to the following week.
Vehicles at site6
Booked out2
Left running4
Dock use, was88%
Dock use, now94%

Capacity is counted from vehicles recorded against a site. Where that is not recorded, the site is reported as unknown rather than counted — see below.

No single system could have made this one. That is the whole argument for the product.

Could not check Fleet

Three depots could not be assessed at all

No vehicle is recorded against a site at three of your depots, so Guardian cannot say whether their capacity is at risk on Thursday or any other day. It reports them as unknown. It does not report them as clear.

To fix: record a depot against each vehicle. The check then runs on those sites without any other change.

An empty result is not a healthy result. Most software reports silence as safety, and silence is where the expensive surprises live.

The one that admits what it does not know. This is the finding that makes the other two worth believing.

Every finding shows its working. An operations manager will not act on “trust me”, and should not.

Where it applies

The reasoning does not know it is about trucks

The three findings above are a distribution operation, because that is the operation connected to it. But none of the three checks is written about distribution. Each is a shape — a rate nobody divided, two correct systems colliding, a check with nothing to run on — and those shapes turn up wherever people run things.

A rate nobody divided

Two systems each hold a true number. Neither divides one by the other, so the number that matters is never produced.

DistributionRunning today

630 units over 90 days is 7 a day. 28 in stock is four days — not a healthy number sitting above a reorder line of 26.

HospitalSame shape

Sterile instrument sets held, against theatre list volume and reprocessing turnaround.

HotelSame shape

Linen par level, against occupancy and laundry cycle time.

Production lineSame shape

A consumable on the line, against takt time and replenishment lead.

Two systems, both correct

Two decisions made weeks apart, by different people, in different systems. Neither is wrong. Together they cost you the day.

DistributionRunning today

Two of six vehicles into the workshop on the Thursday dock use climbed from 88% to 94%.

HospitalSame shape

A theatre list booked against sterile processing capacity that nobody checked.

HotelSame shape

A deep-clean programme scheduled into the week of highest occupancy.

Production lineSame shape

Planned maintenance against a production commitment agreed separately.

The check that could not run

A link between two records was never made, so the check has nothing to run on. Reporting that as healthy is the expensive mistake.

DistributionRunning today

Three depots with no vehicle recorded against a site. Reported unknown, not clear.

HospitalSame shape

An instrument set with no theatre recorded against it.

HotelSame shape

A room with no housekeeping zone.

Production lineSame shape

A consumable with no line recorded against it.

Only the distribution column is running. The rest describe the same question asked of a different operation, and are not deployments. The figures shown are the ones Guardian computes; no figure is shown for an operation that has not been connected.

How Guardian thinks

Five stages, one closed loop

Guardian is not designed as a batch job that runs overnight. It will observe continuously, and every action taken will become training signal for the next recommendation.

Operational Memory
∞

Designed to retain and reuse every pattern

Decision Latency
Real-time

Designed to issue as signal arrives

The heart of Opservor

Everything in. Decisions out.

Guardian sits between the systems that record your operation and the people who run it.

GUARDIAN ENGINE REASONING LAYER DATA SOURCES IOT DEVICES BUSINESS SYSTEMS PEOPLE & PROCESSES INSIGHTS ALERTS RECOMMENDATIONS AUTOMATED ACTIONS
Everything in
Data sources
IoT devices
Business systems
People & processes
GUARDIAN ENGINE Reasoning layer
Decisions out
Insights
Alerts
Recommendations
Automated actions

Signals enter from every system already running. What comes back is not another report — it is the decision, with the evidence attached.

The loop

Ingest, reason, act, learn

01 — INGEST

Data ingestion

Telemetry, transactions and manual entry stream in from every module and every connected external system.

02 — CONTEXT

Contextualisation

Raw signal is joined to the operation that produced it — this site, this shift, this vehicle, this cost centre.

03 — REASON

AI reasoning

Models detect patterns, anomalies and leading indicators across departments that would never be compared by hand.

04 — ADVISE

Recommendations

Actions are ranked by expected impact and issued with a confidence score and the evidence behind them.

05 — ACT

Execution

Recommendations land where the work happens, so the team acts on the highest-impact move rather than the loudest alert.

06 — LEARN

Learning loop

Outcomes will feed back as training signal, sharpening the platform against your operation rather than a generic benchmark.

What Guardian will be asked to do

Guardian Intelligence Packs

Guardian answers one kind of question well today, shown above. These are the kinds it is being built to answer, and the set grows as operations ask for more. They are capabilities of Opservor that a business activates as it needs them — not separate products, and never sold separately.

In design. None of these is built yet. They are listed because the shape of the thing matters more than the order it arrives in.

Growth

The Hummingbird

Opportunity shows up first as a change in pattern, not as a number crossing a line.

Improve

The Honeybee

The same twenty minutes lost every day. Small, repeated, and invisible in a monthly report.

Change

The Chameleon

A change that was agreed is not a change that happened. Adoption fades quietly and nobody reports it.

Resilience

The Sea Turtle

What a site can still cover when two of its six vehicles are gone. Capacity looks fine until the day it does not.

Strategy

The Compass

Three ways forward, and what each one costs. The choice is rarely good against bad.

Risk

The Eagle

Not a warning about today. About the day the consequence lands, while there is still time to move.

Simulate

The Dolphin

Move the service date and see what moves with it — before committing to it.

Root

The Octopus

The symptom was an empty shelf. The cause was three systems each being correct on its own.

Memory

The Elephant

This happened before, and it was dealt with. The person who knew how has left.

Proof

The Lighthouse

A recommendation that is never measured is an opinion. This is what the decision was worth.

Problems it targets

What Guardian is built to solve

The operational problems the intelligence layer is designed to address. Most of them are not hard to understand — they are hard to see, because the evidence for each one sits in a different system. Outcome figures will be published once measured against live deployments.

Fleet optimisation

Inefficient routes waste fuel and hours. Guardian is designed to recommend routing against live traffic, fuel cost and delivery windows.

Warehouse efficiency

Dock congestion slows fulfilment. Guardian will predict peak windows and recommend dock allocation and staffing ahead of them.

Predictive maintenance

Failures cause unplanned downtime. Guardian is designed to read degradation in telemetry and schedule the intervention before the failure.

Inventory balance

Overstock ties up capital, stockouts halt the line. Guardian will forecast demand and move reorder points as conditions change.

Cost optimisation

Inefficiency hides between departments. Guardian is designed to read across cost centres and surface what no single owner can see.

Safety prediction

Incidents get handled after the fact. Guardian is designed to find the patterns that precede them and recommend action while it is still preventive.

Cold chain integrity

A temperature excursion is usually discovered when the load is rejected, not when it happened. Guardian is designed to catch the drift and name the leg it happened on.

Dangerous goods compliance

Placarding, documentation, driver certification and route restriction all have to be current at the same moment. Guardian is designed to check them together rather than one audit at a time.

Workshop and vendor performance

The same repair done twice in a quarter. A vendor whose turnaround has quietly doubled. Both are visible only when the jobs are read together.

Rental fleet cost control

Units stay on hire past the point where buying was cheaper, because the rental sits in one budget and the utilisation in another.

Trailer and equipment utilisation

Trailers are counted as assets and managed as furniture. Guardian is designed to treat them as capacity, and to say which ones have not moved.

Lead time drift

A supplier lead time is recorded once, at onboarding, and then relied on for years. Guardian is designed to measure what the lead time actually is now.

Spare parts obsolescence

Stock held for equipment that has left the fleet. Nothing flags it, because nothing connects the parts list to the disposal list.

Handover loss

The person who knew why a decision was made has moved on. What was tried, and what it cost, leaves the building with them.

Put Guardian to work

See what the intelligence layer finds in your first week.

Schedule a demo →