The Discipline
Every specialist holds a part of a system. Systems ontology takes the whole as its object.
Ask what a hospital is, and the answer is a list: wards, a pharmacy, a laboratory, an emergency department, the people who staff them. Ask how it is doing, and the answer is the same list with numbers beside each item. Every ward has a head. Every number has an owner. Nothing on the list is the hospital.
This is how systems are usually seen: as their parts. The parts are what can be pointed at, staffed, measured, and held to account. But the parts are not where a system begins. It begins with a purpose. The purpose decides what the whole must be, the whole is divided into parts, and the parts are joined so that together they do what none could do alone. The parts come last. They are the last thing the system depends on, and the first thing anyone sees.
Systems ontology reads a system in the order it holds together rather than the order it is seen: first what it is for, then the whole, then each part in its place and the relations between them, and then how the system meets the world it was built for.
Purpose comes first
Every system begins with a purpose: something it exists to do. What a system is for is the first thing to know about it, and often the last thing anyone wrote down. A system can do exactly what it was built to do and still be failing, if what it was built to do is not what it is for. So a reading begins with the purpose. Without it there is nothing to read the system against.
The purpose is read, not received. It is usually already written somewhere: in a statute, a charter, a court order, a company’s governing documents, the regulation it works under. A system inside an organization rarely has a purpose written for it alone. Its purpose is the share of the larger purpose it exists to carry out, and finding it means following the purpose down from where it is written to the system in question. What the sponsor says the system is for is where the reading begins. It is one account among several, and the system is read against what its sources establish.
Every system also has the purpose it actually serves. Usually the two match. Where they do not, the distance between them is the first thing a reading finds.
The purpose is rarely missing outright. More often, something easier to write down has taken its place.
The county jails of San Diego are a plain case. Deputies were required to check every cell at least once an hour and to record that they had, and they did. From 2006 through 2020, 185 people died in those jails, one of the highest totals in California. The policy required the checks to be done and documented, but did not require staff to determine whether the person in the cell was alive. The procedure was followed, and every log showed the work was done, while the purpose the check existed for—making sure the person in the cell was alive and well—went unserved.
That is the common shape of a failed purpose. A procedure, a metric, a project’s goals, a promise: each is easier to count than the purpose, and each can be met in full while the purpose goes unserved.
At the far end is a system that has never been treated as one. The levees along the lower Pajaro River in California were finished by the federal government in 1949 and handed to two counties to maintain. The river is the line between those counties. Each agency had a written purpose for its own part. The whole had no purpose and no one with authority over it. The levees first failed six years after they were finished, and again several times after that. An agency responsible for the river as a whole was formed only in 2021, and in March 2023 a breach flooded the town of Pajaro. When the parts of a system have purposes and the whole does not, a reading may be the first time it is seen as a system at all.
What a system is
Beneath the purpose is the whole: the arrangement of parts and relations the purpose calls for. The whole has properties none of its parts have. Flour, water, yeast, salt: none of them tastes like bread. Bread exists only when the four are in the right proportions, given the right time and heat. A mechanical clock is made of gears, springs, and an escapement, but no gear tells the time. Telling the time belongs to the assembly, present only when the parts stand in the right relations, and it is the thing the clock exists for. Often the property that matters most belongs to the arrangement and not to any ingredient.
Between the parts are the relations, and the relations have properties of their own. A plug and a socket can each be perfectly made and still not fit. The fit belongs to neither of them. It exists only between them, which is why a system built entirely from sound parts can fail, and why no examination of the parts alone will show it. Relations are the least recorded thing in most systems, and a relation no one has written down is still doing its work.
A whole is divided, and every division is a cut. Some cuts run inside the system: departments, units, roles, modules. One runs around it and separates the system from everything else. Both cuts are judgments, and both can be wrong. A company that organizes customer service by channel sends calls to one team, email to another, and the stores to a third. A customer who calls, then writes, then walks in with the same unsolved problem becomes three tickets on three boards. Each team closes its ticket. No one has been given the customer. Cut by customer instead of by channel, and the problem sits in one place, where someone owns it.
The cut around the system matters as much. A passenger railway can run every train it controls on time and still arrive late, if it shares track with a freight company that decides the order trains pass through a junction. The constraint lies outside the line. Every improvement made inside it is real, and not enough.
And a built system acts in the world, which answers back. Every built system carries a model of the world it was made for: the loads a bridge was designed to bear, the demand a schedule was built around, the kind of case a process expects. It also needs a way for the world to correct that model when it is wrong: a complaint that reaches someone able to act, an inspection, a sensor, a count. Where that correction works, the system stays adequate as the world changes. Where it does not, the system goes on running on its model with the same confidence whether the model is right or not.
Where failure shows, and where it lives
Put these together and a pattern follows. A system touches the world through its parts, so that is where its failures show. A wrong purpose, a missing function, a misdrawn line, or an outdated model rarely appears as itself. It appears as a shortfall at a counter, a mistake by an employee, a part that broke. When an institution has no way to read the whole, it mistakes the place where the symptom appears for the cause of the problem. Because it can only see parts, a bad part is the diagnosis it can reach.
Imagine a chain of shops where the till keeps coming up short, at branch after branch. Each shortfall is handled as its own case, and each case has the same explanation: the person running the branch took the money. Every case is handled carefully. Hundreds of careful judgments, each reasonable on its own, and each wrong, because the cause was in the system all the branches shared, and that was the one thing no one was looking at.
That is close to what happened at the UK Post Office. For sixteen years it treated shortfalls in its branches’ accounts as theft, and about a thousand people were convicted. What it had no way to represent was a cause shared by every branch: the system itself. The figure the system produced could be wrong, and nothing in the system could find out when it was.
The reverse happens too. At Wells Fargo, branch employees opened about one and a half million deposit accounts that customers may never have asked for, to meet sales goals the bank had set. Here the people at the counter did exactly what they were accused of. Between 2011 and 2016 the bank fired some 5,300 of them, while the goals stayed in place. The misconduct was real, but so was its source.
Set a goal that can only be met by cutting corners, and some people will cut them. Fire them, and the people hired in their place meet the same goal the same way. The firing looks like a remedy, but it is really a diagnosis. It says the fault was in the people, and it leaves in place the arrangement that produced what they did.
None of this requires carelessness. A way of seeing built to measure parts is strongest at the level that comes last, and weakest at the level everything follows from. It sees every teller and no goal, every branch and no network. The more carefully it examines the parts, the more certain it becomes that the fault is in one of them.
Why the usual checks miss it
Organizations are not short of checks. An audit asks whether the rules were followed. A management review or a program evaluation asks whether the targets were met. A risk function asks what could go wrong within the model the organization already uses. Each takes the system’s picture of itself as given, and each is right to, because that is its job. None of them asks whether that picture is true to the system. Each works within the system’s own categories. Systems ontology works on them.
The disciplines that study systems divide the same way. Systems engineering builds a system to a specification. Control theory studies how a system holds its course. System dynamics studies how a system behaves over time. Each takes one aspect of systems and goes deep, and on its own ground each knows more than this discipline would. Systems ontology takes what a system is as its object. It does not sit beside these fields as another specialty. It sits beneath them as the floor.
The specialists lose nothing by this. The engineer still builds, the auditor still audits, the analyst still models. What changes is that the whole each of them stands inside, and none of them is assigned, becomes someone’s work.
Seven natures
When a reading finds a fault, it names its nature: what kind of thing is wrong. The nature matters as much as where the fault sits, because the nature decides the remedy. A fault whose nature is misread can be patched for years and never fixed.
There are seven natures, and they follow from what a system is. One concerns the purpose that comes before the system. Four concern how the system is made: its whole, its cuts, its parts, and its relations. And two concern how it meets the world: what it assumes, and how it is corrected.
Before the system
Misaligned purpose
The system has no operative purpose, serves purposes that conflict with one another, or the purpose it actually serves has diverged from the one it states.
RemedyA governance decision.
How it is made
Flawed arrangement
The parts and relations are not arranged as the purpose requires — whether because a necessary function was never built, part of the system is no one's responsibility, or local decisions have added up to an arrangement no one designed.
RemedyRe-derive the arrangement from the purpose.
Misdrawn division
A cut drawn where the system does not naturally divide. Whether at the system's outer edge or between its internal parts, it separates what should be together or joins what should be separate.
RemedyRedraw the cut, or reassign ownership across it.
Defective component
A part does not do what its role in the whole requires, or is absent altogether.
RemedyThe specialists: resource it, replace it, or build the capability.
Broken relation
Two parts affect each other, and the relation between them is either mis-specified, unrecorded, or no longer working.
RemedySpecify or establish the interface, or remove one that should not exist.
How it meets the world
False assumption
The system runs on a model of the world that is false — either because conditions have changed, or because the assumption was never true to begin with.
RemedyVerify or re-model.
Missing feedback loop
A region of the system is never tested against the world, or is tested and does not change it.
RemedyBuild or repair the loop.
Where a fault is, and what it is
Every finding carries two coordinates. Its location says where in the system it sits: at a part, in a relation, in the whole, or at the boundary. Its nature says what kind of fault it is. The location tells you where to go. The nature tells you what kind of fix to bring.
A finding is written so that it can be checked. It says what the system is for, what that purpose requires, where the system departs from it, and what kind of fault that is. Producing a finding takes the competence this page describes. Checking one does not. Every step can be tested against the system itself by the people who run it. A navigator can know a chart is wrong from the pattern of groundings without having the right chart. A finding works the same way: it shows where to look.
And sometimes what a reading finds is that the system is sound. That is a finding too.
Why the work runs as it does
The way an engagement runs follows from all of this. It begins by settling the purpose and the boundary, because nothing can be judged before them. It needs a sponsor whose authority covers the whole, because a reading commissioned from inside one part will stop at that part’s edge. Where no one holds the whole, authority has to be assembled before the whole can be read. It needs access to the system in operation, because without it the reading rests on the system’s account of itself, which is the thing being read. And its findings are delivered unedited, because a finding filtered on its way back is a loop that changes nothing.
These are the engagement conditions. They are what a true account of a system requires.
The name
Ontology is the study of what is. Systems ontology turns that study on the things people build, and asks what each one actually is, apart from how it is described, divided, or measured.
In any organization, what exists on paper—the org charts, the policy manuals, the compliance checklists, the stated goals—forms a thick layer of representations. It is easy to govern the representations and believe you are governing the system.
The discipline is to look past what a system is said to be, to read what is actually there.
It is set out in full in its founding text, Systems Ontology: The Competence, Its Ground, and Its Practice.
To see the discipline applied to a complex failure on the public record, read the published reading of the UK Post Office Horizon system.
Begin with the situation
Tell us what you own and what has happened. We will say which engagement fits, or that none does.