You cannot improve a hospital that does not exist

Put plainly, and better than the textbooks put it: a new design starts from nothing and goes to the dream. Working on something that exists means doing whatever it takes to turn what you have into what you want it to be.

The destination is the same in both, and in neither case is it a fantasy. A dream is what you want subject to the real world: what the ground will carry, what the grid will supply, what the regulator will licence, what the market will pay and which clinicians actually exist. A dream that has not been ground against those is not a brief, it is an aspiration, and the commonest failure of a new build project is discovering the difference at tender.

So both cases are constrained and they are not constrained by the same things. Working on something that exists means carrying two layers: the real world, and the decisions somebody else took and fixed in place. A new design carries one. The difference between the two methods is precisely that inherited layer, and it is the only difference that matters.

That is why starting from nothing is the freer case and not the easier one. Nothing has to be accommodated and no decision from 1987 has to be worked around, which sounds like an advantage and is mostly an absence of excuses. Everything this study spends twenty pages on, from land title to water table to road speeds to the paying share to the fire brigade response time, is the real world arriving to bound a dream that started out unbounded.

And the constrained case has a ceiling that the free one does not. Doing what it takes has a limit set by what is already there, which is what the rest of this page is about calculating before anybody commissions a programme that assumes there is not one.

The distinction between the two methods is real and it is smaller than the literature makes it, and it is worth collapsing before it does any damage.

Both are the same discipline. Establish the current state honestly. Establish the required state precisely. Close the distance between them and then hold it. Everything else, including which five letters are used, is a question about what kind of intervention closes that particular distance.

Which exposes a mistake that this page made in an earlier version and that design projects make constantly. It is not true that a design project has nothing to measure. The current state always exists. It simply is not inside the facility that has not been built.

For this project the current state is not an empty field outside Lagos. It is a family selling something to buy three plane tickets, a consultant in Ibadan writing a referral to a hospital four thousand miles away, and a billion dollars a year leaving the country. That is a process. It has a cycle time, a cost, a failure rate and a set of customers who can describe it in detail, and every one of those is observable today without anybody laying a brick.

So the Measure phase of this study was not skipped and it was not substituted. It was run, and what it measured was the process people use when the thing they need does not exist. The patient page, the family testimony and the outbound flow figures are not context or colour. They are the AS IS.

That changes what the model is for, which is worth being precise about. It does not stand in for observing the current state, because the current state was observed. It stands in for observing the future state, which is the only one that genuinely cannot be watched before it is built. A run predicts the TO BE. It does not excuse anybody from measuring the AS IS.

And it supplies the test the earlier chapters are really applying. A project that cannot describe its current state in measurable terms has not finished Define, whichever method it thinks it is using. A project that can describe the current state but not the required one is about to improve something toward no particular destination. Both failures are common, both are visible from outside in about ten minutes, and neither has anything to do with which acronym is on the front of the report.

The two sequences look similar enough that the difference is often treated as terminology. Read the last column, because the divergence is real and it is largest exactly where it is least noticed.

The second row is the one that matters. In an improvement project, Measure means observing a process that runs: how long things take, how much they vary, where the waiting is. It is the phase that supplies the evidence for everything after it, and it is evidence in the ordinary sense, because somebody watched.

The second row is where the two are usually said to diverge most, and the divergence is narrower than that. In an improvement project Measure observes the process inside the thing being improved. In a design project it observes the process people are using instead, because there is always one, and then adds the requirement translated into characteristics the new design can be tested against. Both are observation of a current state. Only one of them finds it in the building it is about.

What a design project genuinely cannot observe is the future state, and that is what the model is for. It predicts the TO BE. It is not a substitute for measuring the AS IS, and a project that uses it as one has quietly skipped the only phase that produces evidence rather than intention.

Three numbers decide which method applies, and only two of them are ever produced.

Current performance is measured constantly. The requirement is usually stated, sometimes carelessly. The third is entitlement: the best this design could do with every special cause removed, every process running properly, and nothing changed physically. It is what the building and the configuration will allow at their theoretical best, and almost nobody calculates it, because doing so means writing down what the existing design cannot do.

With it, the question answers itself. The part of the gap that sits below entitlement is improvement work and it is available. The part above it is not available at any effort, because it is a property of the design rather than of the process running inside it.

Look at what that does to the four measures above, and at how badly the size of the gap predicts the answer. The largest gap on the list is half reachable, because the lift core caps how many patients can move before noon. The smallest is entirely reachable. Nurse walking has a large gap of which barely a quarter is available, because the distance between a bed and a store is a constant. Theatre cancellations look intractable and are entirely a policy matter, because ring fencing a list costs nothing physical.

Two of those four are improvement problems and two are design problems, and nobody could tell which by looking at the numbers everybody already has. That is the argument for calculating the third one.

The failure mode worth naming is the one that follows from skipping the entitlement calculation, and it is depressingly orderly.

An operator is set a target above what the building allows. An improvement programme is commissioned, competently, and it delivers: the reachable part of the gap closes and performance improves genuinely. It then stops, because it has arrived at the ceiling. The target is still missed.

What happens next is the damaging part. The shortfall is attributed to execution, since the method is known to work and the team clearly did not achieve the target. A second programme is commissioned. It delivers nothing, because the first one already took everything that was available. The team learns that improvement work is theatre, the organisation learns that its staff cannot deliver, and nobody involved has done anything wrong.

This is the same structure as the appraisal problem on the constants page, arriving from the methodological side. An operator measured against a performance whose ceiling somebody else set will look like a poor operator, and the only thing that distinguishes a low ceiling from a bad team is the number in the middle column of that table.

Which gives the rule in one sentence. Before commissioning improvement work, calculate the entitlement, and if the requirement sits above it then the honest answer is that this is a capital problem wearing an operational costume, and no programme will fix it.

There is one consequence of that framing which deserves stating on its own, because it is the reason a design project carries a heavier duty than an improvement project rather than a lighter one.

The dream stops being a dream on the day it opens. From that morning it is simply what somebody has, and everybody who comes after is in the second case: doing what it takes to turn what they have into what they want. Every freedom exercised during the design is converted, at handover, into a constraint that a stranger will spend decades working around.

So the new design is not choosing a building. It is choosing which of the real world constraints get absorbed now and which get handed on, and therefore the entitlement of every improvement programme that will ever be run inside it. A ward drawn slightly too long does not fail; it sets a ceiling, and somebody in 2049 will commission work against that ceiling, deliver the reachable part, miss the target and be judged to have underperformed.

That is what makes the freedom uncomfortable rather than pleasant. The person who will pay for a decision taken now has no way to object to it, cannot be consulted, and in most cases has not been born into the profession yet. The only thing that represents them at the table is whether anybody writes down what was chosen and why.

It also settles what a design project owes the operator, which is not a finished building. It is a stated entitlement: here is what this configuration can do at its best, here is the evidence for that claim, and here is the list of the things we fixed in place that you will not be able to change. An operator handed that document can tell the difference between a management problem and a capital one on their first day, which is the distinction the rest of this page exists to protect.

The last phase differs in a way that sounds pedantic and is not.

Control holds a gain that has already been demonstrated. The change was piloted, the number moved, and the work is to stop it drifting back: a standard, a measurement, an owner who notices. It is a maintenance activity and it belongs to whoever runs the process.

Verify proves that a design meets the characteristics it was built to satisfy, before anybody relies on it. It is not maintenance, it is evidence, and it happens once, at the point where a thing is handed from the people who made it to the people who will live with it.

In building projects that phase mostly does not happen. There is commissioning, which demonstrates that the systems work, and there is handover, which transfers responsibility. Neither of them asks whether the finished hospital achieves the characteristics stated at the beginning, and the reason is that the question has no owner: the designer has finished and the operator has not arrived.

Which is what the sixth gate on this project is for. It is the point where Verify hands to Control, and it never closes, because the only instrument capable of verifying a hospital is the hospital running. Everything measured after opening is simultaneously the proof that the design worked and the baseline for holding it.

The whole study mapped onto the five phases, which also makes it navigable, since the pages were written in this order without the order being announced.

The last column is the point of the table. Every phase here has something it could not do, and in each case the limitation is a consequence of designing rather than improving: no process to measure, a decisive input that stays unmeasured, a design that is a shape rather than a specification, and a verification that can only see what somebody thought to model.

Naming those is not modesty. A design method that does not state what its evidence is missing produces the same confident output as one that has the evidence, and the two are indistinguishable from outside.

Three, and the first undermines the central argument of the page, which is the right place to put it.

The entitlement test decides everything and entitlement is never observed, because a process almost never runs at its theoretical best. So it is estimated, and it is estimated by people who usually have a view about whether the answer should be a redesign. The defence is to derive it from physical constants rather than from performance history: lift capacity, walking distance, plate depth and compartment size are measurable, they do not have opinions, and they bound the entitlement whether or not anybody likes where the bound falls.