An assumption arrives labelled as data.

What the study was actually for

The total cost management framework describes feasibility analysis as a step that rates and ranks alternative ideas against criteria the team establishes before it starts rating anything. Cost and value engineers are among those who lead it, and it applies scheduling, estimating and risk work to a scope defined only at conceptual level.

Read that as a job description and one thing stands out: the output is a ranking. The study exists to say this option rather than those options, and everything in it is shaped by that purpose.

An assumption that would be dangerous in a contract can be perfectly sound in a comparison, as long as it is applied evenly to every option. If you assume the same ground for all four sites, the ground assumption doesn't change which site wins. It has done its job. It is also now sitting in a document that will outlive the comparison it was made for.

Two kinds of sentence, and only one of them was tested

Open any feasibility report and the sentences fall into two classes.

Some describe something somebody went and found: a borehole log, a survey, a metered demand, a quoted tariff. Others describe something the team adopted so the work could continue: a rock head level between two known points, a haul route nobody has driven, a connection date the utility has yet to confirm, a labour rate from a neighbouring region.

The study tests which of the second class matters, by moving each one and seeing what happens to the answer. That test is the most valuable thing in the document and it is the first thing lost, because its output is a statement about uncertainty and the tender pack has no field for one.

Where the label comes off

The path is short and nobody on it does anything wrong.

The study is written with its hedges intact. The design consultant appointed afterwards works from its conclusions, because the conclusions are what was approved. Their drawings carry a level, a route, a rate — not a range, because a drawing can't carry a range. The tender documents then assemble what is available about the site and issue it as information, which under the standard forms is exactly what the Employer is required to do.

At no point does anybody delete a caveat. The caveat simply has nowhere to go at each step, and after three steps it is gone.

This matters for how the problem gets investigated later. When a condition turns out different and somebody goes looking for the origin of the number, the search assumes there was a decision to find. There was none. There were three transfers between document types, each of which preserved everything the receiving format was capable of holding.

Which is also why the defence is a format rather than a person. Nobody needs to be more careful. Something downstream needs a field the caveat can survive in.

What the contract deems, and what it never asks

The Employer makes site data available. The Contractor is then made responsible for interpreting all of it, is deemed to have inspected the site and satisfied itself before tendering, and the accepted price is deemed to have been based on that interpretation. That is the architecture, and it is the same architecture in all three books with the balance shifted.

What the contract doesn't ask, anywhere, is which of the data was measured.

That silence is deliberate and defensible. A contract allocates a risk; it doesn't audit the provenance of every figure it hands over, and a form that tried to would be unusable. But the consequence is that the class of a statement stops being recorded at precisely the point where money attaches to it, and from that point onwards it can only be reconstructed by argument.

Foreseeability is judged against what an experienced contractor could reasonably have worked out as at the Base Date — and the Base Date, from week 2, sits 28 days before tenders are due. So the test is applied to a moment when the only thing in existence was the study and whatever the tender pack had made of it. An assumption made three years earlier, for a comparison that is finished and forgotten, is the material the reasonable contractor is deemed to have reasoned from.

Putting the label back on

There is an instrument for this and it is well established in tunnelling and heavy civils. A geotechnical baseline report states, as a contractual reference, what the ground is to be taken to be. Not what it is — what it is deemed to be for pricing. Encounter something worse and the departure is measurable against a stated line rather than argued against a folder.

The international contract law literature records a case where the whole baseline came down to ten key parameters in a simple table, and singles out that clarity as what made it work. Ten rows. The instrument isn't expensive; it is unusual.

And its real function is the one this week is about. A baseline is a place to write down the assumption, at the only moment when writing it down still costs nothing.

System design

The second row is the one the current project management standard names outright and the industry mostly doesn't keep. An assumption register is a list of sentences that aren't yet true, each with the thing that would settle it — and it is the only record in this table whose value goes up the longer the project runs.

Record	Produced by	Required quality	Verified against	Feeds
Feasibility report	The study team, before appointment	Every input marked measured or assumed, and the mark carried forward	The investigations behind it	The case · the reference design
Assumption register	The same team, as it works	One line per assumption: who adopted it, when, and what would disprove it	The document the figure came from	Sensitivity · the baseline · your queries
Sensitivity result	Feasibility analysis	Names which assumptions the answer actually turns on	The ranking it was testing	Where investigation money should go
Site investigation report	The Employer, before tender	Extent and date stated, so gaps are visible as gaps	The physical ground	Site data · the tender price
Baseline of record	The Employer, in the tender documents	States the parameter that governs, in a form a departure can be measured against	The investigation it rests on	Pricing · every later argument about conditions

Read the Required quality column on rows one and two together. Both ask for the same discipline: mark the class of the statement, and carry the mark. Neither costs anything at the time, and after award neither can be reconstructed.

Practical insight

Ask for the feasibility study, and ask for it differently from how you asked for the business case last week. The business case is commercially sensitive and you may be refused. A ground investigation report or a route study sits in a different category, and the request is ordinary.

What you are looking for isn't the conclusions. It is the section where the study says what it assumed, and the sensitivity work if there is any. Those two pages tell you which numbers in your tender pack have something behind them, and which are somebody's reasonable adoption from years before you were appointed.

Then do the cheap thing. Take the five or six inputs that most affect your programme — ground, access, a utility connection, a permit, a delivery route — and for each one write one line saying where the figure came from and whether anybody measured it. Half will trace to a document. The rest will trace to nothing you can find, and those are the ones to raise while raising them is still a clarification rather than a claim. That list is your week-four deliverable and nobody else on your job will produce it.

You won't get this done in month nine. The window is the first six weeks, when asking where a number came from still reads as somebody getting up to speed.

Key takeaways

• A feasibility study distinguishes what it measured from what it assumed, because it needs that distinction to test its own answer.
• The study exists to rank options against criteria set beforehand, so an assumption applied evenly to every option is sound for that purpose.
• The same assumption isn't sound as a contract input, and nothing in the process marks the change of purpose.
• Sensitivity work is the most valuable thing in the study and the first thing lost, because its output is a statement about uncertainty.
• The label comes off in three ordinary steps — conclusions approved, drawings drawn, data issued — and nobody deletes a caveat at any of them.
• The Employer makes site data available; the Contractor is made responsible for interpreting it and the price is deemed to rest on that interpretation.
• The contract never asks which of the data was measured, so the class of a statement stops being recorded at exactly the point money attaches to it.
• Foreseeability is judged as at the Base Date, from material an earlier study produced for a different purpose.
• A baseline states what the ground is deemed to be for pricing, which makes a departure measurable instead of arguable.
• Writing an assumption down is free while it is still an assumption and expensive afterwards.

Records born here. The feasibility report and its assumption register · the alternatives compared and why one won · the sensitivity result · the site investigation report · the basis of estimate · the baseline stating which parameter governs.

What is coming next

The study didn't simply end. It was presented at a gate, and somebody with authority decided it was complete enough to spend the next tranche against.

Those gates come with a price attached — each one is expected to produce an estimate of a stated accuracy, and the accuracy tightens as the gates pass. The number you will be handed at award has a class, that class was fixed at a meeting held before your company was on anybody's list, and it is the reason a variance can appear in month three that nobody on the job caused.