Two kinds of not knowing. Only one of them shrinks.

One question separates them

Forget probability for a moment. Ask this instead.

If I spent money finding out, would the uncertainty get smaller?

For the rock, obviously yes. Two boreholes in the northern half and the question is largely settled. The answer would cost about $6,800 and take a fortnight.

For the gang, no. You could study that gang for a month, measure every hour they have ever worked, interview the foreman and audit the last four jobs, and you still would not know what Tuesday produces. Because Tuesday has not happened. Whether it rains, whether the banksman is off, whether the delivery lands at seven or eleven — none of that has been decided yet by anybody.

Those two answers describe two genuinely different kinds of uncertainty, and they have names.

Why one shrinks and the other doesn't

Epistemic uncertainty is uncertainty about something that already has an answer. The rock is either there or it is not. The ground made that decision in the Jurassic and has not changed its mind since. Our uncertainty is not a property of the site — it is a property of our ignorance, and ignorance is something you can spend money on.

Aleatory uncertainty is variability in a process that has not run yet. There is no fact of the matter about Tuesday's output because Tuesday does not exist. You cannot buy the answer, at any price, because nobody has it. What you can do is measure the range — how much daily output has varied historically — and hold money against the range.

So the two demand different responses, and they are not interchangeable.

Epistemic uncertainty wants information. Aleatory uncertainty wants money.

Put contingency against an epistemic risk and you have paid to stay ignorant. Try to investigate an aleatory one and you will spend a month producing a number that was already in the timesheets.

Nine of the fourteen

Go through the register with that one question and the split is not close.

The rock. The groundwater. The wayleave. The supplier's solvency. The switchgear lead time. The discharge consent. The waterproofing interface. The temporary water. The temporary power.

Nine of the fourteen are epistemic. Every one of them has a definite answer that exists right now, held by somebody, and not currently written down anywhere on this project.

Four are aleatory: steel escalation, the weather, the crane, the pump. Nobody knows those, including the people who would sell you the answer.

One — late detailing — is a genuine mixture, and it is worth saying so rather than forcing it. How many drawings are outstanding today is a countable fact. Whether the designer delivers the next one on time is not.

Now put money against the split. The nine epistemic risks carry $63,880 of expected value between them. The cost of resolving all nine — two boreholes, a winter standpipe reading, a pre-application enquiry, a credit check and five emails — is about $9,300.

Five of the nine are free. They are emails.

What information actually buys

Now the honest bit, because there is a version of this argument that is too good to be true and you should not repeat it in a meeting.

Spending $6,800 on two boreholes does not save $24,000. It does not move a single stone. If there is rock under the north, drilling a hole to look at it does not make the rock go away, and you will spend the money anyway.

What the investigation buys is different, and it is worth more than it sounds.

It tells you which world you are in. If the north is clean, you release the contingency and use it on something that is still uncertain. If the north is rock, you now know before you mobilise — so you bring the right rig once instead of standing the wrong one down, you programme it properly instead of absorbing it, and you put the client on notice at a point when the contract still gives you somewhere to go with it.

The rock costs the same either way. Finding out early changes whether it costs you the money, or costs you the money plus a standing rig, a broken sequence and a claim you filed too late.

And for several of the nine, the uncertainty really does go to zero. The wayleave either exists or it doesn't. The client either is or isn't providing the temporary power. Those are not probabilities. They are unanswered emails.

Why it matters across a company, not just a job

There is a second reason to keep these two apart, and it only becomes visible above project level.

Aleatory variability cancels. Run twenty jobs and some have a wet winter and some don't, some get the good gang and some don't. Across the portfolio the good and bad roughly offset, which is exactly why a company can carry less contingency in total than the sum of what each project would want on its own.

Epistemic error does not cancel. If your company consistently assumes the ground is better than it is, or that designers deliver on the date they said, that assumption is wrong on all twenty jobs, in the same direction, every year. There is nothing for it to offset against.

That is how a business full of capable people, each managing their own project competently, loses the same money annually and never finds the cause. It isn't in any one project. It is in an assumption that every project inherited.

Practical insight

Take your register and put one letter next to every line. E if somebody, somewhere, already knows the answer. A if the answer does not exist yet.

Then for every E, write two things: who has the answer, and what it would cost to get it. Not a plan — a name and a number.

You will find, as we did, that a third of them are emails nobody has sent. Send those this week. They cost nothing and they will take real lines off your register, which is the only way a register ever gets shorter.

For the rest, compare the cost of finding out against the money you are holding. If the investigation is cheaper than the contingency, you are currently paying a premium to remain uninformed — and that is a decision somebody should make on purpose rather than by default.

Key takeaways

✔ Epistemic uncertainty is about a fact that already exists. Aleatory uncertainty is variability in something that has not happened yet.
✔ The test is one question: if I spent money finding out, would it get smaller?
✔ Epistemic uncertainty wants information. Aleatory uncertainty wants money. They do not substitute for each other.
✔ Nine of the fourteen risks are epistemic, carrying $63,880 — and about $9,300 would resolve all nine.
✔ Five of those nine cost nothing. They are unsent emails sitting in front of real money.
✔ Investigation does not move the rock. It tells you which world you are in, early enough to bring the right rig and give notice in time.
✔ Aleatory variability cancels across a portfolio. A bad assumption is wrong on every job, in the same direction, every year.

What's coming next

Aleatory uncertainty wants money. But how much money, exactly?

Up to now every risk on this register has carried a single impact figure — $60,000 for the rock, $30,000 for the steel — as though the consequence had one value. It doesn't. If the rock turns up it might affect six piles or twenty-six, and those are wildly different numbers hiding behind one.

Next week we stop writing a number and start writing a range. Which sounds simple until you discover that the same three figures — best case, likely, worst case — produce different answers depending on which curve you stretch between them, and that most people pick the curve without knowing they have picked anything at all.