Week 3 Session: Naming Your Assumptions

A case file rather than an exercise — one 1957 estimate, one 1973 outcome, and the question of which assumption was holding the first one up

8 March 2027

You will not estimate anything today. Bring last week’s chain anyway.

The link you circled last week — the one you least wanted to defend — is where today starts. That circle was a hunch. This week we turn it into a question you can actually answer, by practising on an estimate nobody in this room made.

The case#

In 1957 the Sydney Opera House was estimated at £3.5 million. The building was completed in 1973, at a cost of approximately A$102 million.

Resist the arithmetic. Do not divide one by the other.

Those two figures are separated by different currencies — Australia moved to decimal currency in 1966 — by sixteen years of inflation, by a client decision to change the interiors substantially, expanding the number of performance spaces, by Jørn Utzon’s departure in 1966 and a different design for the interiors that followed, and by a great deal else besides. A ratio between them would be a number with no referent: precise, impressive, and about nothing. If week 1 landed, that sentence should already feel familiar.

So this is not a case study in how wrong an estimate can be. Nobody here can apportion that gap, and this session does not try. It is a case study in something narrower and more useful: what was the 1957 figure resting on?

VERIFIED — what the sources say#

From the Sydney Opera House’s own published history of the start of construction:

• Around 700 steel-cased concrete shafts were bored into the site, with mass concrete filling unstable rock. Of this the official account says: “This preparatory work had not been budgeted for, setting the building on a course of extended overruns beyond its estimated cost of 3.5 million pounds.”
• Bennelong Point’s geology “hadn’t been properly surveyed when competition guidelines were written.” The ground turned out to be “made of loose alluvial deposits permeated with seawater and completely unsuitable for bearing the weight of the intended structure.”
• On the roof, at that stage, “the load they would have to bear was unknown.”
• And on the decision to begin: Premier Cahill pushed ahead “even before critical design elements had been settled.”

That is the evidence. Everything else in this session is yours to argue.

Your job: the ledger#

In pairs, build an assumption ledger for the 1957 estimate. What must that figure have taken for granted in order to exist at all? Some will be obvious once written down. Some you will only find by asking what a costing needs to know before it can produce a number.

Then rank them — and rank them by the right thing.

Rank by how far the answer moves if this assumption is wrong, not by how likely you think it is to be wrong.

Those are different orderings and the difference is the point of the week. Likelihood and consequence are two separate axes, and a full account of risk needs both; what this exercise deliberately ranks is the second, because it is the one people leave out. An assumption can be very nearly certain and still be the one carrying the whole answer, so that its small chance of being wrong still dominates what is at stake. An assumption can be a coin flip and barely shift the result, in which case arguing about it is entertainment. Most consequential if wrong is not most likely to be wrong. People audit the second and get surprised by the first.

For each entry, keep two columns apart:

• VERIFIED — what you can point to in the material above.
• INFERRED — what you are reasoning to. Reasonable inference is welcome; passing it off as sourced is not.

Only when your ranking is on paper, compare it with what the official account says actually went wrong early. Where did your ranking agree? Where were you looking at the wrong assumption entirely?

The harder question#

Sort your ledger one more time, into two piles:

Checkable in 1957. Whether the site had been surveyed is a fact about 1957, available to anyone who asked. An estimate resting on an unsurveyed site was resting on something that could have been checked with the evidence of the day, or obtained by commissioning it.

Unresolved in 1957. What a roof of that shape would weigh was not a fact anyone was withholding; on the sources above, the load “was unknown” and critical design elements “had not been settled”. That is a different situation from an unasked question — it depended on design and investigation work that had not yet been done.

The distinction matters because hindsight flattens it. From here, every load-bearing assumption that failed looks like negligence, and that reading is both unfair and useless — unfair to people working on questions nobody had answered yet, useless because it teaches you nothing about your own estimate, whose load-bearing assumptions are invisible to you right now for exactly the same reasons.

Leaving#

One line each, out loud: name an assumption in your own week 2 chain that you now think was load-bearing, and say whether you could have checked it at the time or whether it was genuinely still open.

Nothing is due. Next week we stop asking how wrong a number could be and start asking how to say so.