Week 7: Revision, and its imitation

How pressure actually arrives, the test that separates an honest change from a concession, and why the defences have to be built early

19 April 2027

Open the slides

Nobody will ask you to lie. That is the first thing to understand about this week, and most of the reason the problem is hard.

How it actually arrives

The request is never for a different answer. It is for a second look at a particular assumption — and the person asking will have a reason, and the reason will usually be a fair one. Are you sure about that growth rate? Isn’t that contingency conservative? Shouldn’t the comparison include the benefits we always count elsewhere?

Any one of those could be a good question. Asked in isolation, each deserves an honest answer, and sometimes the honest answer is yes.

The pattern only becomes visible in aggregate. Across a dozen such requests, the assumptions selected for a second look are the ones pointing the wrong way, and the assumptions pointing the right way are never mentioned. No individual exchange contains a lie. The estimate drifts anyway, and it drifts in a direction somebody chose.

This is why the defence cannot be scepticism about any particular request. You will lose those arguments, and you should — the requests are often reasonable. The defence has to be a record.

The test

There is a single question that separates a revision from a capitulation, and it is not about how strongly the case was argued.

Did something about the world change, or did something about the room?

New information is a reason to revise. A survey came back, a comparable project completed, an input price moved, someone found a factual error in your working. Revise, gladly, and note what you learned. An estimate that never moves is not rigorous, it is inert.

Who is asking is not a reason. Nor is how senior they are, how much they need the number, how tired you are, or how many times they have raised it. These things feel like reasons in the moment, particularly the last one, because repetition manufactures a sense that the point must have merit.

The test is easy to state and hard to apply live, which is the whole argument for writing your assumptions down before the meeting. Held in memory, an assumption can shift without ever presenting itself as having shifted; you simply find you now believe something slightly different and cannot say when that started. Written down, the same shift becomes an edit — dated, visible, and requiring a reason in the margin.

Widening as a third option

The framing of accept-or-reject is a trap, because it makes every exchange a contest somebody loses.

There is a third move, and it is often the truthful one: leave the estimate where it is and widen the range around it. This concedes something real — that the challenge has genuinely reduced your confidence — without pretending the challenge has told you which way the answer moves. Week 4’s grammar exists partly for this. A point estimate can only be defended or surrendered. A range can be honestly loosened.

Used carelessly it becomes an evasion, a way of agreeing with everybody. Used well it is the most accurate available description of what has just happened to your state of knowledge.

The pressure is structural, not personal

It would be comfortable to treat this as a story about weak analysts and pushy executives. The documentary record does not support that reading.

Infrastructure Australia’s appraisal guidance identifies optimism bias as a significant problem in infrastructure and sets out mitigations for it. A body does not write that into its formal framework because of a few bad actors. It writes it because the incentives point the same way on every project, at every level, in every jurisdiction, and produce the same skew regardless of who is sitting in the chair.

That should be reassuring rather than cynical. A structural problem has structural defences, and the ones that work are unglamorous: assumptions written down and visible, comparison against projects that actually happened, review by somebody with no stake in the outcome. None of these requires courage in the moment, which is exactly their virtue — they work by being in place before the moment arrives.

What this is not

Not a case for stubbornness. An estimator who cannot be moved is no more useful than one who can be moved by anything, and considerably harder to work with. The goal is not an unchanged number; it is a number whose changes each have a reason attached that would survive being read out loud.

Outline