Week 12 Session: Your Own Track Record

The whole semester's set of ranges, read together for the particular way you tend to be wrong — a shape no single estimate could ever have shown you

24 May 2027

Eleven weeks have asked how good one estimate was, one at a time. Today asks a different question of the same material: read as a whole set, what does it say about the particular kind of wrong you tend to be, if there is one at all?

Open the one that has been sealed#

Start with the entry you cannot revise. Take out your week 10 forecast — the sealed low bound, high bound, confidence, and miss-line for Canberra Airport’s daily maximum on 20 May 2027 — and check it, for the first time, against the Bureau of Meteorology’s Daily Weather Observations for Canberra, ACT, the exact product you named as your resolution source two weeks ago. Record what it shows and whether your range held it, honestly, before you touch anything else in the diary. This is the one entry in your record you were structurally prevented from adjusting in hindsight, which is exactly what makes it worth resolving first, while that discipline is freshest in mind.

Count what you have#

Retrieve every other entry — the one from week 4, the two copied out of week 8’s slips, and the eleven you built on your own initiative — and mark each resolved one hit or miss. No formula does this for you. You are doing by hand a small-scale version of what the Bureau does to its own forecasts when it checks whether a stated 75% happened about 75% of the time: the same verification idea — outcomes counted against the confidence stated in advance — on fifteen heterogeneous entries rather than on thousands of comparable ones. The idea carries. The statistical weight does not.

Lay the count out where you can see the whole thing at once — a table, a grid, whatever makes the misses visible rather than buried in prose. You are not computing anything yet. You are making the shape of your own record legible before you try to say what it means.

Say what the shape is#

From the count, write your pattern claim: about 400 words, naming what your ranges have in common as a set, citing at least four specific dated entries as evidence. A pattern is not “I was often right” — being right does not need explaining. A pattern is something like always missing on the same side, or being most overconfident about quantities you’d read somewhere rather than built yourself, or widening every range that involved another person’s number. Name the actual shape, not a summary of how you felt about doing this all semester.

Go looking for the entry that ruins it#

Before anyone else reads your claim, you read your own diary again, this time hunting for whatever does not fit it. This is not optional and it is not the same activity as writing the claim — confirming a pattern and searching for its exceptions use the record in opposite directions, and doing only the first one produces a claim that only ever meets evidence that agrees with it. Write the counter-case honestly, about 250 words: which entries complicate the story, and whether the pattern survives them or has to be narrowed to survive.

Trade diaries#

Swap with a partner. Read their pattern claim and their entries independently of what they told you their pattern was, then go looking for what would undermine it. That search is the work, and it is not optional.

Write one note back. Best case, it names a pattern in their record their own claim missed, or one specific entry that genuinely does not fit what they argued. If you have searched properly and the claim holds, say so and then do the harder thing: name the limits of what it actually covers, or name the evidence that would most seriously threaten it if it turned up. What is not a note is bare agreement — “this looks right” tells the author nothing they can use.

One change, stated so it can be checked#

Close with about 150 words: one specific thing you will do differently the next time you write a range, and what you would expect to see, later, if it actually worked. Not “be more careful” — a change specific enough that a version of you six months from now could look at your next diary and say plainly whether you kept it.

What today is not testing#

Nobody’s diary is judged on how often it was right. A diary with no misses in it is not, by that fact, a strong diary — nor a dishonest one. It is a diary whose ranges have not been tested, and the thing to ask of it is whether those ranges were informative enough to have been at risk, or wide enough to catch almost anything. What is being read is whether your record was genuinely kept — precommitted, dated, not touched up afterward — and whether your account of it is honest about the entries that complicate your story as much as the ones that support it. That is the same discipline this course opened with, turned, for the first time, on the person who has been doing the estimating all along.