You sit an exam on what you covered, and then answer the harder question the exam cannot. Did your mechanism cause that result, and what would you cut from it now
Set on the quarter you cut in week 10, from the material you nominated, and
written to test whether you can use the ideas rather than whether you saw them.
It is not comparable between students and it is not meant to be. Yours is harder
or easier than the person next to you depending on what you picked, which is why
it is not the score being marked — see the
policies page for how that is handled and why we decided it that
way rather than forcing everyone onto one subject.
What it does establish, and nothing else can, is whether learning actually
happened. Every other artefact in this course is something you wrote about
yourself.
This is the hard half, and the honest answers are more interesting than the
flattering ones.
Some of you will have a good result and be unable to attribute it. The wall
turned out to be lower than you remembered; you had a light fortnight elsewhere;
the mechanism ran for four days and then you just worked. Say so. A frank
account of a confounded result is worth more here than a confident story, and
confident stories are easy to spot because they never mention the weeks that
didn’t fit.
The interesting failure is the opposite one, and some of you will have it: the
mechanism ran faithfully for three weeks and you stalled in the same place you
always stall. That is a result, it is worth full marks in this criterion if you
can show it, and it is more useful to the person reading it than another
success.
The evidence you have is your logs: the verdict stream, the attempt lines, the
boss tree with its dates. Argue from them. Show us the stretch where the
mechanism was running and the stretch where it wasn’t, and what the difference
looks like.
The last question, and the one that separates people who built a mechanism from
people who built a hobby.
Adding to a system is easy and feels like progress. Removing something is a
judgement: you have to decide that a part which cost you effort is not earning
it. Every mechanism worth keeping has a shorter version, and the person who can
name their shorter version understands their own design.
Come back to where we started. The failure loop was the only variable that
mattered. Of everything you built across twelve weeks, how much of it was
actually shortening the loop, and how much was you enjoying the machinery?
The slides, as text
Every slide in order, with a description of each figure. Generated from the deck source, so it cannot fall behind it.
1
The exam, and what it cannot tell you
SLOP3733 · Week 12
FigureA stone wall lit from the left, with one patch scuffed pale by repeated impacts at the same height, and four chalk tally marks beside it counting the times it was attempted.
2
The exam
Set on the quarter you cut in week 10, from the material you nominated.
Written to test whether you can use the ideas, not whether you saw them.
3
It is not comparable between students, and it is not meant to be
FigureFour students, four different subjects, four different exams. On the left, their raw scores as bars, struck through and labelled not comparable, four different papers. On the right, the distance each one travelled from the baseline they wrote in week 10, drawn as an arrow from baseline to result. The student with the lowest raw score, mathematical statistics, has by far the longest arrow. The caption reads: the criterion is the exam, not the exam score. You cannot improve your mark by picking an easy subject and you cannot lose marks by picking a brutal one.
4
So the criterion is *the exam*, not *the exam score*
Yours is harder or easier than the person next to you depending on what you picked.
You cannot improve your mark by picking an easy subject, and you cannot lose marks by picking a brutal one.
You can lose marks by picking something you have not actually stalled on, because the evidence of the stall is marked too.
5
What it establishes, and nothing else can
Whether learning actually happened.
Every other artefact in this course is something you wrote about yourself.
6
Did the mechanism cause it?
FigureA broken ring of tick marks closing on itself, with a short bright arc where the loop restarts
7
Some of you will do well and be unable to attribute it
The wall turned out to be lower than you remembered. You had a light fortnight elsewhere. The mechanism ran for four days and then you just worked.
Say so.
A frank account of a confounded result is worth more than a confident story, and confident stories are easy to spot: they never mention the weeks that didn't fit.
8
The interesting failure is the opposite one
The mechanism ran faithfully for three weeks and you stalled in the same place you always stall.
That is a result. Shown properly it earns this criterion in full, and it is more useful to the person reading it than another success.
9
Argue from the logs
The verdict stream. The attempt lines. The boss tree with its dates.
Show us the stretch where the mechanism was running and the stretch where it wasn't, and what the difference looks like.
By now your memory of week 10 has been replaced by a story about week 10. The logs are the only part of this that did not get rewritten.
10
What would you cut?
FigureThree feature panels drawn as faint outlines and struck through: a decay meter, a win-streak multiplier, and an end-of-session summary card. Each was built and then taken out again.
11
Removing something is a judgement
Adding to a system is easy and feels like progress.
To remove a part you have to decide that something which cost you effort is not earning it.
Every mechanism worth keeping has a shorter version, and the person who can name their shorter version understands their own design.
12
Where we started, twelve weeks ago
FigureThree feedback loops on a logarithmic time axis from one second to one month. A Souls boss: ninety seconds to failure, fifteen seconds back. A combo in the lab: three seconds to failure, but about two days before a real match tells you whether it mattered. An assignment: three weeks to failure, and the retry line runs off the axis because there is none.
13
How much of what you built was shortening the loop?
And how much of it was you enjoying the machinery?