Week 8: Treating a subject as a set of bosses, on paper

The paper version of the Souls loop. You attack, you take the one lesson, you go again. The list of what you are fighting grows out of those attempts, instead of being written before you have made any

26 April 2027

Open the deck

Last week you died to a boss twenty times and came out understanding a fight you could not have described beforehand. This week you do the same thing to a subject.

You cannot plan what you have not attacked

The instinct with a hard subject is to sit down and break it into parts first. It feels responsible. It is also the move that has failed you before, and here is why: on day one you do not know enough about the subject to know what its parts are. Any list you write is a guess assembled out of the table of contents, and a confident wrong plan is worse than no plan, because you will follow it.

Nobody walks up to a Souls boss and decomposes it. You go through the fog, you die, and the fight tells you what it is made of.

The loop, on paper

Four lines and nothing else:

  1. What I am attacking now. One thing, and small enough to attempt today. Not “understand regression” — derive the normal equations from a blank page.
  2. The attempt. Short. Short enough that you find out today, not next week.
  3. The one lesson. One sentence, and it has to be about the work, not about you. I can follow the derivation and cannot start it is a lesson. I was lazy is not — there is no next attempt that confirms or denies it.
  4. The one change for next time. Exactly one. If you write three, you will not know which of them worked.

Then go again. That is the whole mechanism, and it is deliberately too small to be impressive.

The list grows; you do not write it

You will name things as you go. By attempt five it becomes clear that the reason you cannot start from a blank page is that you never learned to write down what you are minimising — and now you have a second thing to attack, one you could not have put on a list on Monday.

That is the shape it should have. A boss is something you found by attacking, not something you predicted. If your list is complete and tidy after an hour at your desk, you have written a table of contents and given it a new name.

Some of what you find is small: a piece of notation, a tool you cannot drive, a definition you keep mis-remembering. Call those elite mobs, give them one line each and no ceremony, and do not let them become projects.

The border between the two is genuinely blurred, and it is worth knowing that the games have the same problem. Sekiro has 13 bosses and 29 mini-bosses, and several of those mini-bosses are mandatory and harder than the area boss they stand in front of. Dark Souls III has 19, or 25 counting its DLC, and the wikis argue about which non-respawning enemies belong on the list. Elden Ring needs 13 beaten to finish, has somewhere above 40 unique ones on a strict count, and around 238 boss encounters if you count every arena — a number inflated by the same enemy appearing in thirty catacombs.

Nobody agrees on any of these figures, and that is the useful part: if the designers cannot cleanly enumerate their own bosses, you are not going to enumerate yours in advance either. The labels are a convenience for deciding how much ceremony something gets. They are not a taxonomy, and an hour spent sorting your list into bosses and elite mobs is an hour not spent attacking anything.

Kill conditions, written when you can

For each thing you are fighting, write the check that ends it — one sentence, checkable by someone who is not you.

You often cannot write this on the first attempt, and you should not pretend to. “Done” for a fight you have never had is a guess like everything else. Two or three attempts in, you will know what winning looks like, and then you write it down. What you must not do is leave it unwritten indefinitely: a fight with no kill condition stays open forever, which is the single most common way this mechanism fails.

Why any of this hurts less

A hard subject hurts because failure feels total and unbounded. You do not know how much is left, so every failure is evidence about the whole mountain and about you, and that is unbearable over months.

An attempt makes failure small, expected and local. You did not fail at statistics. You failed at starting from a blank page, once, and you already know what you are changing. That is a survivable size, and surviving it is the entire reason you get to attempt number twenty.

This week

Attack something. Five attempts minimum on a real subject, logged four lines each. Bring the log to week 9, along with whatever list of fights has grown out of it — and if the list changed shape halfway through, bring both versions.

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. Treating a subject as a set of bosses, on paper

    SLOP3733 · Week 8

    FigureA fog gate: a stone archway with gold mist standing in it

  2. Last week you understood a fight you could not have described beforehand

    Twenty deaths, one line each, and at the end of it a model of a boss.

    This week you do the same thing to a subject.

  3. The instinct is to break it into parts first

    It feels responsible. It is also the move that has failed you before.

    On day one you do not know enough about the subject to know what its parts are. Any list you write is a guess assembled out of the table of contents.

    And a confident wrong plan is worse than no plan, because you will follow it.

  4. Nobody walks up to a Souls boss and decomposes it

    You go through the fog, you die, and the fight tells you what it is made of

  5. The loop, on paper. Four lines.

    1. What I am attacking now. One thing, small enough to attempt today. Not understand regression — derive the normal equations from a blank page. 2. The attempt. Short enough that you find out today, not next week. 3. The one lesson. About the work, not about you. I can follow the derivation and cannot start it is a lesson. I was lazy is not. 4. The one change for next time. Exactly one. Write three and you will not know which of them worked.

    Then go again. It is deliberately too small to be impressive.

  6. The list you would have written, and the list you got

    FigureTwo lists of the same subject, side by side. On the left, written on Monday off the table of contents: six chapter headings, all the same size and all dim. On the right, after twelve attempts: derive the normal equations from a blank page; write down what I am minimising without being prompted, marked as found on attempt five; read sigma notation without translating it first, an elite mob, also found by attacking; tell an estimator from an estimate; set up a design matrix for a real dataset, which split out of the first fight; and chapter three worked examples, struck through because it turned out not to be a fight at all. Two of the right-hand entries are marked as having no equivalent on the left.

  7. The list grows; you do not write it

    By attempt five it becomes clear that the reason you cannot start from a blank page is that you never learned to write down what you are minimising.

    Now you have a second thing to attack — one you could not have put on a list on Monday.

    A boss is something you found by attacking. If your list is complete and tidy after an hour at your desk, you have written a table of contents and given it a new name.

  8. Some of what you find is small

    A piece of notation. A tool you cannot drive. A definition you keep mis-remembering.

    Call those elite mobs, give them one line each and no ceremony, and do not let them become projects.

  9. The games cannot agree either

    FigureHow many bosses three games have, on a logarithmic axis from ten to three hundred. Sekiro: thirteen bosses, or forty-two counting the twenty-nine mini-bosses. Dark Souls III: nineteen in the base game, or twenty-five counting the downloadable content. Elden Ring: thirteen required to finish, more than forty unique on a strict count, and about two hundred and thirty-eight encounters if every arena is counted. Each game is drawn as a bar spanning its own disagreement. The caption reads: if the designers cannot cleanly enumerate their own bosses, you are not going to enumerate a subject in advance either.

  10. Kill conditions, written when you can

    For each thing you are fighting, write the check that ends it: one sentence, checkable by someone who is not you.

    You often cannot write this on the first attempt, and you should not pretend to. Done for a fight you have never had is a guess like everything else.

    What you must not do is leave it unwritten indefinitely. A fight with no kill condition stays open forever, which is the single most common way this mechanism fails.

  11. Why any of this hurts less

    FigureForty faint horizontal lines and one bright gold line among them: everything a long session produced, against the one thing short enough to keep

  12. Failure becomes small, expected and local

    A hard subject hurts because failure feels total and unbounded. You do not know how much is left, so every failure is evidence about the whole mountain, and about you. That is unbearable over months.

    You did not fail at statistics. You failed at starting from a blank page, once, and you already know what you are changing.

  13. That is a survivable size.

    And surviving it is the entire reason you get to attempt number twenty.

  14. This week

    Attack something. Five attempts minimum on a real subject, logged four lines each.

    Bring the log to week 9, with whatever list of fights has grown out of it — and if the list changed shape halfway through, bring both versions.

Teaching team