write-learning-sheet

name	write-learning-sheet
description	Use when writing a new learning sheet from scratch for a specific week — reads textbook sections, follows templates, and produces a single learning-sheet.typ file
argument-hint	[week-number]

Write Learning Sheet

Week number N = $ARGUMENTS.

Inputs (read all before writing)

1. coursedesign/schedule.typ — find textbook sections assigned to week N
2. textbook/*.md — extract ALL definitions and theorems from those sections
3. templates/week-template/N.learning-sheet.typ — start from this template (structural comments embedded)
4. week1/1.learning-sheet.typ — canonical reference for prose quality (skip if generating week 1 itself; rely on templates/learning-sheet.typ + coursedesign/weekly-materials-guide.md — week 1 becomes canonical after approval)
5. weekN/plan.md — specific requirements (if exists)
6. week{N-1}/* — ensure NO content overlap with previous week

Output

Create weekN/N.learning-sheet.typ.

Hard Rules

• Include EVERY definition and key result (theorems for math/CS; laws, postulates, principles, governing equations for physics / engineering / empirical sciences) from assigned textbook sections
• Define before use — no forward references
• FIDS is a mindset, NOT explicit boxes
• Blog-like prose, not fragmented bullets
• No page breaks between parts (only first page has pagebreak)
• No author footnotes

Which environment to use

The template provides two parallel sets of boxed environments, same visual style:

Course type	Box for named result	Box for justification
Math / CS / theory	#theorem(...)[...]	#proof[...]
Physics / engineering / empirical	#keyresult(numbering: none, title: [Newton's Second Law])[...]	#derivation[...]

Pick one set per course (read textbook/*.md and weekN/plan.md to decide — if the source material proves things from axioms, use theorem/proof; if it states laws and derives consequences, use keyresult/derivation). The "Proof Quality" criteria in CLAUDE.md apply identically to #derivation — read key insight as physical insight, why non-trivial as what naïve picture this overturns.

---

Writing Craft Guide

The template (templates/week-template/N.learning-sheet.typ) tells you what goes where via structural comments. This section tells you how to write it well. Read both.

Task Box

• Open-ended: multiple reasonable approaches possible; no single "right answer" visible at first glance
• Clued by the theory: after completing the sheet, students see how this week's concepts provide a principled approach
• Self-test: can you return to this problem in the Summary? If not, redesign.
• Generate 3 candidate task framings internally, then pick the best one

Bridge

The 2 paragraphs after the task box are the most important prose in the sheet. Use Tao's "ladder of failures": the reader should feel stuck before the concept arrives, then feel relief when it's named.

Opening Prose (per Part)

The goal is to make the reader want the upcoming definition. Two craft moves:

• Terminology inventory (Bendersky): name every component of the definition informally before the formal box. If the definition has 5 parts, the prose must introduce all 5.
• "What would you try?" (Gowers): pose the question the concept answers. Show what breaks without it.

Formal Definitions & Theorems

Critical rule: every component must have been named in the preceding prose. The formal box should feel like a summary of what the reader already understands, not new information.

Worked Examples

The example should make the definition click, not just illustrate it. Use Kun's concrete-first approach: the reader should understand the concept from the example even if they skip the formal definition.

Proofs

Two anti-patterns to avoid:

1. Rabbit-out-of-hat: every construction must be motivated by what a solver would naturally try
2. Unmotivated key step: name the key insight before the proof body ("The trick is...")

Structure: plain-English goal → key insight callout → formal proof → remark on what would break without a hypothesis.

AI Prompts

• DEEPER (~2-3 per Part): escalate in difficulty within a Part
• BROADER (~1 per Part): connect to real world or frontier research
• Each must be directly copy-pasteable into an AI chat

Summary

Use Aaronson's "ceiling-hitting": celebrate what this week's model can do, then show its limit as the teaser for next week.

---

Advanced Techniques (reference — use when natural)

Technique	When to use	Example from week 1
Ladder of failures (Tao)	Bridge and Part transitions	"You might try parsing manually... but this gets messy fast"
Proof as algorithm (Gowers)	Deeper Mechanics subsections	"Put yourself in the machine's place"
Ceiling-hitting (Aaronson)	Summary "What's next?"	"Are all languages regular? No."
Confession of confusion (Kun)	Before complex definitions/proofs	"This looks intimidating, but the idea is simple..."
Structural metaphors (Aaronson)	When a real-world system maps 1:1	"automatic door = DFA with two states"

---

After Writing

Compile with typst compile weekN/N.learning-sheet.typ to verify no Typst errors.