Lesson 02: Work through a lesson

Lesson objectives:

• Name the fixed parts every generated lesson shares and what each part is for.
• Use completion criteria to judge your own exercise attempt instead of guessing.
• Pick the right copy-to-agent entry point for a stuck moment, a confusing term, or a deeper question.

Prerequisites: A course link from lesson 01 (this course works too) | Previous << 01 Ask for a course | Next 03 Run a review session >>

The page looks like an article. Reading it like one wastes most of it.

You open the link, you see headings and paragraphs, and ten years of reading habits kick in: scroll to the end, nod, close the tab. On this page that habit discards the parts that were built for your hands. Every generated course has the same working surfaces — checklists that let you grade yourself, small in-lesson checks, and buttons that carry your exact situation back to your agent. This lesson walks each one, on this very page.

Explanation

First, the map. The reader's /courses page lists every course on your machine; each course has its own page at its short name¹. The reader runs only on your computer: your reading progress and practice attempts are stored in local files next to the course, and nothing is submitted anywhere¹. That means checking a box records your own judgment for your own later use — there is no server, no grade, no author watching.

Every lesson has the same anatomy, fixed by the skill's authoring rules: objectives at the top, an explanation whose factual claims carry numbered footnotes into a per-course source list, a worked example the author walks through, a half-finished example left for you, exercises, and a summary². The footnotes are not decoration — a generated course is forbidden from stating a key fact it cannot cite, so the number is your path to "says who?"².

Two kinds of practice live inside the lesson body. Interaction blocks are short checks — pick a judgment, order some steps — that grade themselves on the spot; a wrong answer must explain the mechanism you missed, not report "incorrect"². You met one in lesson 01. Completion criteria sit under each exercise: a short checklist specific enough that you can decide pass or fail without a grader². The discipline they replace is "looks done to me," which is the exercise equivalent of nodding at an article.

Then there are the exits — the copy-to-agent entry points. The reader has four, each scoped to a different kind of trouble, and all of them do the same favor: they copy your exact situation, with the course and lesson attached, so your agent starts in context instead of from a cold "I don't get it."

The stuck templates deserve one extra sentence, because they encode a study skill: each one makes you fill in where you are stuck and what you already tried, and each instructs the agent to lead with a hint rather than the full answer³. The deeper-question entry has a triage rule on the other end — if two chat turns can settle it, the agent answers in chat; only a real gap grows into an extension lesson attached below the main one⁴.

Worked example (follow along)

Take lesson 01's Level 1 exercise — the four-part course request — and watch the criteria do their job on a real attempt.

The attempt: "I want to learn photo editing. I take a lot of photos. By the end I want to be better at editing them."

• Each of the four parts is present — fails. Topic, starting point of a sort, goal of a sort; the gap is missing entirely.
• The gap is stated as plainly as "I have never X" — fails. Nothing to check.
• The goal names one action with real material — fails. "Be better" is a feeling; no photo, no action.

Three criteria, two clean failures and one miss — and no grader was needed. The revision writes itself part by part: name the tool, name the never-done thing, name one photo and one fix. That is what criteria are for: they turn "is this good?" into checks you can run.

Your turn (faded example)

You are stuck on an exercise and about to use the stuck card's template. Fill in its blanks so the agent can act on the first read:

I'm on lesson ____ of the ____ course. The exercise asks me to ____. My attempt so far: ____. The completion criterion I can't meet: ____.

Answer: A filled example — "I'm on lesson 02 of the pivot-tables course. The exercise asks me to build a by-region summary. My attempt so far: I inserted the pivot table but every region shows the same total. The completion criterion I can't meet: each region row shows its own sum." Notice what the two attempt blanks force: you name what you tried and where it diverged, which is most of a diagnosis — and why the template forbids arriving with only "it doesn't work."

Summary + what's next

You can now read a generated lesson as equipment: footnotes to check claims, interaction blocks to check judgments, criteria to check your own work, and four copy entry points that carry your exact situation to your agent. Nothing on the page reports to anyone but you. What happens to all those local records is the next lesson's subject — say "quiz me" and they come back as questions, on a schedule tuned to when you'd otherwise forget.