Lesson 03: Run a review session

Lesson objectives:

• Start a review session with a plain request, and say which items enter the review pool.
• Explain how the box schedule decides what is due today and what a right or wrong answer changes.
• Read the closing report and act on tomorrow's number.

Prerequisites: At least one completed lesson in any course | Previous << 02 Work through a lesson | Next 04 Update a course >>

A finished course starts leaking the day you finish it

You completed the lessons, the exercises passed their criteria, and a week later the terms are fog. That is not a personal failing; memory for unrehearsed material decays on its own schedule. The counter-move is old and well-tested — recall things at stretching intervals, timed to land before you would forget them — and the skill ships it wired to the records you already produced in lesson 02. The cost of entry is one sentence to your agent.

Explanation

The sentence is "quiz me" or "what should I review today"¹. Your agent reads the local learning records the reader has been keeping, computes what is due, and starts asking — no flashcard app, no separate account, no setup step².

What can come up? Two kinds of items². Glossary terms, once you have completed at least one lesson of their course — every generated course ships a glossary of key terms with sourced definitions³. And interaction blocks you have already answered inside lessons; a block you never touched stays out of the pool, so the quiz never asks about material you haven't practiced².

The schedule is the Leitner system — the everyday version is flashcards sorted into boxes by how well you know them⁴. Here there are five boxes with waiting times of 1, 3, 7, 14, and 30 days². Answer an item right and it moves toward the slow end, so you meet it less often; answer wrong and it returns to the fast end for frequent review⁴. An item is due when its box's waiting time has passed since you last answered it².

What this diagram wants you to see is that the boxes are a sorting machine for your attention: material you keep getting right drifts rightward and stops costing you daily time, while anything you miss snaps back to the frequent end. Forgetting is not a penalty here — it is the signal the machine runs on.

Inside a session, the rules are few and deliberate². At most ten items per session, so a backlog never becomes an hour. Items you previously missed come first. For a term, you explain it in your own words before the standard definition is shown — the recall attempt is the treatment, so the order never flips. For an interaction block, the question returns rephrased, and a wrong answer earns a short variant follow-up on the same mechanism. "I forgot" is recorded as wrong, and the tone stays matter-of-fact — a wrong answer is scheduling information, not a verdict².

Each item is recorded the moment it is scored, and the session ends with a fixed report: how many you reviewed, how many right, how many wrong, and how many come due tomorrow². Two edge cases are worth knowing so they don't read as errors. If nothing is due, the agent says you are caught up and tells you tomorrow's count — it does not invent questions to fill the silence². If you have never completed a lesson anywhere, there is nothing to seed the pool, and the agent will send you back to the reader first².

Worked example (follow along)

A two-item session, end to end. You say: "quiz me."

Item 1 — a term, previously missed. The agent asks: "Can you tell me what you understand by 'completion criteria'?" You try: "the checklist under an exercise that says when it counts as done." The agent gives the standard definition, notes that you covered the self-judging part but not that the criteria replace a grader, and records the answer as right. The item moves one box toward the slow end.

Item 2 — an interaction block from lesson 01. The original asked which request skips clarifying questions; it returns rephrased: "Your friend writes 'teach me everything about spreadsheets, from zero to expert.' Does the agent start building?" You say yes. Wrong — scope words are not a starting point. The agent explains the mechanism, then follows up with a variant: "Same friend adds: I do my taxes in a notebook and have never opened a spreadsheet. Now?" You say it builds; right. The original item still records as wrong and returns to the fast end.

The close: "Reviewed 2 items today, 1 right, 1 wrong. 3 items due tomorrow." Twenty minutes of your life it was not — two questions, two records, one number for tomorrow.

Your turn (faded example)

Today is Friday. Decide which of these three items is due, using each box's waiting time:

1. "glossary" — box 2 (3-day wait), last answered Monday. Due? ____
2. "snapshot" — box 3 (7-day wait), last answered Wednesday. Due? ____
3. "skill" — box 1 (1-day wait), last answered Thursday. Due? ____

Answer: 1 is due (four days since Monday, past the 3-day wait); 2 is not (two days since Wednesday, five short of the 7-day wait); 3 is due (one day since Thursday meets the 1-day wait). A session today would ask about items 1 and 3 — and if item 1 was ever missed before, it goes first.

Summary + what's next

You can now start a session with one sentence, say what is in the pool (terms of completed lessons, blocks you've answered), trace an item through the five boxes, and read the closing report as a schedule rather than a scorecard. Wrong answers are routing, not judgment. The records that make this work are keyed to the course's files — which raises a question for the next lesson: what happens to them when you ask the agent to change the course itself?