The Weekly Loop: Read, Reconstruct, Make, Defend, Log

Reading a hard book feels like learning, and often is not. The weekly loop is a five-step routine that forces the reading to produce something you can inspect: an argument laid bare, a made thing, and a record of what changed. The loop and its AI rules are the author's design, so they are Provisional; the learning-science ideas underneath them are better supported, as noted below.

Why reading alone is not learning

Fluency is a poor guide to understanding. A passage that reads smoothly can leave you unable to restate it. Research on retrieval practice finds that trying to recall material improves delayed retention more than rereading does, although rereading can score better on an immediate test (Roediger and Karpicke 2006; Dunlosky et al. 2013 rate practice testing as high utility). Established for retention of facts and text; weaker evidence for deep understanding of arguments. The loop builds retrieval and production into every week.

READ  ->  RECONSTRUCT  ->  MAKE  ->  DEFEND  ->  LOG

The five steps

1. Read

One primary text, slowly. Fewer pages and more thought. Methods for difficult books are in How to Read a Great Book: Levels, Modes and the Completion Standard. Resist reading the summary first.

2. Reconstruct

State the argument as numbered premises leading to a conclusion. A reusable template:

1. Thesis in one sentence, in your own words.
2. Argument: P1, P2, P3, therefore C.
3. Hidden premises the argument needs but does not state.
4. Weakest premise, and why.
5. Strongest objection, and the author's likely reply.
6. Your verdict.

For a worked example, see Peirce on the Fixation of Belief. Practice at spotting valid and invalid inferences is in Proof and Precise Reasoning: From Arguments to Theorems if you want a formal footing.

3. Make

Produce something: an essay, a proof, a program, a diagram or a write-up. The made thing is the evidence that you understood. Writing as a way of thinking is the subject of Technical Writing as Reasoning.

4. Defend

Face the strongest objection. A person comes first when one is available; an AI second. Revise your draft in response. If you cannot answer the objection, say so in the draft.

5. Log

One line: what I did, what was wrong, what I changed my mind about.

The minimum viable week

If a week has only ninety minutes, do Reconstruct and Log on one passage. That counts. A plan you can keep at its minimum is worth more than an ambitious one you abandon.

A 20-minute weekly review

Ask three questions only:

1. What did I actually make this week?
2. What was wrong, and what did I change my mind about?
3. What do I cut or keep for next week?

Rules for AI

Treat an AI as a tutor, adversary and lab assistant, never as author. The reason is the idea of desirable difficulties: effortful processing tends to improve long-term learning even when it feels slower (Bjork 1994). Established as a general finding in cognitive psychology; how it applies to AI tools is Provisional, since the tools are new.

The reconstruct-without-it test: after any AI session, can you rebuild the result without it? If not, you consumed rather than learned. A longer treatment is in Technology, AI and Human Judgment.

A reasoning notebook and an error taxonomy

Keep one lightweight notebook rather than an elaborate system. When you find an error, label it. A useful taxonomy:

• Observation: you misread or mis-recorded what happened.
• Definition: a key word meant two things.
• Inference: the conclusion did not follow.
• Causality: you took correlation or sequence for cause.
• Probability: you misjudged how likely something was.
• Systems modelling: you ignored feedback, delay or coupling.
• Execution: the plan was sound and you did not carry it out.
• Judgment: you chose badly with the information you had.

Reviewing which labels recur shows where practice is needed. Procedures and abilities for mathematics are treated in Learning Mathematics: Procedures and the Six Abilities.

Audio protocol for walks

Audio suits the first contact with a text but is not reconstruction. A sequence that respects that:

1. Listen straight through; state the thesis in sixty seconds.
2. Relisten, pausing at each section to say its claim aloud.
3. At a desk, reconstruct in writing with the text open.
4. Walk again without audio and think only about the weakest premise.
5. Write 300 to 700 words, take one round of objections, revise, log.

Failure modes

• Collecting lists. Hold one text and one question.
• Reading without producing. No week counts without something made.
• Rebuilding the plan instead of running it. Revise the method only at a scheduled review, and only from evidence.
• AI doing the thinking. Produce first, always.

Over the long run, organise questions with Principles Ledger and Question Arcs.

Try this

1. Pick a two-page text. Reconstruct its argument with the six-line template and mark the weakest premise.
2. Run the 20-minute review at the end of the week and cut one commitment.

Further reading

• Mortimer J. Adler and Charles Van Doren, How to Read a Book (1972).
• Dunlosky et al. (2013), "Improving Students' Learning With Effective Learning Techniques".
• George Polya, How to Solve It (1945).