Review that actually happens

Recalling a note before you look at it is retrieval practice, and that is the mechanism session 1 put the numbers on. In Roediger and Karpicke's prose study the tested group beat the restudy group 56% against 42% at one week, having lost to them at five minutes. A meta-analysis across 159 effect sizes puts the average benefit at g = 0.50, medium and reliable. So the move is not a stylistic preference. Asking yourself what is in a note before opening it is the same operation, applied to your own material.

The practical catch is the part that decides whether your review survives. Rereading feels better while you are doing it. In that same study the students who restudied most were the most confident they would remember it a week later, and they recalled the least. A review you build around what feels productive is optimising for a signal that points the wrong way.

Which means the good version of this will feel worse than the bad one, every single Friday, permanently. That is worth saying out loud before you design your own, because the pull back toward a comfortable reading session is constant and it never announces itself as a downgrade.

When you open the note and compare it against what you just said, there are only three outcomes, and each one points somewhere different.

• You recalled it accurately. The note is doing its job. Leave it completely alone, do not tidy it, do not improve the wording. This is the most common outcome and it should cost you no time at all.
• You got the gist but not the specifics. The note itself is fine. What is usually wrong is the title: it is carrying more meaning than it can hold, so you remember the shape of the thing and none of its contents. This is the session 2 problem showing up two months later, and the fix is a narrower title or a split into two notes.
• You had no idea what was in it. Two possibilities, and you have to pick. Either the title is wrong, which is again session 2 and takes thirty seconds to fix. Or the note should not have been kept at all, which is a discard decision - and the capture rule in session 2 was deliberately fast and deferred exactly this decision to here, on purpose, because deciding at capture time costs you the capture.

Notice that only the third outcome can send a note out of the collection, and it is the only place in the whole loop where that can happen. A system with no third outcome grows forever.

Cepeda and colleagues pulled together 839 assessments of distributed practice across 317 experiments. Of the 271 comparisons where massed and spaced study could be set against each other, only 12 showed no benefit or a negative one. The authors' own description of the spacing effect is that it is "quite robust". As findings in this area go, that is about as settled as they get.

The part with the most practical bite is not the headline. It is that the interval that maximises retention grows as the delay before you need the material grows. There is an optimal gap, and it moves. More spacing is not automatically better, and neither is more frequency.

Which reframes the whole scheduling question. A quarterly review is not a weekly review done badly. It is the right instrument for a different horizon: things you will need in a year want a wide gap, and the project you ship next Thursday wants a narrow one. Trying to run one interval over an entire collection is what makes the pass too long, and a pass that is too long is a pass that stops.

So the weekly pass in part 3 covers a small, changing set - this week's captures and your two live projects. The rest of the collection is on a much longer clock, and most of it is on no clock at all until you need it.

You will see spacing quoted with a number attached. The two most common are that it "doubles recall" and that it gives "2-3x better retention". Both were searched for directly in that paper's own raw text, and neither is there. Not phrased differently, not in a table, not in the abstract. They are confections that grew in the retelling, and they get repeated confidently because nobody who repeats them has opened the paper.

There is a second, quieter version of the same error: quoting one pooled effect size for the whole meta-analysis, usually as a single d value. That paper does not report one. Most of the studies it covers did not report enough to compute an effect size at all, so what it reports instead is raw accuracy differences and the 271-comparison split you saw above, with effect sizes available for only 23. Anybody handing you a single pooled number for it is quoting something else, or quoting nobody.

This matters more than it looks. The honest version of the claim is stronger, not weaker: 12 nulls in 271 comparisons is a more persuasive sentence than a round multiplier anyone can wave away, and it survives being checked. The dishonest version has the property that the first person who looks it up stops believing the rest of what you said.

• Empty the inbox, five minutes. Everything you captured this week gets a home in one of the four categories. Five minutes is enough for a normal week's capture, and if it is not, that is information about your capture rule rather than about your review.
• Recall-check three notes, four minutes. Three, not ten. Titles only, say what is in each, then open and mark the gap. Rotating three a week gets you through a hundred and fifty notes a year without ever feeling like a study session.
• Repoint the project notes, four minutes. Your two live projects change faster than anything else in the collection, so the links out of them go stale fastest. This is the move that keeps the thing session 4 measured - links moving retrieval far more than filing does - actually true a month from now.
• Mark one thing for discard, two minutes. One. Archive it or delete it, and do it during the pass rather than making a list of things to consider later.

That last move needs its reason said plainly. The capture rule in session 2 is deliberately fast and errs toward keeping, because a capture rule that makes you deliberate is a capture rule you stop using. That choice is only defensible if the deferred decision gets made somewhere, and this is the somewhere. Without it the collection has an entrance and no exit, which is the shape session 1 measured on everybody's real notes.

One item a week is deliberately small. It is not a cleanup and it will never feel like one. But fifty-two decisions a year, each one made while you are actually looking at the note, is enough to change the direction a collection is travelling, and that is the whole ambition.

The quarterly pass has exactly one job: find everything in a project folder that has not been opened in three months, and apply one rule. For each of them there are only three honest answers.

• The project is finished. Nobody declares a project over; it just stops having next actions. Its notes belong in the archive, and moving them there is what makes the remaining project list mean something.
• The project is stalled. This is the valuable one. Three months of not opening something is a fact about the project, not about your filing, and it is a fact you would otherwise never be told. Either it gets a next action this week or it goes to the archive and stops pretending.
• It was never a project. No finish line, no goal, just an ongoing part of your work or life. That is an area, or it is reference material and belongs in resources. Miscategorised areas sitting in a project list are the most common reason a project list feels permanently overwhelming.

The thing that makes all of this possible is that archiving is not deleting. The archive keeps everything, searchably, forever, and asks nothing of you. Almost nobody can delete their own notes, and almost everybody can move them somewhere quiet. A collection with a real archive is one where discarding costs nothing emotionally, and that is the only reason the discard step in the weekly pass survives contact with a human being.