How do you decide a curated problem list is finished rather than merely completed?

## Two different stopping conditions "I completed the list" and "I am done with the list" sound like the same statement and are not. Completion is an inventory fact: every row is checked. Finished is a capability claim: for every category on the map, you can regenerate the approach from nothing. Only the second predicts anything about an interview, because an interview supplies a blank page and no prompt. The reason this distinction gets lost is that checking rows is measurable and regenerating approaches is not, unless you deliberately build the test. ## Recognition versus recall The central failure is mistaking recognition for recall. - **Recognition** is the response you get when the material is in front of you: you read your old solution and every line looks obviously right. The feeling is strong, immediate and almost entirely uninformative about whether you could have produced it. - **Recall** is generating the material with nothing on screen. It is slower, less pleasant and the only one of the two that matches interview conditions. Rereading builds recognition efficiently and recall barely at all, which is why the most common form of fake readiness is a candidate who has reviewed their solutions many times, feels fluent, and stalls in the round. The correction is mechanical: to test recall you must remove the material and add a delay. A same-day self-check measures short-term memory. A week later, the trace is weak enough that what comes back is what was actually consolidated. ## A concrete done-criterion A criterion is only useful if it is decidable — you can run it and get a yes or no, not a feeling. A workable one: 1. **Pick a category at random**, not one you feel good about. Self-selection destroys the test. 2. **State the technique in one sentence** — what it maintains and when it applies — from memory, out loud. 3. **Take a problem from that category you last touched at least seven days ago.** Do not reread your solution first. Sketch the approach: the structure you would use, the state you carry, the cost, and the boundary that is easiest to get wrong. 4. **Then** open your old solution and compare. The gap between what you regenerated and what is written down is the actual measurement. 5. **Score the category, not the list.** Pass means you regenerated the approach unprompted; "I would have got it with a hint" is a fail, because interviews give hints at a price. Run that across every category on the map. A list is finished when the map passes, not when the counter does. ## Reading the result What you do next depends on the shape of the failures: - **A couple of categories fail, the rest pass.** Normal, and the cheapest fix available. Reopen those categories with a short block; do not restart the list, and do not start a second list. The second-list instinct is the count metric reasserting itself — a new list re-covers the categories you already passed and gives the failing ones the same one-pass treatment that just failed. - **Most categories fail.** The problem is the practice mode, not the list. Usually it means solutions were read rather than derived, so nothing was ever encoded for recall. - **The map passes cleanly.** The list has done its job. The remaining gains come from mixed unlabeled problems, from depth in the categories your target loop weights, and from problem-solving under real time pressure — not from more rows in the same list. ## Why the delay is not optional Candidates routinely run the recall test immediately after finishing a category and pass it easily. That result is nearly meaningless: it measures whether the material is still in short-term memory. Interviews happen weeks after most of the studying. The seven-day gap is the cheapest available approximation of the interval that actually matters, and it is the parameter people are most tempted to drop because dropping it makes the test feel better. ## Guarding against a criterion that drifts Write the criterion down before you start the list. A stopping rule invented at the end of a long effort will be shaped by the desire to be finished — the bar quietly becomes "I recognize these" because that is what you can currently pass. Deciding in advance, and choosing categories at random rather than by comfort, is what keeps the test honest. A useful sanity phrasing: if someone described a problem to you over a phone call, with nothing on screen, could you talk them through the approach and the cost? That is exactly the interview's information environment, and it is the standard the done-criterion is trying to approximate.