No published study settles this, and anyone who tells you otherwise is selling something. In practice the decision is made for you by two things: how many questions you can genuinely complete and review in the days you have left, and what your wrong answers are actually made of. Work out the arithmetic first, classify your misses second, and the choice usually makes itself.
Start with capacity, not philosophy
Write down three numbers.
D: days left before the exam, minus the days you will lose to assessments, travel, and the day you take off before test day.
T: questions you can answer and review properly in one day, sustained. Review is the slow half. If a 40-question block takes an hour and reviewing it carefully takes ninety minutes, your real throughput is closer to 60 questions a day than the 120 you might imagine.
Q: questions in each option: unused items, incorrect items, marked items.
Now multiply. If D × T comes to 1,200 questions and a full second pass is 2,400, the second pass is not a strategy; it is a plan to skim 2,400 questions without reviewing them. Skimming is where the value of a question bank goes to die.
Most students discover at this point that they have capacity for one of the two options, not both. That is useful information to have before week three instead of after it.
A worked example
Exam in 24 days. You lose two days to full-length assessments and one to a rest day, leaving 21 usable days.
A 40-question block takes about an hour to answer and about ninety minutes to review properly. Two blocks a day is five hours of question work, which is already a heavy day when reading and notes have to fit around it. Call throughput 80 questions a day.
Capacity: 21 × 80 = 1,680 questions.
Now the options. Suppose you have 900 unused items and 640 incorrects. Both fit, only just, with nothing left over for the second look that hard items deserve. Suppose instead you have 1,900 unused items: a full second pass was never available, and the only real question is which subset of them to use.
Do this arithmetic on paper before you commit. The common failure is choosing a strategy that quietly requires 140 reviewed questions a day and then blaming yourself in week two for falling behind a plan that was never achievable.
The questions you got right by luck
Your incorrect list is systematically incomplete. Every item you narrowed to two options and guessed correctly is filed as a success, and it is exactly the material most likely to fail you under exam pressure.
Fix it with marking discipline: mark any question where you were down to two choices, where you needed more than a moment to commit, or where your reasoning was different from the explanation's. Review incorrects and marked items together.
Expect that list to be substantially larger than your incorrects alone, which is a reason to redo the capacity arithmetic above rather than a reason to skip the marking.
Then classify what your misses are made of
Pull your last 200 wrong answers and sort them into four buckets:
- Knowledge gap: you did not know the fact or the mechanism.
- Misread: you knew it and answered a different question than the one asked.
- Timing: you ran short and guessed, or rushed the last ten items.
- Second-guess: you had the right answer and changed it.
Only bucket one is repaired by content review. Buckets two through four are process problems, and re-reading explanations does almost nothing for them. They are repaired by fresh questions under exam timing, where the process is actually exercised.
If your first bucket dominates, incorrects-only makes sense. If buckets two, three and four together make up most of the pile, you need fresh timed blocks far more than you need to reread anything.
What each option actually buys
A full second pass buys breadth and repetition. It also contaminates your percentage: you remember items, your accuracy climbs, and the number stops being a measurement. The exposure is real; the feedback is not.
Incorrects only buys the highest density of material you have already proven you do not know. Its blind spot is the questions you got right by elimination or luck; you will never see them again, and they are not distinguishable from the ones you knew cold.
Fresh unused questions in random timed blocks buy the cleanest signal of the three, because nothing is remembered and nothing is filtered. If you still have a meaningful pool of unused items, this is usually the best use of them.
A reasonable compromise, when capacity allows it: incorrects plus unused items, mixed into random timed blocks rather than studied as separate lists. Mixing preserves the retrieval practice that a filtered list removes.
The measurement problem nobody mentions
Whichever you choose, your question-bank percentage will stop telling you whether it worked. Second-pass accuracy is inflated by recall. Incorrects-only accuracy is depressed by selection, since you are attempting only the items that already beat you.
So do not judge the decision by the percentage. Judge it with a scored assessment sat under exam conditions, converted the same way each time. Running a raw result through the Step 1 conversion tools before and after a two-week block gives you something a qbank average cannot: a comparison on a stable scale.
It also helps to write down, before the fortnight starts, what result would count as the strategy working. "My next form estimate is at least five points higher" is a testable prediction. "I will feel more confident" is not, and confidence is the first thing a hard block of questions takes away.
Keep the assessment schedule fixed when you make this choice. Changing the study method and the measurement instrument in the same fortnight leaves you unable to attribute the result to either.
Defaults by timeline
These are starting points, not prescriptions. Individual situations differ enormously, and a plan you will actually follow beats an optimal plan you abandon.
More than three weeks out. Mix incorrects with unused items in random timed blocks. Keep one full assessment scheduled at the midpoint.
Two to three weeks out. Prioritise incorrects tagged as knowledge gaps, especially in high-yield systems you have not touched in a month. Keep at least one block of fresh questions daily for pacing.
Under two weeks. Stop trying to cover new territory. Work incorrects in your weakest two or three systems, keep a daily block for rhythm, and protect the time you have set aside for your final assessment.
If you want the schedule written out against your actual exam date, the study plan builder turns a current estimate and a date into a week-by-week structure. To keep a minimum daily habit on the days everything else falls apart, one question a day is better than nothing, and it costs four minutes.
Quick answers
Does a second pass raise your score? Nobody can answer that with evidence. There is no published, verifiable study comparing second-pass and incorrects-only strategies for these exams.
Is my second-pass percentage meaningful? Only as a rough consistency check. It is inflated by recall, so it should never be compared against your first-pass number.
Should I reset my question bank? Resetting removes your incorrect list, which is the most valuable artefact you have built. If you reset, export or screenshot the tags first.
How many questions a day is realistic? Count review time, not just answering time. Whatever number you pick, if you are not reviewing every miss the same day, the number is too high.
What if I have both a large unused pool and a large incorrect list? Mix them. A random block drawn from both is closer to exam conditions than either list studied on its own.
What this framework cannot do
It cannot tell you which content you are weakest in; only your own error log can, and building one is the prerequisite for every recommendation above. It cannot account for a rotation schedule, illness, or a bad fortnight.
And it makes no promise about points. There is no honest conversion from "questions completed" to "score gained"; if there were, this site would publish the calculator for it. If you want more free questions to work with while you decide, the practice set here needs no account and no email address.