An all-hard practice form is not a harder version of the test. It is a diagnostic with one job: telling you whether difficulty is actually what is going wrong, or whether the problem is pacing, or careless errors, or a handful of topics.

We serve 37 standard forms that mirror a real sitting, 21 all-hard forms where every question comes from the top two difficulty levels, and 6 single-module Math gauntlets. The counts differ for a structural reason worth explaining, because it says something about how hard questions actually work.

What a hard label means here

Every question in the bank carries a difficulty level, and the top two are what an all-hard form draws from. Nothing else about the question changes. A hard question is not longer, does not use rarer vocabulary by design, and is not built to trick you more than any other.

The label needs its own scrutiny, which we learned the unglamorous way. When an item's author supplies its difficulty along with the question, a real share of what comes back labelled hard is not hard. It is a mid-difficulty question written by someone who found it demanding to write. Correctness checks do not catch this, because the question is fine. Only a separate pass that asks how difficult this actually is will catch it, so difficulty is reviewed rather than inherited from whoever wrote it.

That is the honest limit on the word. Difficulty is a judgement we check deliberately, not a measurement we take, and the standard we hold it to is whether it behaves like the hard end of a real digital SAT.

What every question passes before it can be called anything

The verification is the same for hard questions as for the rest, and it is the part that matters more than the label.

Each question is solved independently by reviewers that did not write it, and disagreement stops it shipping rather than starting an argument we resolve ourselves. Maths is checked symbolically, which catches a specific and nasty failure: an answer that is correct arithmetic for the wrong quantity. A value checker passes those, because the number really does follow from the working. Only re-solving the actual question asked catches them.

The hard items get this treatment twice over, because a question at the top of the difficulty range is where a plausible second answer is most likely to hide. An equally defensible wrong option is the most common real defect in a hard question, well ahead of a wrong answer key.

The Math gauntlet throws its own score away

The gauntlet is a single Math module: 22 questions in 35 minutes, every one of them hard. It reports a raw score rather than a scaled one, and it is deliberately kept out of your score history.

That last part is the point. A deliberately brutal session should not drag down the trend line you use to judge whether you are improving. An all-hard module is not measuring the same thing your practice tests measure, so averaging them together tells you less than either does alone. If it fed your history it would make your progress look worse the more of them you did. That is an incentive pointed the wrong way.

So the gauntlet is for the days when you want to work only on the hard end, without paying for it in a number you are trying to watch.

Why there are fewer hard forms than standard ones

This is the structural bit. A practice form has to cover the blueprint, meaning the right mix of skills in the right proportions, not just any 22 hard questions in a row.

Hard questions are not spread evenly across the skills in our bank. Some skills produce difficult questions readily. For others we hold far fewer, because writing one that is genuinely hard and still tests that skill honestly is simply harder to do. Build a form that needs a hard question from each skill and the scarcest skill sets the ceiling on how many distinct forms you can make before you start repeating questions.

That thinnest cell, not the total number of hard questions, is what caps the count. It is also why adding hard questions in bulk does not raise it: adding more of a skill that already has plenty changes nothing. The binding constraint moves only when the scarcest skill gets deeper.

We would rather serve fewer forms that each cover the blueprint than more forms that quietly overweight whichever skills happened to be easiest to write hard questions for.

What an all-hard form is not

It is not a prediction of your score, and reading it as one will mislead you. It has no easy questions to bank, so the number it produces is lower than a real sitting by construction.

It is also not what test day looks like. A real digital SAT opens each section with a broad mix of easy, medium and hard questions. How you do on that first module then routes you to a second module that is either more difficult or less difficult. Either way, no real form is uniformly hard. That routing is described in how the adaptive second module works.

Use an all-hard form for the question it answers. Say your practice scores sit above your real ones and you suspect the questions are too easy. A form with the easy items removed settles that faster than another full-length test. It is also worth ruling out the other explanations first, which whether practice scores predict anything covers.

How these counts are produced

The form counts above are generated from the same served index the app reads, and recounted on every build, so a retired form cannot linger in a number on this page. The gauntlet's question count and time limit are read off a served form rather than from the tool that built it, because checking a generator against itself proves nothing. If any of these numbers stops being true, the build fails rather than the page going stale.