How the tests are made

How the tests are made

Every Margin test is written from scratch. Each passage, study, researcher and data set was invented for the test, so treat the content as practice material rather than as a source.

The structure is the digital SAT’s, module for module: two Reading and Writing modules of 27 questions in 32 minutes each, a 10-minute break, then two Math modules of 22 questions in 35 minutes each. The calculator, reference sheet, directions, timer, Mark for Review and answer eliminator all work the way they do in Bluebook.

Module 1 runs from medium to hard. On the real test, how you do on Module 1 decides whether Module 2 is easier or harder. Margin always gives you the harder Module 2, because that’s the only route to a top score.

Every answer key was checked by solving each question again from scratch, and every Reading and Writing question was reread to make sure exactly one answer holds up.

How they follow College Board

College Board publishes the blueprint for the digital SAT: the four content domains in each section, the skills inside each domain, and roughly how much of the test each domain gets. Margin is built on that blueprint.

• Same skills. Every question is tagged with College Board’s own domain and skill names, the ones the official Question Bank uses, from Words in Context to Rhetorical Synthesis and from linear equations to circles.
• Same order. Reading and Writing questions are grouped by skill and run in College Board’s order: Craft and Structure, Information and Ideas, Standard English Conventions, then Expression of Ideas.
• Same formats. Passages run 25 to 150 words with one question each. In Math, 6 of the 22 questions in each module ask you to type your own answer, close to the real test’s quarter.
• Close to the same mix. The table compares each test’s share of questions by domain with College Board’s published targets.

Tests 1 and 2 lean a little toward Craft and Structure. Tests 3 to 5 bring Reading and Writing closer to the official balance. All five run slightly lighter on Algebra and slightly heavier on data analysis and geometry than the real Math section.

How drills work

A full test shows you where you stand. Drills are how you fix what it finds. There are two kinds.

• Module drills. Open the menu next to any unlocked test and pick one module. You get that module alone, with its real time limit and a raw score at the end. Use them to practice the harder Module 2s, or to retake a section after you’ve reviewed it.
• Redo drills, in beta. Every question you miss, or spend more than one and a half times the test pace on, goes into your Redo bank. A redo drill pulls up to ten of them from one section, missed questions first, and times you at test pace: about 71 seconds a question for Reading and Writing and 95 for Math. Get a question right within that pace and it leaves the bank. Miss it again and it stays.

A third kind of set lives in the question bank: Tailored sets, built from bank questions aimed at the skills and traps your misses point to.

After any drill, open the report, filter to the questions you missed, and tag each one with why: the setup was right but the last step flipped, you didn’t know the rule, you picked an answer that echoed the text, you misread, or you ran out of time. A few drills in, the tags show you which habit to work on.

The question bank and Tailored practice

The bank holds 628 questions written to be harder than the tests: 250 in Reading and Writing and 378 in Math, spread across all 30 of College Board’s skills. None of them appear in the five tests. Each is labeled Hard, Harder, Hardest or Extreme. Extreme questions go past anything on the real test on purpose: long multi-step setups, hidden constants, and geometry that needs two or three facts chained together, so that test-day questions feel a step easier.

Built to beat the calculator. Desmos is allowed on the whole Math section, and it can solve any equation you can type in, fit a curve to a table, or show whether two expressions match. So the bank’s Math questions are written so that typing it in isn’t enough: the equation has an unknown constant, the answer is an expression or a count, the graph only exists after a word problem is set up, or the figure has no coordinates at all. Desmos still helps you check work, and the explanations say when it does.

To decide what “hardest” means, the bank starts from College Board’s own description of its top score band, which calls for early-college texts, subtle inferences, unknown constants, the most useful form of an equation, and circle and trigonometry facts that lower bands don’t need. Reports from recent real tests added the rest: factoring with unknown constants, triangles where the right angle isn’t given, percent changes applied one after another, and transitions that depend on context. Wrong answers are built from the mistakes strong students actually make, such as an answer true to one sentence but not another, the right idea in the wrong direction, a percent taken of the wrong base, or stopping one step early. Every explanation names the trap.

Every Math answer was checked by a computer algebra system that solves the question independently. Every Reading and Writing question was reread to make sure exactly one answer holds up.

How Tailored practice chooses questions. It reads every answer you’ve given: on the five tests, in drills, in earlier tailored sets and in the bank. From those it keeps a running estimate of how likely you are to get a hard question right in each skill, with a method used by adaptive practice systems (an Elo rating, as in chess; see Pelánek, 2016). Then it builds a set in four steps.

• Reviews come back on a schedule. A bank question you miss returns a day later, then 3, 7 and 16 days after each right answer, until it sticks. Spacing reviews out like this helps memory last (Cepeda and colleagues, 2008). Reviews fill at most 30% of a set.
• Points at stake decide the mix. Each skill’s share of the set is set by how often it appears on the real test and how likely you are to miss a hard question in it, with a boost for recent misses and a first look at skills you haven’t tried.
• Each question is pitched just past comfortable. Inside a skill, it picks questions you’re predicted to get right about 70% of the time, hard enough to teach without being hopeless (Wilson and colleagues, 2019). It also leans toward the traps you tag in your reports and toward the topics you missed on the tests.
• Skills are mixed. No two questions in a row come from the same skill when that can be avoided. In a classroom study, students who practiced with mixed problems scored 61% on a test a month later, compared with 38% for students who practiced one type at a time (Rohrer and colleagues).

Matched or hardest. Matched to you is the default described above. Hardest only skips the matching and draws only from Hardest and Extreme questions, extreme ones first, for practice that should feel harder than the test.

Timed sets run in the test screen at test pace and get a full report, and their questions feed the next set. Untimed practice shows each explanation as soon as you answer. Everything stays in this browser.

Concept drills

Some misses come from one idea you haven’t nailed down yet, like mixture problems, work rates, the discriminant, or completing the square for a circle. Concept drills take 26 of those ideas one at a time. Each starts with a short primer: the key idea, the trap that catches most students, and what Desmos can and can’t do for that kind of question. Then come practice questions in three levels: a warm-up, test level, and beyond the test. After four right in a row you can move up a level.

The drill questions are generated from templates with exact arithmetic, so every answer is right by construction, and each explanation works through the steps. Your drill answers count toward the skill estimates that Tailored practice uses, and a missed bank question links to the drill for its skill.

How scoring works

Each full test gets an estimated score from 400 to 1600, made by converting your raw score in each section with a fixed table. College Board scores the real test with a statistical model that weighs each question’s difficulty, which nobody outside College Board can reproduce, so treat the number as a close estimate. Single modules and drills show raw scores only.

The Art of Solving Problems

Taking a test on paper

Every test also comes as a PDF laid out like a paper SAT booklet, with an answer sheet, an answer key, a table for estimating your score, and explanations at the back. Paper mode opens it in the browser so you can write on it with a pen, an Apple Pencil or a mouse, time each module, and fill in an answer sheet that checks itself. Your markup stays in this browser.

About Margin

Margin is independent practice made by Lamai Hons. It isn’t affiliated with or endorsed by College Board, which owns the SAT and Bluebook trademarks.