Why no honest SAT practice test can give you a real 400–1600 score

The short version

On the old paper SAT, scoring was mechanical: count the correct answers, look the raw score up in a table printed with the test, read off the scaled score. Practice tests could reproduce that exactly, because the table was published.

The digital SAT does not work that way. College Board states that scores are "a product of several factors, characteristics of the questions they answered right or wrong (e.g. the questions' difficulty levels), and the probability that the pattern of answers suggests they were guessing." The consequence is stated plainly: two students who answer the same number of questions correctly may earn different section scores.

Why the difficulty of your questions matters

The digital SAT is multistage adaptive. Each section has two modules. The first is the same broad mix for everyone; how you do on it determines whether your second module is the harder or the easier panel. A student routed to the harder second module is answering questions worth more, in the statistical sense, than a student routed to the easier one.

So "35 out of 44 correct" is not a score. It is not even enough information to produce a score, because the answer depends on which 35.

The machinery that turns your answer pattern into a 200–800 section score is a set of calibrated item parameters and an ability-estimation model. College Board does not publish either. Nobody outside College Board can run that calculation, and anyone who claims to is estimating.

What about the questions that do not count?

There is a second reason a raw count misleads. College Board embeds unscored pretest questions: "Two pretest questions are also included in each module ... Student responses to these pretest questions don't affect their scores." With four modules, that is 8 of the 98 questions you answer that do not count toward anything, and you are not told which ones. Our practice contains no pretest items, so every question you see here counts — which makes our counts and the real exam's counts different things.

So what should you measure?

Useful measurements exist; they are just not a fabricated 1600.

• Accuracy by skill. "I get 40% of inference questions right and 85% of linear-equation questions right" tells you what to do tomorrow morning. A composite score does not.
• Time per question. Running out of time and not knowing the content are different problems with different fixes, and timing data separates them.
• Error patterns. The same misconception usually produces the same wrong answer repeatedly. That is findable, and fixable.
• Trend. Whether your accuracy on a skill is climbing over weeks is more informative than any single session.

This is why our results pages report exactly those things and no scaled score. It is not modesty; producing a number we cannot justify would be the dishonest option.

The one place you can get a real score

Official full-length practice tests taken in College Board's own Bluebook application are scored by College Board's own model. If you want a scaled-score estimate, that is the way to get one that means anything. Use ours for diagnosis and drilling, and theirs for score prediction.