How we verify our content

Last bank-wide pass: September 7, 2026 · applies to the current SAT and ACT question files

A wrong key can teach a student the wrong rule. That is why we publish the checks we can demonstrate and the gaps we cannot honestly call proven. “No issue found” is not the same as a guarantee, and this page does not pretend otherwise.

1. What the bank-wide checker covers

Every stored item in both exam banks is checked for answer indexes that point to a real choice, the expected number of choices, complete hints and explanations, valid question types, valid difficulty labels, duplicate choices, exact duplicate question signatures, and agreement between the full curriculum and the smaller files the browser loads. It also flags drafting artifacts and language that deserves an age-appropriateness review.

The latest pass found no blocking structural faults, no exact duplicate question signatures, no drafting artifacts, and no sensitive-content or pressure-language flags. A small set of repeated generic stems and answer-position patterns remains as review notes; the questions themselves differ, and answer choices are shuffled for students.

2. The blind re-solve, and what it does and does not prove

In September 2026 every one of the 3,575 questions was handed to an independent reviewer exactly as a student sees it — stem, passage, data and choices, with no answer key, no explanation and no hints — and solved from scratch (maths with a computer algebra system). The reviewer’s answer was then compared with the stored key by value. All 3,575 items were re-solved; the reviewers disagreed with the stored key on 5, and in all 5 cases the stored key was checked by hand against the printed item and stands. Two items were tightened editorially (a sentence-addition stem and a science stem) without changing their keys. This is the first time the ACT English, Reading and Science keys and every SAT Reading & Writing key have been independently checked; SAT Math had previously been re-solved by a pattern-based checker that could parse 883 of 1,305 items.

What this does not prove: the reviewers were AI systems working blind with tools, not human subject experts, and their own error rate on this bank was about 0.14%. The wrong-answer rationales and hints were not part of this pass; roughly 1,300 older SAT rationales and the hints remain unproven and are described as such here rather than counted.

Disagreements are leads to investigate, not verdicts: every one is re-solved by a person against the printed question before anything is changed.

3. Test format and policy facts

Details that can change — section lengths, scoring rules, optional sections, calculator policies, and the Digital SAT structure — are checked against the official College Board and ACT sites during a release review. The September 2026 review re-checked every factual claim in the site’s lessons, FAQs, guides and cheat sheets against College Board and ACT pages read on September 6, 2026, confirmed the Digital SAT’s two-stage adaptive structure and timing and the enhanced ACT’s 50-question English, 45-question Math, 36-question Reading, optional 40-question Science and optional Writing format, and corrected nine statements — among them a rule that told students they could revisit questions across a whole section rather than within a module, and a grid-in tips list that omitted the character-limit and fourth-digit rounding rules.

4. What remains editorial judgment

Automated checks cannot prove that every reading inference is the best interpretation, that every distractor is equally fair, or that every explanation is the clearest possible way to teach a student. The blind re-solve covers answer keys only; it does not cover explanations or hints. Those areas require independent editorial re-solving and ongoing reports from students and educators.

5. Practice tests, difficulty and score estimates

Difficulty labels. In September 2026 every one of the 3,575 questions was rated for difficulty, blind, on a five-point scale, by two independent AI reviewers working from different instructions — one judging the steps and traps an item demands, one estimating the share of real test-takers who would answer it correctly — each shown only the question and its choices, never the old label. They agreed exactly on 69% of items and within one level on 99.4%. The 21 items they disagreed about by more than one level went to a third reviewer. Where they disagreed by exactly one level (31% of items) we took the lower rating, deliberately: a question wrongly called hard inflates the “Hard” filter and the score a student thinks they earned, while one wrongly called easy only understates us. That choice matters and here is its size — under the shipped rule 1.3% of the bank rates as genuinely hard (4 or 5 of 5); taking the higher rating instead it would be 4.8%. Either way the conclusion is the same and it is not flattering: this bank is easier than the real tests. The “Hard” filter on many pages is therefore short or absent, and the higher-difficulty second module in our SAT practice tests is the hardest content we have rather than as hard as College Board’s. These are judgements about items, not measurements of students; no rating here comes from watching real test-takers answer.

Form structure. SAT practice tests follow College Board’s Assessment Framework. It publishes domain counts per section rather than per module, so our per-module quotas are those counts halved and are our arithmetic, not College Board’s. We follow its published rules exactly where it does state them: the Reading & Writing question order (Craft and Structure, Information and Ideas, Standard English Conventions, Expression of Ideas, easiest to hardest within a domain), Math ordered easiest to hardest with about a quarter student-produced responses, and a second module that is a different, lower- or higher-difficulty set chosen from Module 1 performance. College Board sets that cut at the population median on an ability estimate it does not publish; this site routes at 18 of 27 (Reading & Writing) and 16 of 22 (Math) correct, figures from independent analyses of official practice tests, not from College Board. ACT practice tests follow ACT’s published reporting-category shares, with Math ordered easiest to hardest as ACT states, and Reading and Science assembled from whole passages with their question clusters intact rather than from single questions. Our Reading section still carries more words per question than a real form, and our English questions come in much smaller clusters than the real test’s five or ten per passage; both are said plainly on the practice-test page. No question appears in more than one form, which is why there are 6 SAT and 4 ACT forms rather than more.

Score estimates. SAT ranges are read from the raw-score conversion tables College Board publishes for its paper practice tests 4–11 (the median lower and upper value across those forms), with the raw score scaled to the paper form’s length. A student routed to the lower-difficulty module has the top of the range held at about 600 (Reading & Writing) or 590 (Math). College Board does not publish a ceiling for that route at all; what it says is: “If you get a couple questions wrong on the first module, scoring an 800 for the section won’t be possible even if you get every answer right in the second module.” The 600/590 figures come from independent analyses of Bluebook practice tests (Applerouth; The Test Advantage), not from College Board, and they are the least certain numbers on this page. ACT scores are read from the conversion table ACT printed for its official practice form, with ACT’s published standard errors shown as ranges — 1.9 points in Math up to 2.7 in Reading, and 1.3 on the Composite, rounded to whole points on screen, so Reading appears as ±3. The real SAT is scored by an item-response model that College Board does not publish; no outside site can reproduce it, and this one does not claim to.

Measured accuracy. Students who have taken the real test can report their official score anonymously alongside our estimate — the pair of numbers and nothing else. We publish a measured error for an exam once 20 students have reported one, because an average built on a handful of students is noise wearing a number’s clothes. SAT: 0 reports so far. No measured error is published for it yet; that needs 20. ACT: 0 reports so far. No measured error is published for it yet; that needs 20. Until an exam reaches that mark, the honest statement about it is that our estimates are built from published tables and have not been checked against a single real score.

6. Privacy and student safety

The learning interface is designed to support short practice, review, and private reflection. It does not need an account for practice, and the progress image is created on the student’s device. Privacy details and the rules for younger students are in the Privacy Policy and Children’s Privacy Notice.

Found something wrong? Tell us.

The last line of defense is you. If any answer, explanation, or fact looks off — even slightly — please report it here. We read every report, verify it, and fix real errors fast. That feedback loop is part of the method, not an afterthought.

This is an independent study tool. Our accuracy process is our own; it is not endorsed by or affiliated with the test makers.