The October 2025 SAT: the hardest math of the autumn

Yes, and unusually so in math. October generated roughly three times as many difficulty complaints as reassurances, the widest gap of any recent sitting, and the reports came disproportionately from students who describe themselves as consistently strong in math. When people who average close to full marks on practice forms walk out unsettled, the form was hard rather than the cohort weak.

Reading and Writing drew far less comment, which is itself informative: when one section dominates the conversation after a test date, the other one behaved. Nor were there many timing complaints. October was hard in the sense of demanding ideas students had not rehearsed, not in the sense of being long.

Two things followed from that difficulty, and both are reasoning mistakes rather than content gaps. Students started trying to work out what their raw score would become, and they started deciding which questions had not counted.

The version circulating after October went roughly like this: every test-taker starts at 1600, and each wrong answer subtracts a fixed amount that depends on the difficulty of the question, so an easy question costs you more than a hard one. It is stated with real confidence, it produces satisfyingly specific predictions, and none of it is how the exam works.

What actually happens has two stages. First your raw score is counted: the number of questions you answered correctly, with no deduction for wrong answers and no weighting by difficulty. A hard question and an easy question are worth exactly the same one point. There is no penalty for guessing, which is why leaving anything blank is a straightforward mistake.

Then that raw score is converted to the scaled score through equating, which adjusts for the difficulty of the specific form you sat and the module you were routed into. This is the part that makes per-question arithmetic impossible from the outside. The same number of correct answers converts differently across forms, and the conversion is not published. That is also why an exam that felt hard does not translate into a worse score than an easy one: the difficulty is already accounted for.

The practical damage of the folk model is that it makes students believe they can compute their score from memory, which turns two weeks of waiting into repeated arithmetic on numbers that do not exist. It also produces the reasonable-sounding but wrong conclusion that a hard form is bad news.

The second theory October produced was a rule for spotting unscored questions: if an item seems far harder than everything around it, it is probably a pretest question that does not count. Students applied it confidently, and named the specific items they were sure about.

The SAT does include unscored pretest questions, so the premise is sound. The rule built on it inverts the purpose. Pretest items exist to gather performance data on questions that may appear on future forms, which only works if students treat them exactly like scored ones. An item recognisable as unscored would be worthless, so they are written and placed to be indistinguishable.

What the rule actually detects is difficulty, which is why it always nominates the questions the nominator found hardest. It is a way of feeling better about a specific question you could not do, and the cost is that you stop treating it as something to go back and learn.

October’s hard math had a recognisable signature: equations containing an unknown constant, where the question asks what that constant must be for some stated property to hold. A quadratic whose two solutions sum to a given expression. A linear equation with no solution. A factored form that has to match a given expansion.

None of these can be brute forced, and this is the specific reason strong students found the section hard. A student who is fast and accurate at solving is not automatically good at working backwards from a property to a constraint, and October asked for the second skill repeatedly. Geometry appeared too, weighted towards volume and surface area of solids rather than coordinate work.

Reading and Writing drew comparatively little comment, and what there was concentrated on evidence questions and on a handful of near-synonym vocabulary items. The evidence questions followed the standard shape: a hypothesis drawn from a single site or a single study, then a question asking which finding would weaken it.

These are more mechanical than they look. A hypothesis built on a correlation observed in one place is weakened by evidence that the same conditions elsewhere produce a different result. You are looking for the choice that breaks the link, not the choice that sounds most negative about the researchers.

Original questions written around what this exam tested, one for each argument above. Nothing here is a real test question.

A hard form is not a worse outcome. Equating means the difficulty of the exam you sat is already priced in, so the useful response to walking out of a hard test is not to estimate your score, which cannot be done, but to write down the questions you could not do while you still remember them.

That list is the most valuable thing you leave a test centre with, and the two theories October produced are both, in effect, reasons not to make it.