Score band and ranking
A score range you can trust, not a number we made up.
A projected Reading and Writing score, a projected Math score and a total, each reported as a range with a stated confidence — ±60 on the total after one mock, ±45 after two, ±40 after three or more, and ±30 / ±25 / ±20 on each section. Underneath: where your points actually went skill by skill, where your time went question by question, which second module you were routed into, and what your wrong answers have in common. And once you have sat the real thing you can tell us what you scored, which is how we find out whether our ranges were any good.
Where this actually stands
Live. Section and total bands with stated confidence, every attempt charted as a band rather than a point, per-skill loss analysis, skill estimates shown as ranges, pacing against the real module clocks, a routing map, an error profile, a written debrief, and your rank this season. The bands stay deliberately wide until there are enough real responses to narrow them honestly, and the calibration check exists so that narrowing can be earned rather than announced.
Every score on this platform is a range. The total carries ±60 points after your first mock, ±45 after your second and ±40 after three or more; each section carries ±30, ±25 and ±20 across the same progression.
Those are wider than the real SAT's own standard error of roughly ±40 on the total — deliberately, because the real test has genuine equating behind it and we do not. Converting a raw count to the 400–1600 scale needs the College Board's equating data, which is not published. An uncalibrated practice form deserves a wider band than the real test, not a narrower one, so the single-mock case starts above that figure and tightens towards it as evidence accumulates.
Scores are reported in ten-point steps, as the real test does. A practice platform that shows you a 1247 is telling you something about itself.
Every mock you have sat appears as a band, with your target as a fixed line you set yourself. The obvious version of this chart is a bar per attempt with the number printed under it, and that version breaks the one rule the product is built on in the largest type on the page — exactly where a student goes looking for progress.
It also does not extrapolate. A line curving towards your goal is the most persuasive thing this screen could draw and the least honest one: two mocks do not establish a gradient and three do not either. The bands are drawn where they were measured, the target sits where you put it, and the space between them is left alone. If the gap is closing, overlapping bands say so without anyone having to claim it.
After you sit the real SAT you can tell us what you actually scored. That number is compared against the band we projected, and the result is aggregated: how often our range contained the real score, and the mean size of the error when it did not.
Almost nothing in this category does this, and the reason is obvious — it is the one measurement that can prove a scoring claim wrong. It is here because a claim of honesty that cannot be checked is just a nicer kind of marketing, and because it is the only mechanism by which the bands could ever be narrowed on evidence rather than on confidence.
Until enough scores have been reported, the page says what the current basis is rather than showing a rate computed from a handful of people.
The score is a summary. The screens underneath it are what produce a study plan.
Per-skill loss analysis sorts by points lost rather than by accuracy percentage, because a skill you got 60% right across twenty questions cost you more than one you got 40% right across five, and sorting by accuracy hides that. Pacing measures your time per question against the real module limits and shows where in each module your wrong answers cluster — the difference between running out of time, not knowing the content, and burning the clock early is three different problems with three different fixes, and the score distinguishes none of them.
There is also a routing map showing which second module you were sent to, which is the single most misread thing on any adaptive score report, and an error profile that looks for what your wrong answers have in common.
Alongside the score, each skill carries an estimate of how well you know it: a probability, from Bayesian knowledge tracing, that moves with every answer that counts — practice tests once submitted, lessons and drills on your first attempt of the day. It is shown as a range, and the range narrows as evidence builds.
Answers given too quickly to have been read are left out of those estimates. They still count on your score, exactly as on test day — the point is only that a lucky or unlucky click should not distort what the platform believes you know.
Each results page opens with a short written debrief: what capped the score, which kind of mistake cost the most, what to do next. The wording is drafted by an AI model; every figure in it is filled in from your own results by our software, and a draft that states a number of its own is rejected. It is labelled as drafted by AI.
What the model receives is a summary of numbers we have already calculated, with nothing that identifies you, sent only to providers that keep nothing. When no such provider is available, or the daily limit is reached, the same debrief is written from our standard wording instead.
Each part is useful alone. Together they are the point.
Missing something you need? Request it.
