Every first call with a parent arrives at the same question, usually inside ten minutes. How many points, and how long. It is the right question. For years the honest answer has been a shrug dressed up as expertise, because nobody had counted.
Mike McGibbon at Mathchops counted. Working from about 3,500 students in his platform’s data, he put the average cost of 100 Math points at roughly 600 questions attempted. At a minute a question, that is about ten hours of answering.
The Number
Six hundred questions, on average
Three of these four come from the Mathchops analysis. The fourth is ours, and it is the reason the article does not end at the third.
students in the dataset behind the estimate
Mathchops, Sept 2026
questions attempted per 100 Math points, on average
Mathchops, Sept 2026
of answering time, at roughly one minute per question
Mathchops, Sept 2026
score promises we will make to any family, before or after seeing this data
Corngold Prep
What the average hides
Six hundred is not the price for everyone
McGibbon breaks the number apart himself, and the split is the most useful thing in the piece. Hold volume constant at about a thousand questions: students starting below 600 in Math averaged gains north of 120 points. Students starting above 650 struggled to reach 100 on the same volume.
Same reps. Roughly a third less movement at the top. That is not a fluke of one platform, and it matches what happens in sessions.
The mechanism is hit rate. A student sitting at 520 has gaps that are common and foundational — sign errors, fraction arithmetic, slope-intercept form, percent multipliers. Almost any set of thirty questions contains three of them. Every practice block lands on something worth fixing, so reps convert to points at a high rate.
A student sitting at 690 has already cleared everything common. What remains is rare: the disguised system, the function-notation composition, the geometry problem that hides a similar-triangle ratio inside a coordinate setup. Those show up a few times per test at most. To encounter one, the student has to work through many questions that teach nothing, because they were already answerable. The reps are not less valuable. They are less frequently relevant.
What it costs
The same 100 points, three different bills
Volume figures below come from the Mathchops data. The schedule translations and the coaching load are ours — arithmetic on their number, plus what these programs actually look like from inside a session.
Fastest conversion
Starting Math near 500
The cheapest 100 points on the test, and the ones most families underestimate. The gaps are common, so nearly every practice block finds one.
- Roughly 600 questions puts this student at or above the average gain, and the data suggests better.
- Ten weeks at 60 questions a week — about fifteen questions, four days a week.
- Fluency first. A student computing 7×8 by counting is spending working memory the problem needs. Five minutes of fact drilling daily is not optional here.
- Coaching load is moderate. Most misses are diagnosable from the work on the page.
The stall zone
Starting Math near 600
Content is mostly present and retrieval is slow. This is where most students plateau, and where more material is the wrong prescription.
- Roughly 600 questions is a fair planning figure, weighted toward timed sets over untimed ones.
- Twelve weeks at 50 a week, run in sets of five to ten under a clock rather than in full modules.
- The leak is usually execution — a skipped check, not a missing concept. Error logs matter more than new topics.
- Desmos fluency starts paying here. Ten minutes a day on the tool itself.
Most expensive
Starting Math 680 and up
Despite having the strongest foundation, this student pays the most for the next 100 points. The remaining errors are rare, so they are expensive to even locate.
- Well past 1,000 questions on the Mathchops figures, and the last 40 points cost more than the first 60.
- Real released College Board questions at Medium and Hard only. Authored imitations do not contain the rare cases.
- Sessions shift to error analysis — not how the concept works, but why this specific trap worked on this specific student.
- Coaching load is highest. Finding a gap that surfaces twice in a thousand questions is the job a platform alone does slowly.
The part that is easy to misread
Six hundred questions on a platform that grades you instantly is not the same as six hundred questions from a book nobody ever checked.
Corngold Prep Academy
Where we add something
The reps in that number were all reviewed
This is the detail worth holding onto. Every one of those 600 questions was answered inside a system that told the student immediately whether it was right and showed the solution. Feedback was built into the rep. It was not a separate act of discipline that a tired sixteen-year-old had to choose at 10 p.m.
Handing a student 600 questions out of a book and grading them Sunday produces a different result from the same 600 on Mathchops or in Bluebook, and the number above does not transfer. A rep that ends without the student learning why the answer was wrong is not a smaller rep. It is a different activity.
That is also the argument for a tutor rather than against one. The platform closes the loop on what was wrong within seconds. It cannot always close the loop on why — whether the student misread the setup, knew the rule and skipped the check, or has a fluency hole three layers down that keeps producing the same slip in different costumes.
A rep that counts
- Graded within minutes, not days.
- The miss gets named — question type, and the specific check that was skipped.
- Logged in two columns, so the repeats become visible by the fourth day.
- Re-attempted cold a few days later, without the solution nearby.
A rep that doesn’t
- Answered, checked, moved on. A red mark is not a diagnosis.
- Reviewed by reading the solution and nodding. Recognition is not retrieval.
- Filed as a careless error, which is where real error patterns go to hide.
- Done in a full-length test that gets scored and never opened again.
What 600 reviewed questions looks like on a calendar
Volume and the ten-hour answering figure are from the Mathchops analysis. The review time, the totals, and the weekly split are our estimates, built from what review actually takes in our sessions. Session time with a tutor sits on top of this, not inside it.
Where Corngold Prep Stands
Why we still won’t promise you 100 points
A number this good makes a promise tempting, and we are not going to make one. An average across 3,500 students describes a population. It forecasts nothing about one sixteen-year-old, and the variables that decide the outcome sit mostly outside a tutor’s control: whether the reps happen on the weeks nobody is watching, how the fall sports schedule lands, sleep, a bad Module 1 that sends the adaptive second module somewhere unhelpful.
Every prep company that advertises a guaranteed point gain is quoting a population average at an individual family and charging for the confidence. The data does not support that transfer. It never did.
What is promised here instead: the specific weak areas get named by question type rather than by chapter, the reps get reviewed while the memory of the attempt is still warm, and the read a parent gets is honest in both directions. When something is not working, that shows up in the update near the top, not buried in paragraph three.
Despite the size of that dataset, the most useful thing in it is the split, not the headline. A family planning around 600 questions for a student at 690 is planning with the wrong number, and will read a real result as a failure. Knowing the actual price is what makes the commitment survivable.
The average is real. Your kid is not the average.
From Corngold Prep
Want the honest version of this number for your student?
It starts with a diagnostic and an error log, not a package. For focused SAT and ACT preparation built around the gaps a diagnostic actually shows, we’re always glad to talk.