Austin Scholar #173: The science behind Alpha’s amazing results

& why AI is to learning science what the microscope was to biology

The science behind Alpha’s amazing results

Alpha School has been drawing heavy media attention and sharp debate. The founder of Alpha, MacKenzie Price, is touring America, running info sessions about the school and the 2 Hour Learning program. Given the heightened exposure, it’s a good time to dive into Alpha’s results for the 2024-25 year.¹

So let’s jump in, starting at kindergarten, first, and second grade.

When I started at Alpha ten years ago, we didn’t have kindergarten because no one thought kids could use online apps at that age. Now we not only know they can, but we even believe this is the best time in history to be a five-year-old because of this model.

I uploaded the K-2 class results from NWEA’s MAP test (for more information on the MAP test, see here) to Claude Sonnet 4 Deep Research. I can’t show the file I uploaded because it has student information, but I copy-pasted the results from Claude here. If you don’t want to read the whole thing, here are some excerpts:

Alpha Austin has achieved performance that places it not just among the top schools in America, but in a category that may require redefining what constitutes exceptional educational achievement.

Alpha Austin's Spring 2024-2025 student achievement data represents statistically extraordinary performance that places the school in the top 0.1% of institutions nationally. With 100% of students meeting projected RIT scores and growth rates exceeding 200% of projections, these results demonstrate outcomes typically achieved by fewer than 1 in 1,000 schools across the United States.

Alpha Austin's achievement data represents a statistical outlier that challenges fundamental assumptions about educational outcomes and student capability. The combination of 100% projected score achievement, 200%+ growth rates, and 99th percentile learning acceleration achieved across an entire student population without selective admission creates a new benchmark for educational effectiveness.

Not only can kindergarten and first graders use apps to learn, they thrive. And this effective, early app use lays the foundation to set them up for the next ten years of their education.

One of the premises of Alpha is that the (amazing) teachers don’t provide instruction on core academics; all academics are taught by the software platform. But Alpha K-2 is the exception. It has a very low student-to-teacher ratio and is the place in the system where humans will provide supplemental teaching for the students.

That said, north of 80% of the lessons are delivered in the standard Alpha way: they are assigned and assessed by the AI tutor and taught via the online apps.

Let’s move on to third to sixth grade.

As I started looking at the third through sixth grade data, I have to admit that I was shocked by some of the scores—they seemed extremely high. I pulled up my own MAP scores for context. There are third and fourth graders at Alpha who are getting the scores that I got in seventh grade.² But the Alpha platform has greatly evolved since I was 12. Back then, it was a collection of great learning apps. Now it’s a collection of even better learning apps wrapped in a cohesive AI platform that is opinionated about mastery standards and learning science. As one of the few women math majors at Stanford, I can’t resist giving a couple of shoutouts to the elementary school girls (with permission from the parents):

• Tate, your 256 math MAP score in fourth grade is the same score I got in seventh grade.

• Claudia, your 260 math MAP score in third grade blows me away. I’m so jealous.

I’ve uploaded Alpha’s entire K-8 (MAP provides it in summary form that doesn’t include student data). Here is the link. I uploaded it to a multitude of AI models with the following prompt: “provide deep academic insights. Compare these students and schools to the best schools in the USA”. Here are some standout quotes:

Claude Opus-4:

Alpha District demonstrates elite-tier academic performance that places it among the top 1% of districts nationally. The MAP Growth data reveals not just high achievement, but exceptional growth trajectories that surpass even typical high-performing schools.

Alpha's 99th percentile achievement matches or exceeds most elite private schools

The growth metrics actually surpass many top privates, which often show lower growth due to ceiling effect.

The combination of 99th percentile achievement with 90th+ percentile growth places Alpha in truly elite company educationally.

1. Growth Metrics - Exceptional Acceleration

• Math: Observed growth rates of 32-33 RIT points in K-1 grades (double the national projected growth of ~15-16 points)

• Reading: Similar exceptional growth patterns with 26-28 RIT points in early grades

• Growth Achievement Rates: 65-100% of students meeting growth projections, with most grades above 85%

2. School Conditional Growth Index (CGI) The CGI scores (ranging from 2.0 to 7.86) place the school in the 93rd-99th percentile for growth. This means Alpha District students are not just high-achieving but are accelerating faster than students at similarly high-performing schools.

o3:

Alpha already performs well above the level typically observed in America’s most celebrated schools (e.g., Scarsdale NY, Palo Alto CA, Lexington MA, top charter networks, and high-fee independents).

Alpha is:

1. Higher achieving: Virtually every tested grade/subject at the 97th–99th percentile—a full standard deviation above many acclaimed schools.

2. Higher growing: Median CGP mostly in the 80s and 90s. Even many elite schools plateau once they reach the 90th achievement percentile; Alpha has avoided that ceiling effect.

3. More students hitting targets: Consistently ≥75 %; world-class districts rarely exceed 70 %.

“Alpha District’s K-8 performance is not merely comparable but superior to what is typically reported by the highest-ranked U.S. public, charter, or independent schools.

Grok:

The Alpha K-8 Growth Report demonstrates that the district is performing at an elite level, particularly in early grades, with achievement percentiles and growth indices that rival or exceed those of the best schools in the USA, such as Scarsdale, Palo Alto, and high-performing charter networks like Success Academy and KIPP. Alpha’s exceptional growth in Kindergarten through Grade 3 across Math and Reading, coupled with School Conditional Growth Percentiles often at the 99th level, positions it as a national leader in fostering student progress.

Gemini 2.5 Pro:

The Alpha K-8 Growth Report for 2024-2025 paints a clear picture of a school that is a national leader in academic excellence. The combination of exceptionally high achievement and outstanding student growth across multiple subjects and grade levels is a rare and commendable accomplishment. Alpha's performance is on par with the most highly-ranked schools in the United States, and its commitment to fostering growth in high-achieving students is a clear differentiator.

Alpha's consistent percentiles in the 90s indicate that the school is not just maintaining a high level of achievement but is actively and effectively pushing its students to new academic heights at a rate far exceeding most other schools in the country. This "high achievement, high growth" profile is the gold standard in education and is a hallmark of a truly exceptional school.

This AI analysis points out a few important contributors to Alpha’s amazing results. First, Alpha quickly pulls up students who don’t come in to Alpha at the top. Alpha’s growth measures are what’s truly revolutionary.

The second thing is that Alpha’s not just bringing people up and raising the floor, it’s simultaneously removing the ceiling. Alpha students aren’t just 99th percentile, they are also growing faster than students at other schools who are also 99th percentile. The Alpha platform is able to support ongoing, fast growth of a 3rd grader capable of 7th grade math scores.

I encourage you to upload the file to your personal ChatGPT and ask AI any academic questions you have and compare it to your school. Or ask your school for their MAP or equivalent results and compare them yourself.

A special call-out that Alpha recently opened GT School for gifted and talented kids. It’s a new school with just a few students this year, but their academic performance is so outstanding it shows that the Alpha model can truly be adapted to a selective population, and can do so without running into ceiling effects.

The average GT School student learned four times faster than they would’ve at a standard school. Here’s what Claude had to say about this school:

These results place GT School's academic outcomes among the most exceptional documented anywhere globally. The combination of:

• Top 1% achievement levels maintained consistently

• 2-3x accelerated growth rates

• 100% student success rates

• Excellence across all core academic domains

...represents performance that exceeds what's typically seen even in the most elite educational contexts worldwide, including top international schools, specialized gifted programs, and the highest-performing national education systems.

The data suggests GT School has developed educational methodologies that optimize learning for high-ability students in ways that surpass conventional educational approaches, potentially offering insights for broader educational innovation.

Let’s now move on to high school. Alpha relies heavily on standardized tests because, without them, nobody would believe these results. When I applied to Stanford, it was important that I had achieved a 1600 SAT score because it showed them that, while I might not have had ordinary teachers, I did have a head full of knowledge. Alpha High’s end of year SAT scores for 2025 came back, and they remain the highest in Texas (senior class average of 1535). The most impressive stat that I can bring out is that both the median and average SAT of the freshman class was 1410. Most high schools don’t even allow freshmen to take the SAT, but Alpha’s freshman class outscores most school’s juniors and seniors.

Two more call-outs highlighting that every girl can be a “math girl”:

• To one high school student, whose AlphaX project is to create and perform a musical on Broadway, congrats on the 790 math SAT score.

• And to another student, who’s competing to be Miss Teen USA, congrats on your 790 math score.

I’m not going to lie: even to me, these results sound unbelievable. So how is this possible?

Well, I asked AI using this prompt: “Alpha is using an AI-powered tutor that is based on the principles of learning science and cognitive load theory. It is mastery based and creates personalized lesson plans for each student. These students only spend 2hrs/day to achieve these outstanding results. Is that possible? Provide academic backing for this.”

Here’s what Claude Opus had to say:

“The data from Alpha District, if achieved through 2 hours of daily AI-powered instruction, represents a potential paradigm shift in education. Here's the academic backing for why this is not only possible but theoretically predicted by learning science:

Theoretical Foundation

1. Bloom's 2 Sigma Problem (1984) Benjamin Bloom demonstrated that students receiving one-on-one tutoring with mastery learning performed two standard deviations better than conventional classroom instruction. Alpha's 99th percentile results align perfectly with this finding. AI tutoring can theoretically provide this one-on-one experience at scale.

Key Research: "The 2 Sigma Problem: The Search for Methods of Group Instruction as Effective as One-to-One Tutoring" - Bloom, Educational Researcher

Cognitive Load Theory Application

2. Optimized Cognitive Processing John Sweller's Cognitive Load Theory suggests learning is maximized when intrinsic load is matched to student capability and extraneous load is minimized

AI systems can achieve this by precisely calibrating difficulty to each student's zone of proximal development and eliminating irrelevant information.

Supporting Research:

Sweller et al. (2011) - "Cognitive Load Theory"

Kalyuga (2007) - "Expertise Reversal Effect"

Enough with the AI summaries, here is my in-depth narrative on how this is possible.

Every educator will say you need 2 things to educate a student:

1. A motivated kid.

2. Putting kids in lessons of the correct difficulty—not too easy, not too hard.

Regarding #1: Motivation is everything. Alpha's position is that motivation is 90% of the solution, and they build their entire school around that idea. In fact, the code name of the AI education platform is “Timeback” because giving kids "their time back" is the most powerful motivator available.

Regarding #2: Giving a student a lesson of the correct difficulty is a perfect job for AI. It can figure out what a student knows and doesn't know, then give that student the appropriate lessons.

So let's dive into the science of how a motivated student can learn 10x faster and better with Timeback.

For the last 200 years, schools have operated by putting a teacher in front of a classroom of 20-plus students. More recently, around 30-40 years ago, the Schools of Education at leading universities like Stanford, Northwestern, and Harvard started studying "learning science", which is the study of how children learn and develop. They quickly realized that children could learn 2 times, 5 times, even 10 times faster than they were currently learning, but it wouldn't be possible to achieve in a teacher-in-front-of-the-classroom model.

Bloom's 2 Sigma research from 1984 is the seminal paper, even though it does receive some criticism today. Bloom argued that the best intervention in education, with no close second, was 1-on-1 tutoring with mastery—which resulted in 2 standard deviations better performance. The problem has been that providing personalized tutors for every student was infeasible and cost-prohibitive. Until now.

This is where AI comes in. If you look back at the development of sciences like biology, chemistry, and physics, they all had a wilderness period, much like learning science has had over the last 40 years. Early on, doctors were still blood-letting and chemists were blowing themselves up. But each of these sciences took off with the invention of an instrument that dramatically improved the precision of measurement and allowed the scientific method to flourish. For biology it was the microscope, chemistry the analytical balance, and for physics it was the telescope. AI is that crucial instrument for learning science. AI allows for the precise instruction and measurement of each individual student that's impossible in a teacher-in-front-of-a-classroom model.

The key combination that Timeback implements from Bloom's research is both mastery and personalized lesson plans. No other educational platform implements mastery at a rigorous enough standard. While AI chatbots like ChatGPT can generate personalized lesson plans, they don't enforce mastery at a level that will deliver top 1% academic scores.

This isn't just Alpha who's going to be able to acknowledge and implement this. A 2024 Harvard study found that AI tutors performed better than Harvard teachers. The research community has known this was possible for 40 years. We just finally have the technology to make it happen.

Because of its dramatic effect on unlocking every kid's potential worldwide, I believe learning science will impact humanity more than any other science over the next decade.

---

If you want to dive deeper into this research, here are the best things to read:

Books:

• The Math Academy Way (if you read nothing else, read this. It is quite possibly the best collection of research on learning science and description of implementation.)

• "How Learning Happens" by Carl Hendrick

• "The Science of Learning" by Edward Watson and Bradley Busch

• "ABCs of How We Learn" by Daniel Schwartz, the Dean of the School of Education at Stanford

Substacks:

• The Learning Dispatch

• The Bell Ringer

• The Center for Educational Progress

Additional resources:

• This direct instruction BrainLift (you can read this newsletter of mine to learn about a BrainLift)

Academic papers and studies:

• Check out this list from Perplexity of seminal papers on learning science

• And this list of papers on cognitive load theory

• Special call-out: on memorizing multiplication tables (because schools have stopped teaching this important skill)

1. Caveats for my EDU PhD readers:

• This is not research data. Think of this as the data outflow of an engineering project, rather than center-of-the-page education research (e.g. a learning science study, or policy research).

• Student performance is the result of the Alpha system as a whole, not just the software platform; no one at Alpha thinks the software platform is a standalone intervention.

• The most natural comparison class for Alpha and its sister schools are other elite schools, almost none of which publish their K8 data.

2. The 3rd and 7th grade MAP tests are different. But they draw from the same question bank and are adaptive. A 3rd-grader getting a 260 will see many of the same questions that a 7th-grader at 260 will. The 260 score at different grade levels obviously means something different in terms of the student’s position relative to their peers. But it indicates the same level of absolute proficiency (in this case late algebra); MAP’s RIT scale is grade-independent.

---

Thanks for reading. Go crush the week! See y'all on Sunday.