Research that belongs to everyone

Published in full, open by default

Nothing we fund disappears into a drawer.

We pre-register our methods and publish our results in full, including the ones that don't go the way we hoped. Benchmarks, evaluations, curriculum, and de-identified data are released under open licenses for anyone to read, reuse, and build on.

Open science isn't a compliance step for us. It's the point. Findings only compound when other people can pick them up and carry them further.

Artifacts people can use

A paper is a start, not the finish. We invest in the durable outputs around the research, like curriculum, assessment items, benchmarks, evaluations, tooling, datasets, and reference implementations, and open-source them so researchers and industry alike can put them to work immediately.

If a study we fund produces something useful to the field, we would rather it live in the open than sit behind a login.

Research we expect to complete

Over the course of the study, here's what we hope to put into the world:

• → A rigorous, pre-registered estimate of what sustained 1:1 tutoring does for grades 3–4 in math and ELA, measured against a classroom control.
• → Open curricula and assessment banks, refined across a full year of real instruction.
• → Learning progressions that map how each skill builds on the ones before it.
• → Benchmarks for what genuine proficiency looks like at each grade.
• → A cost-effectiveness analysis: what these outcomes take, and how cheaply they might be reproduced.
• → A public dataset of de-identified learning trajectories for others to study.

These are the goals we're aiming at. We'll report honestly on the ones we reach and the ones we don't.

Better together

When two groups are circling the same question, we would rather connect them than watch them duplicate the work. Where it helps, we actively encourage collaboration across labs and institutions: shared data, shared instruments, shared credit.

The goal is to answer the question well, together, not to win it alone.

A note on overhead

One small formality: on our grants, institutional overhead is set at 20% of direct costs. It keeps as much of each dollar as possible pointed at the work itself.

What data we provide

The core dataset follows individual students through a full year of 1:1 instruction. Because we measure the same children repeatedly over time, you can study not just how far they got, but how they got there.

Everything is de-identified and released under open licenses. Here is what each part is:

• → Assessment responses. Every question a student answered, item by item, not just their scores. You can see exactly what they got right, what they missed, and how that shifted over the year.
• → Learning trajectories. A timeline for each student: which skills they worked on, when they reached mastery, and where they stalled. Built for modeling how learning actually unfolds.
• → Curriculum and materials. The real lessons, in the order they were taught, so any result can be traced back to the instruction that produced it.
• → Instructional logs. A record of each tutoring session: what was covered, which techniques the tutor used, and what appeared to move the student forward.
• → Decision logs. After each session, what the tutor noticed and how they designed the next day's lesson in response: the data they checked, the research they consulted, and the prep they did to build tomorrow's plan.
• → Benchmarks and evaluations. The instruments we used to measure progress, the rubrics that scored them, and the results, so others can calibrate against the same yardstick.
• → Documentation. Codebooks, field definitions, and file formats, so the data is usable on day one without reverse-engineering it.

Student privacy comes first. Before any release, an independent research organization such as Mathematica verifies that every dataset is properly de-identified. Nothing that could identify a child is ever made public.

Have ideas for this data?

If you can see a use for what we're building, whether a paper, a product, or a follow-on study, email ben@recess.gg with a one-page proposal. We'd love to work together.