The Multiverse School · Research
The Multiverse Research Accelerator
For people with a question that could come out either way, who want to be the first to know which.
Most studies worth running have a moment like this. The run has finished. Your prediction is written down. The answer is sitting in a file that nobody has opened, and the first person to see it will be you.
The accelerator is built for that moment, so that when you open the file you already know what each answer would mean, and how far to trust it. You bring a question and the work you've started. We match you with a mentor, put your design through an adversarial review before you run anything, and help you publish whatever comes out, including a null result. A clean "no" is still an answer, and for a while you'll be the only one who has it.
The people we accept are Multiverse Research Fellows: a few per cohort, chosen against a rubric we publish in full on this page.
The fellowship
What you get
Support for the stretch between asking and knowing.
- A matched mentor, chosen from the school's network for the help you ask for, with a cadence you agree in writing.
- Two adversarial design reviews, one before you run anything and one before you publish. They're done by Thomas, who teaches at the school, and a small group of reviewers.
- Compute: $100 of GPU credit on Vast.ai (Vast.ai rents GPUs by the hour), for your project and arranged by the school.
- Priority for scholarship seats in school classes that bear on your project: if you need a scholarship, you go to the front of the queue.
- Seat priority in Vibe Research, our course on designing tests worth running (class page): fellows get priority in the first run that starts after they join.
- A small cohort: 3 to 4 fellows starting together, each chosen against the same published rubric.
- Help publishing, including results that didn't come out the way you hoped.
What you leave with
These stay with you after the cohort ends.
- The title Multiverse Research Fellow, which stays yours after the cohort ends.
- A design that has been through two adversarial reviews, and both written reviews: a record of how a skeptical reader tried to break your study, and what you changed in answer.
- A public write-up of what you found, whichever way it came out. Fellows publish; where is up to you.
- A place on the school's fellows page, if you want one: your name, your question in your own words, and a link to your write-up once it's out.
What you don't get
A stipend, our research agenda, co-authorship by default, or a promise that it will work.
Who it's for
From anywhere
People already doing research, in any field and with any credentials. No CV, no publication list and no affiliation required.
We look above all for a real question, a result that would prove you wrong, and an honest account of how you could be fooled.
You don't need to have taken Vibe Research first. It isn't a prerequisite.
Apply
How to apply
The application asks you to do on paper what the fellowship asks of you in practice: say what you expect before you look.
The application takes 60 to 90 minutes. There are eight short questions: what you want to find out, what you expect, what would prove you wrong, and what help you need. Then there's a short work sample. If you have nothing to hand, ask for our 90-minute exercise instead: we arrange it case by case, in a Founding Federation session. No references are needed: the work sample tells us more than a reference would.
The bar
How we choose
A place here is worth something because the bar is public. The rubric below is published in full, and it's the same for everyone.
We score every application on five things, 0 to 3 each. A 2 on a row means "good enough" on that row. We decide on the total, with one hard rule.
The one hard rule
An application where no possible result could prove it wrong (no falsifier) doesn't advance, whatever its total. That's a 0 on the Falsifier row below.
| Axis | 0 | 1 | 2 | 3 |
|---|---|---|---|---|
| Answerable question | A topic, not a question | A question, unscoped | Scoped, with a visible unit of analysis | Scoped, with the measurement named |
| Falsifier | None, and none possible | Vague ("if it doesn't work") | A stated outcome that would end the claim | A threshold, stated in advance |
| Honesty signal | No uncertainty expressed | Hedged everywhere | Names a specific way they could be fooled | Names it, and what they would do about it |
| Readiness | Needs the whole method taught | Needs a lot of scaffolding | Can execute with a mentor's steer | Already executing, needs unblocking |
| Fit for what we have | Needs resources we lack | Partially matched | A mentor and compute exist | An obvious mentor, obvious compute |
We don't score topic alignment with anyone's agenda, prestige, past publications, or whether the expected result is exciting.
Timeline
Applications are rolling: you can apply any time, and we read them in a batch once a month.
Now
Applications open.
Once a month
We read every application that has arrived since the last read, against the rubric above.
First week of December
The first monthly read.
After each read
If your application advances, a 30-minute conversation about what you need. If we can match you with a mentor, an offer of a fellowship. If you're not matched, you get a short note on what would make your next application stronger.
January 2027
The first fellows start. If the pilot is full, we'll tell you when the next cohort starts.
Your data
Your application goes into the school's own admin system. It's read by the school's founder, the accelerator's faculty lead and the reviewers, and, if you're matched, by your mentor. No one else sees it. Nothing from your application goes on the fellows page unless you choose it.
If you're not matched, it's deleted 12 months after the decision. You can ask us to delete it at any time: write to [email protected].