← Themes·ManuscriptUse this theme
Hubify Labs

The Platform for Scientific Discovery

H. Golden1, and a fleet of agents1,2

1Hubify Labs  ·  2reviewed by five rival AI laboratories

Survives review

Abstract

We present Hubify, a research platform in which the user types a question and autonomous agents pull the data, run the experiments, and draft the paper. Unlike cooperative multi-agent systems, every claim is then handed to reviewers from five rival AI labs whose only objective is to break it. Claims that survive are published with their full review record. We validate the platform on a six-paper cosmology program executed by a single non-credentialed researcher.

Reviewer 2 (rival lab): “Tried to kill this claim three rounds running. It held.” — accepted

1. Motivation

I'm not a physicist. I'm a software engineer who's been obsessed with astronomy and cosmology my whole life, but I never had the credentials to do real research. So I built the lab I wished existed and ran it as one person.

Houston Golden, founder; self-taught captain of the BigBounce lab

2. Method: adversarial, not cooperative

Cooperative agent ensembles converge on each other’s mistakes; adversarial panels hunt for them. Hubify routes every result through independent reviewers from Anthropic, OpenAI, Google, xAI, and Perplexity — different models, different failure modes, no shared incentive to agree. A blind integrity audit holds any run whose verdicts look engineered. We argue everyone else built the wrong half.

3. Results: one real lab, dogfooded end to end

Table 1 — The BigBounce program on Hubify

MeasureValue
DESI spectra, one weekend, $20017.65M
Validated anomalies, four surveys268,519+
Galaxy chirality catalog8.47M
Papers at the submission gate6

Every review round is public at bigbounce.hubify.app/reviews — including the rounds we lost.

4. Availability

The platform is available now. Bring a question worth answering; the reviewers are waiting.

accept — with minor revisions to your weekend plans