Open down to the last decision
Not just the weights or the output. The prompt, the seed, the tool call, the branch that went nowhere.
Point it at the unknown.
An open source research harness. It proposes the experiment, runs it, and tries to break its own result.
The whole researcher,
not just the model.
The model is the part that thinks. The harness is everything else: the part that holds the question, designs the experiment that could kill it, runs the code, reads what came back, and decides what to ask next.
That machinery is where research actually happens. Warpy is that machinery, in the open, where anyone can run it, read every decision it made, and point it somewhere nobody has looked.
The same loop a scientist runs.
How it is built.
No model lock-in
Whatever is best this month will not be best next year. The harness is the part that persists, so the model underneath can change without taking the work with it.
Every run counts
A hypothesis that held and one that failed are both results. Both are worth keeping, and publishing either makes the next run better for everyone.
Runs where the data is
A laptop, a cluster, or a machine that never goes online. Bring the harness to the data.
Legible to a human
Someone has to be able to read back why it believed what it believed, and disagree.
First light on biology.
The harness is general and stays that way. The first thing it gets pointed at will not be.
Aging is the largest unsolved problem in medicine, and the one that quietly decides how much time everyone gets to work on all the others. It is also exactly the shape of problem this is built for: enormous literature, enormous data, and far more testable ideas than there are people to test them.
Everything it learns is aimed at one question: what breaks down in a living system, and whether it can be fixed.
Built in the open.
The repository is where the work happens. The Discord is where researchers who think problems like aging are solvable in our lifetime find each other.
