Proteins nature never made.

1. The Search Problem

You cannot test your way there.

A protein is a chain of amino acids, twenty choices at every position. The number of possible chains gets out of hand immediately.

2. Design To A Spec

Start from the behavior, not from a sequence.

The old way is to take a protein nature already made and mutate it, one change at a time. That keeps you tethered to whatever evolution settled on. We write down the physical behavior we want and generate sequences built to it.

• Nothing is copied. These sequences appear in no organism, which is the point, and it keeps the intellectual property clean.
• Not one lucky hit. The output is a library, spread deliberately across different ways of solving the same problem.
• Buildable from the first step. A design that cannot be expressed is not a design. That constraint is applied up front, not discovered at the end.

3. From Thousands To A Panel

Why so few, and why these.

We never search the space above. We skip it, and generate straight into the small region that meets the specification. Everything after that is narrowing, and each step has a reason. Step through it.

Bar widths are illustrative. The first row is not a stage we filter through, it is the space we deliberately never enter, and no bar on a screen could represent its size.

4. What The Score Does

And what it does not.

Every design gets a score, worked out from the sequence alone before anything goes near a lab. It tells us whether a design can actually be made, and whether we already have a dozen others just like it.

That is why the panel is locked before any measurement comes back, so nobody can go back later and decide the model was right all along. Two parts of the pipeline are deliberately not published. Everything else is standard practice or open tooling, and ten provisional patent applications cover the methods and the designs themselves.

5. Why Now

None of this was buildable five years ago.

The underlying biology has been in the literature for years. What changed is everything around it. Every entry below is public.

1. 2012
   Fast motion, not hardness, is tied to stability. Published work shows that how much a dried glass rattles tracks how well it protects, more closely than the temperature at which it softens does.
2. 2017
   The tardigrade mechanism is pinned down. Their shield proteins are shown to be disordered, and to turn glassy as they dry rather than folding into a fixed shape.
3. 2020
   The cold chain becomes a front-page problem. A global vaccine rollout puts ultra-cold logistics, and the losses that come with it, in front of everyone at once.
4. 2023
   It is shown to work on a real drug. Tardigrade proteins are published stabilizing human Factor VIII, a fragile clotting therapy, through drying and rehydration.
5. 2023 to 2025
   The tooling catches up to disordered proteins. Generative methods mature enough to specify behavior instead of editing an existing sequence, and simulation built for shapeless proteins arrives. Standard structure tools had been blind to them, because they assume a fixed shape.
6. 2026
   Still Velocity is founded. Ten provisional patent applications filed within the first five months, covering the design methods and the sequences themselves.

6. The Short Version

Too many proteins to test. So we design to a spec, narrow honestly, and let the lab decide.

The engine is built and running. What it has not done yet is prove that its picks protect a medicine better than the alternatives, which is exactly what the first round of measurements is for.