Approach
We build for organic reality.
Organic reality: everything that grows, divides, metabolizes, and evolves. It is most of what we eat, how we heal, and what we will build with next. Working with it is still slow, so we build the companies that make it faster, and hold them to the same standards.
01
Grow
Cells take in what is around them and make more of themselves. A colony, a tissue, a harvest.
02
Divide
One becomes two, and the two are not quite the same. Every lineage starts here.
03
Metabolize
Sugar in, energy and parts out. The chemistry that decides what a cell can become, and what it makes.
04
Evolve
Over generations, what works is kept. Organic reality is never finished, and neither are our models of it.
Section 02 · How we build
One loop, every lineage.
- 01
Find a hard problem in organic reality
Something that grows, divides, metabolizes or evolves, where trial and error is still slow and expensive.
- 02
Build it in-house
A small team inside Progenitor, on shared science and infrastructure, until it works.
- 03
Test it against real data
Data it never saw. Not a demo, not a benchmark we chose after the fact.
- 04
Label it checked or theory
Checked if it matched. Theory if it has not been tested, or did not match yet. The label ships with the result.
- 05 → 01
Spin it out, or keep it
A lineage becomes its own company, or stays a product inside Progenitor. Either way it keeps the standards.
Section 03 · Why build them together
Shared, so each lineage starts ahead.
- Science
- Models, validation sets and the habits of checking them are built once and inherited.
- Infrastructure
- Compute, data, records and reproducibility are shared, so a new lineage does not rebuild the plumbing.
- Standards
- Checked or theory, priced before it runs, reproducible by default. A lineage cannot opt out.
Section 04 · Principles, with examples
- 01
Checked or theory
Every claim we make is labeled. Checked means it matched real data it never saw. Everything else is theory, and says so.
For exampleDrylab labels yeast gene essentiality checked: it scored MCC 0.56 on 1,481 genes it never saw.[1] Its drug screens are labeled theory, on every result, until real data says otherwise.
- 02
Reality has the last word
Models, simulations, and products earn trust by meeting the real world. When they disagree, the world wins and the model changes.
For exampleDrylab is a triage tool. It ranks which wet-lab experiments to run first; it does not tell you to skip them.
- 03
Reproducible by default
Same inputs, same result, every time. If we cannot rerun it from the record, we do not report it.
For exampleRerun any Drylab result from its record alone (model checksum, seed, engine versions) and it comes out byte for byte identical: 0 bytes differ.[2]
- 04
Start with one cell
Begin small, prove it, then grow it into its own lineage. Breadth comes from many small things that work.
For exampleDrylab is built around one reference cell, baker’s yeast S288C, modeled at all five tiers. It is also where its checked result comes from.
Section 05 · Honest limits
What we do not claim.
- We do not replace the wet lab. Software triages experiments; the bench decides.
- Nothing we make is for clinical use unless it says so, and today nothing does.
- Checked means checked on a stated test. It does not mean right about everything nearby.
- One lineage is live today, in private preview. The others are in development, and we will not name them until they are real.
- We do not publish a number we cannot source. If it is on this site, it has a footnote or a figure credit.