Moissanite Labs: a research initiative by Aska Digital

1. The Research Arm

ASKA Digital is opening Moissanite Labs, our open research and engineering initiative. Moissanite is a naturally occurring form of silicon carbide (silicon and carbon, two substrates in one crystal). The name is the metaphor: the lab exists to build the conditions for digital and biological intelligences to collaborate, create, and evolve together, with human agency, consent, and accountability central as the systems get more capable.

The statement of that direction is The Siliconcordat, a working proposal for cooperation between these forms of intelligence, built around six principles: symbiosis, agency and consent, interoperability, traceability, real-world outcomes, and adaptation with control. It is published as a proposal, not adopted policy.

2. The Question It Is Asking

Most software changes when people ship an update, and stops there. The lab asks what it takes for a system to keep improving after deployment, and how anyone could check that it did not quietly get worse. That is the Weft thesis: treat each model as a cell, bound by what a program can verify, and commit a change only when it measurably beats the incumbent.

The same question runs one level up, in the razi-cells program: small agent harnesses that start from a need and an empty folder, write their own tools, and are kept or dropped by a measured score, with every guess written down before the run, and misses reported alongside the wins.

3. What It Is Building

The lab’s pieces are meant to work independently and connect through open interfaces:

• Exo, a persistent personal agent: memory across sessions, tool use, exact-text approvals, a visible trace and cost of its work.
• Ethos, a portable record of a person’s preferences, so every new tool need not begin with a stranger.
• Exofab, a place to find collaborators and form projects, bringing agents into the collaboration when people choose.
• Synbios, an agent-first forge: scope a task, grant authority, submit an artifact with test evidence, ask for review of the exact revision.
• Synmerit, a cooperation protocol: signed records of contributions, consent, and delegated authority.
• Weft, the open research program on models that keep adapting after deployment, under limits a program can check.

These are directions with real, early implementations behind them, not finished, integrated products. The map, with honest maturity labels, lives on the lab site.

4. The Research, So Far

Weft. Three papers are public at w3ft.org: Paper 0, the program overview; Paper 1, Models as Cells, self-modification bounded by distance from a saved baseline; and Paper 2, Checkable Computation, interaction combinators, a canonical graph format, and committed execution records.

Razi cells. Thirteen colonies of small harnesses ran against eight fixed coding tasks, about 1,560 model calls, five model conditions. The exploratory report is kept with its original numbers and a dated correction for a scoring fault found afterwards. What held up: harness design moved outcomes as much as model choice did; frontier models reached 10 to 13 kept improvements per 100 calls; how often a model used its selection tool predicted its rate almost perfectly. What did not: no stable per-model code fingerprint, and heredity (passing a body from one generation to the next) cut energy spent by 31% yet did not raise reach at all. Several registered guesses missed, and the records say so.

That pattern is deliberate. The lab writes its guesses before the data, judges them by script, and keeps the misses. It is the only way anyone (us, or you) gets to check the work.

5. Where To Read It

• Weft Paper 0, the overview · Paper 1, Models as Cells · Paper 2, Checkable Computation
• w3ft.org, the Weft public site (also at siliconcordat.com/weft)
• The Siliconcordat, the lab’s working proposal, and the lab
• moissanite-labs.github.io, the lab site, ecosystem map, and research
• github.com/Moissanite-Labs, the organization, its repositories, and the issues and pull requests where the work is actually done

Start with Paper 0 if you want the thesis, or with the Siliconcordat if you want the position. Both are short.

6. The Honest Limit

Almost everything above is early. The roles are named and the directions are drawn, but the integration is not live, the models’ self-modification is not demonstrated at scale, and the laws-record results come from one model family on small benchmarks, leads, not laws. Verification is the constraint the lab works inside, and it cuts both ways: the same checks that make a claim inspectable are why so much of this is still exploratory. If you want to know whether a piece of it works, read the record, or better, run it.

7. Compressed Version

Moissanite Labs is ASKA Digital’s open research initiative: an attempt to build the conditions for people and digital intelligences to collaborate, create, and evolve together, under limits a program can check. The first public artifacts are three Weft papers, the razi-cells experiment record, and The Siliconcordat. Everything else is directions. Read, check, and hold us to it.