Why Spinel?
Spinel is a gemstone and a family of crystal structures. The name fits the project in two ways: it is a real material with a strong visual identity, and its lattice structure is an unusually good metaphor for a protocol built from simple, exact local rules that must produce one global state.
Historically, spinel was often mistaken for ruby. Some famous “rubies” turned out to be spinels. That hidden-gem quality fits a research project looking for overlooked structure in the relationship between AI compute and proof-of-work.
The thesis
The most important arithmetic in modern computing is increasingly low precision. AI accelerators are built around tensor operations in FP8, FP4, and related microscaled formats. Proof-of-work, meanwhile, requires scarce computation whose result can be independently validated.
Spinel asks whether those two worlds can share a primitive without compromising either side.
How we work
- Define semantics before optimizing kernels.
- Keep an independent CPU oracle.
- Measure real hardware rather than marketing throughput.
- Preserve negative results and architecture limits.
- Require cross-vendor evidence before claiming portability.
- Publish test vectors before asking outside miners to trust an implementation.
Current scope
Spinel is currently in Research R0. The public scope is deliberately limited to the technical program: FP8 arithmetic, MXFP4, transcript construction, cheap verification, native mining kernels, and a future devnet.
There is no public sale, no production token, and no mainnet. Those are downstream consequences, not starting assumptions.