https://github.com/name970/Protocol
A Proof-of-Useful-Work consensus where the "work" is verifiable matrix multiplication: exact INT8 arithmetic for bit-reproducible results, plus a cheap probabilistic audit that certifies a product without recomputing it.
https://github.com/name970/Protocol
consensus cryptography rust verifiable-computation zero-knowledge
Last synced: 1 day ago
JSON representation
A Proof-of-Useful-Work consensus where the "work" is verifiable matrix multiplication: exact INT8 arithmetic for bit-reproducible results, plus a cheap probabilistic audit that certifies a product without recomputing it.
- Host: GitHub
- URL: https://github.com/name970/Protocol
- Owner: name970
- License: mit
- Created: 2026-06-30T11:42:04.000Z (about 1 month ago)
- Default Branch: main
- Last Pushed: 2026-07-13T08:41:33.000Z (19 days ago)
- Last Synced: 2026-07-13T10:20:25.068Z (19 days ago)
- Topics: consensus, cryptography, rust, verifiable-computation, zero-knowledge
- Language: Rust
- Homepage:
- Size: 1.11 MB
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 5
-
Metadata Files:
- Readme: README.md
- Contributing: CONTRIBUTING.md
- License: LICENSE
- Code of conduct: CODE_OF_CONDUCT.md
Awesome Lists containing this project
- awesome-cryptography-rust - Protocol - scheme decomposition) and a Freivalds-style audit, as the compute layer of a Proof-of-Useful-Work consensus. (Cryptography / Zero Knowledge Proofs)
README
# Protocol
> A Proof-of-Useful-Work consensus where the "work" is verifiable matrix multiplication:
> exact INT8 arithmetic for bit-reproducible results, plus a cheap probabilistic audit that
> certifies a product without recomputing it.
> [!WARNING]
> **Status: research / proof-of-concept. Unaudited. Not for production use.**
---
## Why this is interesting
Quantizing to INT8 is the canonical *lossy* step in machine learning. This project uses the
same INT8 path the other way around: it reconstructs a **bit-exact** floating-point matrix
product from bounded integer slices. Exactness is the point — a result reproducible down to
the bit is the precondition for *verifying* a computation cheaply (a Freivalds-style check,
`O(n²)`) instead of recomputing it (`O(n³)`). On top of the exact product sits a commit–reveal
scheme, so a prover can generate its own challenges non-interactively without being able to
grind a passing forgery.
The decomposition technique is **not novel**: it builds on the Ozaki scheme and the line of
work putting it on integer / INT8 tensor cores. The contribution here is using it as the
compute layer of a verifiable-work consensus, with the audit and soundness analysis that requires.
## What is open
- Exact `INT8 → INT32` partial products with `FP64` reconstruction — implemented and covered by tests (bit-exact against a direct reference).
- Non-interactive commit–reveal challenge derivation — designed, not yet implemented.
- Post-quantum time-seal (delay function) — open design question.
- Data-availability layer (FRI) and private transaction layer — specified, not implemented.
- GPU kernels — current code is a CPU reference.
- **Formal security proofs.** Soundness of individual pieces is being discussed in the open
(see [Cryptography, honestly](#cryptography-honestly)); none of it is audited.
## Quick start
```bash
git clone https://github.com/name970/Protocol
cd Protocol
cargo test # exactness + forgery-rejection tests
cargo run --example demo # bit-exact INT8 matmul vs an f64 reference (adjust to your layout)
```
## Architecture
A short map from whitepaper sections to code lives in [ARCHITECTURE.md](ARCHITECTURE.md).
## Contributing
Contributors are very welcome — especially people who want to write the code.
- New here? Start with the **`good first issue`** label: self-contained tasks that touch the
numerics and need no protocol-wide knowledge.
- Read [CONTRIBUTING.md](CONTRIBUTING.md) for dev setup and PR conventions.
- Questions and design discussion go in [Discussions]( #community), not issues.
Every issue references the relevant whitepaper section via `ARCHITECTURE.md`, so you can see
where a task sits in the whole before picking it up.
## Community
- **Discussions:**
- **Discord:** https://discord.gg/aVwCK8WEYd
## Cryptography, honestly
This is research, and the soundness arguments are being worked through in the open rather than
asserted. Public discussion so far:
- Hash-based nullifier without an elliptic-curve key tree''
- Commit-reveal hash binding sufficient to prevent adaptive forgeries in a Stochastic Matrix Audit
If you can break something, open an issue — adversarial review is the most useful contribution.
## A note on the token
The broader protocol design includes a token. **It is not the point of this repository.**
This repo is about the technology; issues and discussion here are technical.
## License
MIT LICENSE