Every number on this page is measured against the live repo

The Build Dashboard

A real STARK-verifiable virtual machine: Rust + Winterfell, an HTTP + MCP proving service, and Foundry contracts with a documented on-chain trust model.

37
Rust Tests
8
Solidity Tests
~11 KB
Proof Size
~5.5ms
Verify Time

✓ measured 2026-08-31 · cargo test --release --workspace + forge test · examples/fibonacci_like.zkasm, 10-run avg, release build

# One-line pitch
"A small, real zkVM: a custom ISA with branching and registers, a from-scratch AIR,
a STARK prover/verifier over Winterfell, an HTTP + MCP proving service, and
on-chain attestation. Every gap is named in the docs, not hidden."
Current State — Aug 31, 2026

Built and Verified

ISA

zkvm-isa

.zkasm ISA: ADD/SUB/MUL, JZ/JNZ branching (forward-only), a 4-register file (LOAD/STORE), label-based assembly

✓ 16 tests passing
STARK

zkvm-stark

STARK prover/verifier over Winterfell, no-secret-witness AIR — sound without a lookup/permutation gadget

✓ 8 tests passing
API

zkvm-host-server

axum 0.8 HTTP server, ProverBackend trait + ProverRouter, hardened with body/timeout/concurrency limits

✓ 13 tests passing
CLI

zkvm-cli

run, prove, verify, deploy, demo subcommands

✓ Complete
MCP

MCP Server

rmcp 3.1, port 4478, prove + verify tools, verified against a real client handshake

✓ Complete
⛓️

Smart Contracts

Foundry: ProofOrchestrator, AttestedVerifier, UnimplementedStarkVerifier, IProofVerifier

✓ 8 tests passing

End-to-End Scripts

Both Pass
scripts/onchain_demo.sh

Full loop: task → prove → verify → attest → pay (anvil)

scripts/mcp_demo.sh

Real MCP transcript capture

What's explicitly NOT here yet

Loops, addressable memory, RISC-V compatibility, formal verification, fuzzing, external audits, a second real proving backend, trustless on-chain verification (today's AttestedVerifier is a documented trust bridge, not a cryptographic verifier). See docs/ROADMAP.md for the full list.

Suggested Next Slice

In order of validation value

This is the actual order in docs/ROADMAP.md — not a fixed calendar. Each step only starts once the previous one is solid.

Next Not started

Loops

  • → A dynamically-sized execution trace (bounded, padded/truncated by actual run length) so JZ/JNZ can jump backward
  • → The "verifier re-executes and asserts everything" trick still works — the verifier can just run the loop itself
  • → Real design needed for trace-length selection and the resulting proof-size variability
After loops

Addressable memory

  • → Dynamic addresses (e.g. LD r0, [r1]), distinct from today's 4 static register slots
  • → Open question: does the no-secret-witness trick still hold, or is a real lookup/permutation argument needed?
Only after 1–2 are solid

RV32I compatibility

  • → Considered explicitly out of order — not worth attempting before loops and memory are real
Ideas Backlog — Not Built

100 Ideas. Zero Shipped.

A brainstorm of what a full platform could eventually include — none of this exists yet. Listed here so it doesn't get confused with the Status section above.

Non-Negotiable

Quality Constraints

The actual working discipline for this repo — see CONTRIBUTING.md.

1. Verify Every External API

Never assume. Read the crate source at the exact version tag, or run a minimal repro before writing production code.

2. No Stubs or Mocks

No todo!(), no return true stubs, no component named after a real product it doesn't implement. If something isn't built, it's in ROADMAP.md.

3. Honest Scope Docs

docs/ROADMAP.md and docs/HOST_SERVICE.md list what's real and what's not. Every gap is named with what it actually takes to close. No "coming soon."

4. Test Under --release

Winterfell's debug-mode degree check is over-strict for data-dependent constraint degrees (confirmed: not a soundness bug — see the code comment in zkvm-stark/src/lib.rs).

5. No Fabricated Specificity

If you cite a paper, a date, a benchmark, or a project name, it must be verifiable. Unverifiable numbers are not acceptable — see the "measured" tag on every metric on this page.

Scope Control

What NOT to Build (Yet)

✗

Billing / payments

No users, no auth persistence

✗

Edge / WASM SDK

No WASM backend exists

✗

SP1 integration

Investigated and declined — ISA mismatch, 518 deps for a hello-world. See docs/HOST_SERVICE.md.

✗

A token / tokenomics

No token contract exists

✗

Compliance / regulatory product

No legal team, no customers

✗

Decentralized prover marketplace

No prover supply to route to — see /depin for the design sketch, explicitly unbuilt