Notes from building a small, real zkVM

Each post below is grounded in the actual repo — real code, real measurements, real test names. Where a number appears, it was measured before it was published, not estimated.

• AIR design

  Why This zkVM's Soundness Doesn't Need a Lookup Argument

  How branch outcomes and register accesses are asserted directly instead of derived algebraically — and exactly where that stops working.

• AIR design

  A Register File Without an Algebraic Mux

  How a one-hot selector proves which register a LOAD/STORE touched, and the negative test that checks lying about it fails.

• Investigation

  We Investigated Adding an SP1 Backend. Here's Why We Didn't.

  518 dependencies, a Go FFI requirement, and a fundamental ISA mismatch — the real numbers behind a real "no."

• Winterfell

  A Winterfell Debug-Mode Assertion That Isn't a Soundness Bug

  The exact expected-vs-actual degree numbers from a real reproduction, and why release mode is correct here.

• Scoping

  Why Forward-Only Jumps, and What Loops Would Actually Require

  The specific reason JZ/JNZ can't jump backward yet, and what changes when trace length depends on runtime behavior.

• On-chain

  An Honest On-Chain Verifier: What AttestedVerifier Actually Guarantees

  Why the contract is named AttestedVerifier and not StarkVerifier, and what a real one would require instead.

• Security

  Hardening a STARK Proving Service: Real Benchmarks, Not Guesses

  An initial instruction-count limit turned out to be wildly wrong. Here's the measurement that fixed it.

• API design

  A Code Review Caught a Real API Footgun. Here's the Fix.

  A function that took a program and silently ignored it — verified, fixed, and confirmed through real CI.

• MCP

  Connecting Claude to a Real STARK Prover via MCP

  Two real MCP tools, verified against an actual client handshake — and why they run on their own port.

• Benchmarks

  What Sub-10ms STARK Verification Actually Requires

  The precise methodology behind a benchmark claim, and the trace size where it stops holding.

• Positioning

  What zkvm.host Actually Is vs. the Open-Source Repo It's Built On

  A line-by-line breakdown of what's built, what's architecturally real but single-backend, and what's pure pitch.

• Security

  The Trust Boundary, Written Down: What's Trustless Here and What Isn't

  The proof is trustless STARK cryptography. The payment for it is not. Exactly where that line sits.

• CI / DevOps

  Proof as a CI Check: What zk-ci.yml Actually Verifies on Every PR

  The real workflow that boots a live prover and proves plus verifies every example program before a PR can merge.

• Contracts / Testing

  Testing an On-Chain Task/Reward Orchestrator with Foundry

  Eight tests, each named for the specific thing it refuses to let happen to a paid attestation flow.

• AIR design

  Reading a Winterfell AIR: One Degree-4 Constraint, Two Kinds of Check

  A line-by-line walk through the real transition and boundary constraints behind this VM's proof.