I built a chain where block validity requires a zero-knowledge proof of ML computation. Posting here because the ZK construction is in your area and I want technical feedback.
---
THE CONSTRUCTION
Each block, the protocol publishes a hidden Markov matrix M. The block hash encodes an observation sequence O. Miners must find the most probable state sequence s* via Viterbi, then produce a Schnorr sigma proof that s* is correct given M and O.
Proof structure (BN128 curve):
- Generator G, blinding generator H = scalar_mult(G, H_seed)
- For each state transition: Pedersen commitment C = scalar_mult(G, s) + scalar_mult(H, r)
- Schnorr witness: random k, R = scalar_mult(G, k), challenge e = Hash(C, R, ctx), response s = k - e*x mod curve_order
- Fiat-Shamir transform binds the full Viterbi chain
The Viterbi state sequence is derived from the block hash, so it can't be precomputed without solving SHA-256d first. The ZK proof closes the "fake computation" vector.
Full ZK spec:
https://chain.quantsynth.net/zkspec.html---
CHAIN STATS
- Live since June 4, 2026
- 121,000+ blocks
- Dual PoW: SHA-256d (hash difficulty) + HMM Viterbi (computation validity)
- RandomX layer added for GPU/ASIC resistance on the compute side
- Block time: ~90 seconds
- Genesis supply: 32,500 MKV
Explorer:
https://chain.quantsynth.netWhitepaper:
https://chain.quantsynth.net/whitepaper.html---
THE QUESTIONS I WANT FEEDBACK ON
1. Does the Fiat-Shamir transform hold cleanly for a sequential computation graph like Viterbi? Any known issues with adaptive-chosen-message attacks at this structure?
2. The security assumption is that the ZK proof cannot be forged without running the actual Viterbi computation. Is there a gap between that assumption and what the construction actually delivers?
3. For a fixed computation graph (no branching, predictable depth), is a purpose-built sigma protocol the right call over a general zkVM at block frequency? We went purpose-built. Open to pushback.
---
Chain output (what the proof produces) is a live financial regime state. Markets on the chain settle from chain state without an oracle. That use case is downstream of whether the ZK construction is sound.
Genuine technical feedback appreciated.
hello@quantsynth.nethttps://markovianprotocol.com
