Solving the oracle issue in prediction markets at the protocol level

I created a blockchain where a zero-knowledge proof of machine learning computation is required for block validity. Just sharing this here because the ZK aspect is relevant to your interests and I’d love some technical input. THE CONSTRUCTION In each block, the protocol reveals a hidden Markov matrix M. The hash of the block encodes an observation sequence O. Miners are tasked with figuring out the most likely state sequence s* using Viterbi, and then they need to generate a Schnorr sigma proof that verifies s* is correct based on M and O. Proof structure (using the BN128 curve): - Generator G, blinding generator H = scalar_mult(G, H_seed) - For every state transition: Pedersen commitment C = scalar_mult(G, s) + scalar_mult(H, r) - Schnorr witness: pick a random k, R = scalar_mult(G, k), challenge e = Hash(C, R, ctx), response s = k e*x mod curve_order - The Fiat-Shamir transform ties together the complete Viterbi chain. Since the Viterbi state sequence comes from the block hash, it can’t be pre-computed without tackling SHA-256d first. The ZK proof eliminates the chance of "fake computation". Full ZK spec: https://chain.quantsynth.net/zkspec.html CHAIN STATS - Active since June 4, 2026 - Over 121,000 blocks - Dual PoW: SHA-256d for hash difficulty plus HMM Viterbi for computation validity - RandomX layer added to resist GPU/ASIC on the compute side - Block time: about 90 seconds - Initial supply: 32,500 MKV Explorer: https://chain.quantsynth.net Whitepaper: https://chain.quantsynth.net/whitepaper.html THE QUESTIONS I’D LIKE SOME INPUT ON 1. Does the Fiat-Shamir transform apply neatly for a sequential computation graph like Viterbi? Are there any known problems with adaptive-chosen-message attacks in this setup? 2. T