Ragged multi-instance GKR for Poseidon2b: one walk, unequal regions, no max-width padding

An ethresear.ch post presents a ragged multi-instance GKR construction that batches Poseidon2b permutation executions across nine committed regions of unequal Boolean widths (12–17) into a single GKR walk, using an implicit selector embedding instead of max-width padding that would expand physical rows from 360,448 to 1,179,648. Benchmarks on an Intel Core i7-1365U show the ragged walk with a 10.830s prover median versus 17.609s for nine independent native walks (1.626x faster), 6.108x faster verification (0.802s vs 4.896s), and a 6.256x smaller transcript (363,264 B vs 2,272,512 B). The author notes Ethereum relevance to hash-heavy proving workloads such as post-quantum validator aggregation and WHIR-based stacks, where Poseidon2 Merkle hashing was reported at 58% of GPU time, and invites comparison with batching strategies in leanVM and AIR layouts.