The Ledger
Delving Bitcoin user posts STARK proof for Bitcoin’s UTXO set
A Delving Bitcoin user says a 1.6 MB STARK proof verifies Bitcoin’s UTXO set after block 972,500, with a browser check taking about half a second. The proof starts from a Bitcoin Core snapshot at block 935,000 and does not verify scripts or signatures.

A compact proof for the coin set
A STARK proof described in a Delving Bitcoin post covers the Bitcoin UTXO set after block 972,500 in a 1.6 MB file, with a browser verification time of about half a second, according to the post’s author. The post, published November 13, says the proof links a header chain from genesis to the tip with replayed blocks from 935,001 through 972,500 and the resulting coin set.
The author says the proof’s coin-set hash matches the value reported by Bitcoin Core’s gettxoutsetinfo and dumptxoutset commands at block 972,500. It also commits to a Merkle root over the set, enabling inclusion and non-inclusion proofs for individual outpoints; the author puts their size at about 1 KB. The demonstration page retrieves a roughly 2 MB segment of the set for each lookup.
What the proof covers
For blocks 935,001 to 972,500, the proof replays raw block data against Merkle roots and checks transaction identifiers, block and transaction structure, amounts, and each coin created or spent, the post says. Its header-chain portion checks SHA256d linkage, proof of work, difficulty retargeting, median time past, minimum version and exact chainwork. One stated exception is Bitcoin Core’s wall-clock time-too-new rule; the post also says a verifier must compare the chainwork against a trusted minimum, as in an SPV-style check.
The proof does not cover scripts or signatures. The author says the replay begins with Bitcoin Core’s assumeUTXO snapshot at block 935,000, whose starting set is checked against a hash pinned in Core’s chain parameters. Earlier blocks are represented by the header chain, not a replay of their coin activity, so the post describes the result as having the same trust as loading that Core snapshot rather than as a proof from genesis.
Lean circuits and remaining trust
The circuits are written in Lean 4, while a Zig prover interprets constraints exported from Lean, according to the post. The implementation uses Stwo, described there as circle STARKs over M31, along with a CUDA backend that produces byte-identical proofs. The author reports that a large block with 2^23 trace rows proves in 2.7 seconds on an H100; blocks are aggregated recursively by epoch before the chain, headers and set passes are joined into one proof.
The Lean work includes machine-checked proofs for header rules, SHA-256 matching FIPS 180-4, block checks, set transitions and the recursion that composes them, the author says. The specifications were also differentially fuzzed against Bitcoin Core’s C++ implementation across about 750,000 header-rule cases and 100,000 block and transaction cases, with zero mismatches. The post stresses that those comparisons are tests against Core’s code, not proofs of that code; it also lists the soundness of Stwo, FRI and LogUp, Fiat–Shamir assumptions, and some native verifier checks among the trusted components.
Snapshot verification and next step
As a practical test, the author says a snapshot dumped from a mainnet node running Core v31.1 at block 972,500 contained 164,539,066 coins and was 9.39 GB. The snapshot check passed in 72 seconds, using 0.5 GB of memory on an arm64 machine. The post also says there has been no external review of the proof system.
The author plans to try a Bitcoin Core patch that would accept a snapshot attested by a verified seal and skip background validation. That is a proposal, not a feature the post says is already available. The author estimates that proving block proofs for the whole chain would require roughly 325 H100-hours and says a proof from genesis would cost probably $1,000 to $1,500; the current demonstration instead starts from the snapshot at block 935,000.
Sources
- Original source Delving Bitcoin delvingbitcoin.org
Related stories
- Proposed Node Attestation Protocol would let Bitcoin holders sign statements weighted by coin value
- Delving Bitcoin post sets out options for recovering unmigrated coins after a quantum break
- Draft Multisig Session Protocol would let remote Bitcoin cosigners set up wallets without a coordinator
- Delving Bitcoin post raises hypothetical BIP54 timing attack
- Lightning Labs ships Pi extension for bitcoin-paid AI inference