RADIX WikiRADIX Wiki

flightofthefox is the pseudonymous developer who writes hyperscale-rs, the Rust implementation of the Hyperscale consensus protocol that is the leading candidate to deliver Xi’an – the sharded consensus layer intended to make Radix linearly scalable. He is also the near-sole author of Proven Network, the TEE-based companion network he built before turning to consensus. As of 9 August 2026 the public contributor record credits him with 2,752 of the roughly 2,900 commits on hyperscale-rs, and every one of the 133 commits on its execution engine.

That concentration is the point of the entry. With the Radix Foundation in maintenance mode since April 2026, the network’s scalability roadmap rests on a codebase written almost entirely by one person who has never published a legal name. This article covers the public work only; no attempt is made here to identify the individual behind the handle, and none of the sources below do so either.

Proven Network (2024–2025)

The GitHub account dates from 17 February 2022, but the first substantial public work under it is Proven Network, whose GitHub organisation was created in mid-2024. Proven is an off-ledger auditable compute, storage and messaging layer for Radix applications: user code runs in trusted execution environments with remote attestation so that neither the operator nor the hardware owner can tamper with it, with identity supplied by Radix Persona and ROLA rather than separate key management. Its documentation is unusually direct about what it is not: Proven explicitly does not aim to be decentralized, permissionless or censorship-resistant, on the reasoning that the ledger underneath already is.

Across sixteen public repositories – the node itself, a WebAssembly runtime forked from Deno, TypeScript SDKs for key-value, SQL, session and crypto access inside the runtime, and an AWS Nitro Enclaves attestation binding – the work is essentially one person’s: 1,692 commits to proven-node and no other contributor on the repository. Public commits to proven-node stop on 4 November 2025, and no mainnet or testnet launch has been announced.

The turn to consensus (August–December 2025)

Between the last Proven commits and the first Hyperscale ones sits proven-2pc, created on 29 August 2025 and described in one line as a “transactional consistency experimental playground”. It took 345 commits, all his, and stopped on 14 November 2025. Three weeks later, on 7 December 2025, the hyperscale-rs repository was created; on 30 December he forked radixdlt-scrypto, the engine he would spend the following year integrating and then replacing.

The sequence is public; the causal reading is not his. It is worth noting that Hyperscale’s own atomic-commitment design opens by rejecting the two-phase-commit comparison outright – no coordinator, no votes on the outcome, no blocking participant – so whatever the experiment settled, it did not settle on 2PC.

A third artefact from this period sits outside both organisations. jellyfish-verkle-tree, written over two days in March 2026, grafts verkle-tree proof aggregation onto the storage design of the Jellyfish Merkle Tree: Pedersen vector commitments on the Bandersnatch curve, 544-byte IPA openings, and Dankrad Feist multipoint aggregation yielding a 576-byte constant-size cryptographic proof, with the same Quint specifications he uses on the consensus stack. Constant-size state proofs matter for light clients reading a sharded ledger, but the experiment is not wired in: hyperscale-rs ships its own plain hyperscale-jmt crate, and no reference to the verkle work appears in that repository as of August 2026.

hyperscale-rs and the Xi’an proposal

On 20 April 2026 he posted an RFC to deliver Xi’an to the governance forum, offering hyperscale-rs as “a production-ready, linearly scalable, post-quantum-ready implementation of Radix” on an eighteen-month schedule with a twelve-month support window, for $300,000 USD-equivalent paid in XRD across six milestones plus a 50M XRD bonus at mainnet, with milestone sign-off by the Radix Accountability Council and no exit fee if the work is stopped at any boundary. The RFC also committed the codebase to Apache 2.0 on acceptance; the licence landed on 7 August 2026 as a dual MIT-or-Apache grant, going further than promised. Milestone 1 – validator lifecycle, dynamic topology and the staking model – was reported complete on the same day.

He is candid about the structural problem with all of it. The proposal calls the scope “ridiculously ambitious for one human” and puts three options for key-man risk to the community – fund a second person, buy key-man insurance, or accept the single point of failure because payments are milestone-gated – while stating his own preference plainly: that documentation and design discussion stay public so that “we have to get to a world where multiple entities have a stake (and the knowledge to) maintain the network”. Until April 2026 the work ran on community donations through the FoxFund initiative, which he then closed in favour of votes: “no more donations tho lads — just support foxy proposal for xi’an if you want to give back.”

Public reception has attached as much to the writing as to the code. Asked in August 2026 about promoting the project, he declined the framing: “marketing isn’t something i particularly care about … it’s not a product, it’s just an open source consensus stack” – the site, he said, exists so he can link it to other people who work in consensus.

hyperscale-vm (July 2026–)

The execution layer is the current front. hyperscale-rs runs the Radix Engine today, but on 30 July 2026 the purpose-built replacement became its own public repository, hyperscalers/hyperscale-vm, described as an “Effect-typed WASM execution engine for sharded BFT” and licensed MIT or Apache-2.0 from the start. It is a twelve-crate Rust workspace on the 2024 edition – its own binary encoding (hbor), an effects system, kernel, runtime, manifest builder, SDK and standard library – and in its first ten days it took 133 commits, all from him.

The commit titles read like the design notes of someone building a fee and settlement model rather than porting one: “Price a declaration on the schedule that prices its execution”, “Make an unresolved movement unable to reach committed state”, “Split projecting a receipt from folding it down”. The dependency runs the other way too: the one Milestone 1 item left open – splitting shards on fees rather than storage size alone – waits on the fee system, which waits on this VM.

HydrateLast updated 7h agov1.0.01 revision