Radix Week in Review: Hyperscale Answers the Sharding Questions

The week in brief

• The lead developer on hyperscale-rs worked through the standard list of hard sharding problems and answered each one against Hyperscale's design, in public.
• The same channels argued out what Hyperscale keeps from Cerberus, the original Radix consensus paper, and what it does not.
• A proposal for penalising provable cheating would impound a validator's stake for a period rather than confiscate it.
• The Foundation said the DAO is expected to launch in about a month, with the permanent council seated after it.
• A four-part model for the handover circulated: milestone-based treasury releases, sub-DAOs, a legal wrapper, and three phases ending in the Foundation dissolving.
• Radix Seal shipped: signing documents, encrypting files and private chat, all from a wallet signature and all in the browser.

Most claims in this industry are made and never checked. The useful thing about this week is that two of Radix's were written out where anyone can mark them.

Answering the sharding questions

In December 2017 Vitalik Buterin published a Sharding FAQ, which became the standard catalogue of what makes splitting a blockchain into parallel slices hard: keeping a transaction that touches two shards all-or-nothing; making sure the data stays available; deciding which validators work on which shard; and stopping an attacker who takes over one shard from taking over the network. Nearly a decade later it is still the exam a sharded design has to sit. This week flightofthefox, who leads hyperscale-rs, went through it point by point against Hyperscale and posted the result for discussion; a community member reposted it to X, where it became the most amplified Radix post of the week.

The central answer is the one Radix has given since the beginning. Conventional sharding makes you choose between shards passing each other messages and hoping, executing optimistically and rolling back when it goes wrong, or fencing applications into separate walled gardens so the problem never arises. Hyperscale keeps transactions atomic across shards instead: a transaction declares up front which state it touches, every shard that owns any of that state takes part, and the whole thing commits everywhere or nowhere. The claim is not new. Setting it beside the specific objections and answering them in order is.

The channel did not treat any of it as settled. One long thread asked directly whether Dan Hughes's original consensus work had been abandoned. The answer that formed was that Hyperscale takes the principles of Cerberus, the consensus design Radix was founded on, and adapts them to the current implementation and the goals of the Xi'an release, rather than reusing the old code; and that the original paper starts from a fixed set of validators and never addresses what happens when validators join and leave a running network, which is the problem a live system actually has.

A second thread argued about penalties. If a validator can be proved to have cheated, by voting two ways on the same thing, should it lose its stake? The proposal that emerged was to impound the bonded stake for a fixed period instead: the operator loses the use of the capital rather than the capital itself, with the length of the impounding set by governance. Nothing there is decided.

The developer also started writing up how Hyperscale is tested, aimed at the Rust community rather than at Radix holders: the architecture that keeps protocol logic separate from network plumbing, so it can be run without a network at all; a simulator that replays an entire network deterministically from a single seed, so a failure can be reproduced exactly; and the Quint models that let the consensus rules be checked by machine. Putting it in front of readers who did not arrive already interested is the same move as the FAQ comparison.

The DAO gets a rough date

On the developer channel, with the argument running in parallel on RadixTalk, the Radix Foundation said the DAO is expected to launch in about a month, with the permanent Accountability Council established and identity-checked after that. A date, even a rough one, changes every governance conversation from someday to soon.

It also made the interim awkwardness concrete. The Foundation, in maintenance mode and with no marketing staff, spent part of the week deciding who should run its social accounts during the handover. The proposal was a narrow, vetted contract for a returning contractor to publish community-submitted posts under Foundation guidelines, because those accounts are Foundation property that will pass to the DAO and cannot be handed to an unvetted third party weeks before the transfer. It is a small question that shows the shape of the large one: an organisation designed to dissolve still has to hold the keys until the thing replacing it exists.

A widely shared sketch set out four parts for the transition: a treasury that releases funds against milestone votes rather than in a lump; sub-DAOs that divide the work into specialised units, so holders vote on leadership and budgets instead of every line item, which is the usual remedy for people not turning out to vote; a legal wrapper along the lines of an unincorporated association, so the DAO can sign real contracts such as exchange listings; and three phases of handover ending with the Foundation dissolving.

Underneath it sat a smaller and more immediate question. The volunteer who hosts Stokenet, the public test network, cannot keep paying for it indefinitely, and put continuation and compensation to a community vote. The framework that would decide such things is itself still in review.

A wallet that signs more than transactions

The clearest shipped work of the week was Radix Seal, released inside genkipool's community console. It uses ROLA, the scheme that lets a Radix wallet prove who holds it without a password, to do three things entirely in the browser: sign documents, encrypt files so only the recipient can read them, and hold a private one-to-one chat. The encryption keys are derived from a wallet signature, so no server ever holds the keys or the content, and a signature can optionally be recorded on the ledger as a non-transferable NFT to leave a permanent audit trail.

It is a concrete answer to something ROLA has always implied and rarely shown at this reach: what a wallet is for once it is treated as an identity rather than a way to spend money.

The week on the ledger

Read from the Radix Gateway as the ledger stood at midday on 26 July 2026, epoch 329,417, against the same reading a week earlier.

Stake up 51 million XRD on the week, and one validator off the registered list.

Concentration watch

Eight validators held a third of all staked XRD, unchanged, and 15 held half.

This week on the wiki

Read hyperscale-rs alongside the FAQ comparison, and Radix governance for where the DAO and the council sit relative to each other.

What to watch

• Whether the DAO's one-month estimate survives the legal review still in front of it.
• How the testing writeup lands with the Rust community, which did not come to be convinced.
• The Stokenet funding vote, and what happens to the test network if it fails.

Corrections

Nothing this week.

Next: whether the DAO date holds, and whether the sharding answers persuade anyone outside Radix.

Radix Week in Review series, Issue #5: ← Previous: Issue #4, Jul 13–19, 2026 · Next: Issue #6, Jul 27 – Aug 2, 2026 → · All recaps · Subscribe

Previously: The consensus half of the Rust rebuild is essentially built and formally verified, and after a week of worry, candidates began standing for the council.