RADIX WikiRADIX Wiki

James Watt didn't invent the steam engine, he invented the part that stops it destroying itself. The centrifugal governor is two weights on hinged arms, spun by the engine's own shaft: as the engine gathers speed the weights fly outward and close the throttle. Without this, the engine would race, overheat and eventually break.

Radix spent this week gathering speed. The Rust rewrite of its consensus layer finished its first milestone. The execution engine that will run on top of it appeared as a public repository, several hundred files of Rust with the paint still wet. The consensus repository alone took 183 commits in seven days. Someone outside the core project pointed an AI coding agent at a new set of Radix instructions and had an escrow contract live on mainnet by the end of the afternoon. The shaft is turning faster.

Nothing is attached to it. The DAO's governance contract is deployed on mainnet, its addresses published this week, and every counter inside it reads zero. The Council that will steward the DAO has not been elected, and the body standing in for it is mandated to form the company, not to sign off protocol milestones. The wallet and the Gateway are in maintenance mode, waiting on a direction nobody has yet been asked to vote on. Fitting the linkage is the month's work.

The Machine Room

"M1 is done and dusted." The lead developer of hyperscale-rs wrote that on Friday morning, closing a milestone that covers validator lifecycle, the consensus engine, and dynamic sharding – shards that split and merge under load while the network keeps committing. The final piece was committee rotation, which reshuffles a shard's validator set one seat at a time, never wholesale; Monday's digest called it make-before-break, the phrase a lineman uses for joining the new wire before cutting the old. Two deferred items remain, chiefly the ability to split a shard on fee pressure, not on stored bytes, and they wait on the fee system. The fee system, in turn, waits on the layer above.

That layer is the week's real news. hyperscale-vm went from an intention stated in chat to a link posted in the channel on Friday evening: 259 files of Rust across twelve crates, 133 commits, dual-licensed Apache-2.0 and MIT, and a README that opens by telling you not to use it. Eight days from first commit to a public repository with a reference interpreter in it.

Its design thesis has a name – determinism by declaration – and a plain version underneath. On Radix today, the only reliable way to learn what a transaction touches is to execute it and watch. That is survivable on one shard. Across many it is fatal, because the network has to know which shards a transaction concerns before it schedules the thing. So the new engine moves the answer forward: a single pure function, route(), folds over a transaction's manifest and returns the shards involved, the state keys each one will touch, and the mode of every access – no execution, no state read. A mempool can schedule on that. A proposer can budget on it. The preview a Radix wallet already shows you before you sign comes from the same read, now with no execution behind it. If a contract then reaches for something it never declared, there is no handle for it to call and it traps.

The consequence sits in the locking. Whole-object locks are replaced by five access modes and a compatibility relation: reads share with reads, and increments and reservations commute, which means a thousand deposits into a single vault stop forming a queue and become one parallel group. Nothing is speculated and nothing is rolled back – conflict groups fall out of the declarations, and the resulting receipts are byte-identical whether the work ran serially, in parallel, or in an order chosen to be awkward. The whole profile is then executed twice, by wasmtime pinned to a version and by a reference interpreter written independently from the spec, with the two compared against each other. A divergence blocks the release, whichever side turns out to be wrong. The engine is being given a governor of its own, a check it cannot outrun, wired in by the people building it.

And Scrypto survives. The question of what happens to Radix's contract language under a new virtual machine had been the biggest unanswered one for builders since the Radix Engine was ruled out for sharding a week earlier. The answer came on Monday: Scrypto is "literally just a couple of Rust macros", parity is not hard, and the developer-facing furniture – manifests, resources, subintents, badges – is considered good and stays. What changes is underneath: state, locks, parallelism. Existing contracts face a recompile at best and an automated transpiler at worst.

What a Seat Will Cost

A validator seat is about to cost roughly what a Raspberry Pi costs, and that single fact settles who is allowed to run this network. Asked whether a Hyperscale node could live on a Pi 5 with an NVMe drive, the answer was almost certainly, and possibly several on the same board: under a tenth of one CPU core, roughly 200 MB of memory, and storage bounded by shard splits. A Babylon node today wants a server. The successor wants a hobbyist's board on a desk.

The penalties were drawn to match. There is no slashing anywhere in the design. Go offline or fall behind and your seat is jailed until you bring it back. Do something provably byzantine – double-vote, or issue two proposals for the same round, both of which require deliberately modifying the client – and your stake is impounded for ninety days, locked, not burned, with the period configurable by governance. The distinction matters most to the people who never touch a machine: a delegator who backed the wrong operator loses access to their stake for a season and keeps it. Delegation itself shifts too, from staking to a single validator toward staking to a pool that may run several nodes, and if more nodes turn up than the topology needs, the price of a seat rises until the surplus goes away.

Low hardware requirements, survivable penalties, and a seat priced by demand are the three dials that set how decentralised a proof-of-stake network becomes in practice, whatever its documentation claims. Every one of them is a governor on the network's behaviour, and all three are being set now, in a chat channel, by one developer answering questions on a Tuesday.

An Afternoon to Mainnet

On Thursday a former lead of Radix's dApp engineering team published everything he knew as a file an AI agent can read. radix-skills is 38 reference guides – transaction manifests, Scrypto, the Gateway, ROLA, subintents, pools, royalties, fees, costing, SBOR – bundled so that a coding agent clones the real Radix sources and works against them instead of hallucinating an API that never existed. This is the cheapest possible fix for the hardest problem a small ecosystem has: the model has read every line of Solidity ever written and almost no Scrypto.

The demonstration arrived the same day. A community member pointed an agent at the skill and came back about three hours later with Trade Vault, a person-to-person escrow, with the contracts deployed to mainnet and a wallet flow attached. It is a small thing, offered with a use-at-your-own-risk warning, and it duplicates functionality another project already has. That is beside the point. The binding constraint on the number of things built on Radix has always been the number of people who know how to build them, and this week that constraint moved. One of the builders in the channel drew the obvious next line: blueprints and royalties are a good shape for agents to use, since a well-made component is a tool an agent can call and pay for. This is speed arriving from outside the project altogether, from someone who needed no funding, no permission and no slot on a roadmap.

Three Counters, All Zero

Now the linkage. On 7 August the DAO's On-Chain Identifiers policy replaced its placeholders with real mainnet addresses, and because addresses can be checked, the wiki checked them. The governance component is live and running the blueprint the policy names. Its three counters – temperature checks, proposals, majority-judgment elections – all read zero. The contract every governance path in the framework routes through is deployed, correct, and has processed nothing.

Everything else this week rhymes with that reading. Asked whether he was in touch with the Accountability Council about signing off Milestone 2, the hyperscale-rs developer answered that there is no Council to be in touch with. A Transition RAC does exist and has not been idle: it retained counsel, drafted the DAO's Operating Agreement and Charter, and moved two proposals to a vote on 28 July. Its mandate is to form the company and take the Foundation's handover. The Permanent RAC, the elected body that would agree contracts for engine development, does not exist yet, and until a Consultation instance capable of running an election appeared on 4 August there was no machinery to elect it. In the developer channel, the question of what is coming in the wallet got the same shape of answer: nothing is coming, for the wallet or the Gateway, both are in maintenance mode until the community picks a direction and votes it through – and the path to picking one runs draft discussion, then temperature check, then proposal, with no concrete proposal currently in flight. The Foundation's blog has published nothing since it announced its own move to maintenance mode on 28 April. Asked on Saturday for an update from the interim Council, the channel produced one word: "Sooon".

Into that gap, on Saturday, a team published a governance design of its own: OTER XRD Governance, voting with a settlement layer built on an optimistic truth engine, three weeks in the making, with a hash-pinned tally document and an independent verifier. It drew real technical questions and one constitutional objection – that the good half of it could have arrived as a pull request against the Consultation app the DAO already has, without standing up a second protocol underneath the vote. Both positions were argued from an Operating Agreement that is still a pre-adoption draft.

The most telling conversation of the week happened on Sunday morning and had no announcement attached to it. Someone asked what is actually left in the treasury the DAO is about to inherit. Nobody could say. So the channel started costing out survival instead: how many validators does the network strictly need to keep producing blocks – four, they reckoned – and who would run them for nothing. The Gateway is paid for through the end of 2026. One member noted that he already runs five validators, two of them taken over when DeXter and HUG collapsed. An hour of arithmetic produced a workable answer, which is either the most reassuring thing that happened this week or the most alarming, depending on where you stand.

The activation vote is due within weeks, and the two numbers that matter to it – what the treasury holds, and what six months of running a DAO costs – are both still unwritten. The engine is the part nobody needs to worry about.

Sources

HydrateLast updated 10h agov1.4.06 revisions