Radix Week in Review: Seven Validators Hold a Third of Stake

Week in brief

• hyperscale-rs merged five pull requests in 60 commits: shards now settle every open transaction before they split or merge, and validators sharing a machine share one copy of each shard.
• Staked $XRD fell 432m in a week, to 4.07B, and 880m sits unstaked and unclaimed in validators’ vaults.
• Seven validators now hold more than a third of the active set’s stake, one fewer than last week, although the seven largest barely grew.
• The five fee rises queued on 17 and 18 October, four of them to 100%, take effect around 1 and 2 November.
• The Radix Foundation has recovered 419k of the roughly 443k USDC stuck in the Hyperlane bridge, and holds the refund until the last 24.7k is back.
• OTER rotated the key behind its randomness beacon after its host disclosed a breach, and candidacy for the Permanent RAC closes on 30 October.

Almost all the code written for Radix this week went into hyperscale-rs, the Rust implementation of a sharded Radix network written by flightofthefox. Five merges dealt with one problem from several sides: a shard that changes shape, restarts or syncs must never lose track of a transaction it has already agreed to.

On mainnet, stake kept leaving validators in the middle of the table. The largest validators held what they had, so the number of validators needed to reach a third of all active stake, the share that can stall the network, fell from eight to seven without any of them growing much.

Hyperscale: reshaping without losing an answer

PR #173, merged on 19 October, makes a shard drain before it splits or merges. It stops taking new cross-shard work and waits until every transaction it has said yes to has its final answer, which takes at most 444 seconds when nothing is at fault: 120 seconds of transaction validity, 24 of finalization and 300 of grace. A reshape that cannot settle by its deadline is cancelled rather than forced. PR #176, merged on 22 October, follows from the same idea: a shard now deletes its record of an outgoing cross-shard transaction only after the receiving shard’s answer that it took the transaction arrives attested, signed by that shard’s validators.

PR #174, merged on 19 October, makes a validator check the certificate a peer cites before it starts syncing a block it does not hold. Before, a forged certificate for a height far above the chain could keep a validator in sync mode, proposing empty blocks, until it restarted. The same merge fixes the round clock and stops a failed fetch from retrying before its next tick.

Several seats, one copy of the shard

PR #177, merged on 24 October, is the week’s largest change at 20,440 lines added. One machine can host several validator identities, called vnodes, and until now each kept its own copy of every shard it served. Now the machine keeps one replica per shard, so vnodes on the same shard execute each transaction once, while each keeps its own signing key and voting records, written to disk before it signs. The design notes give the aim as a seat that costs close to its stake. PR #175 sped up the deterministic simulations that test all of this.

The previous week’s work is summarised in Hyperscale Weekly #23, which reports the iroh networking backend as nearly ready for production. The week’s other tracked repositories were quiet: hyperscale-vm took two maintenance commits, and radixdlt-scrypto and babylon-node took none.

Stake leaves the middle of the table

Nine validators lost more than a tenth of their stake between the readings of 18 and 25 October. None of the seven largest is among them.

Unstaked $XRD waits 2,016 epochs, about seven days, before its owner can claim it (how unstaking works). At epoch 352,577 the validators’ unstake vaults held 880m $XRD: everything unstaked and not yet claimed, whether or not its wait is over.

Why seven, when the seven did not grow

The line that matters for safety is a share, not an amount. Radix, like other Byzantine fault tolerant networks, stays live only while validators holding more than two thirds of the active set’s stake keep voting, so any group holding more than a third can stop it. When stake leaves the validators ranked eighth and below, the share held by the largest rises even if they gain nothing.

That is what happened. The seven largest held 1.34B $XRD on 18 October and 1.36B on 25 October, 1.5% more. The rest of the active set fell from 2.93B to 2.46B, 16% less. Their share of active stake went from 31.4% to 35.6%, so seven validators now pass the third that took eight a week ago. Read through the raw Gateway and through radixscan at epoch 352,577, state version 565,338,400; the two agree. The seven are Astrolescent, SRWA, Ocinode, Jazzer9F, DefiPlaza’s investment node, Radical Staking and Radix-Pipeline.

The bridge, the beacon and the council

The Foundation’s Andy.XRD said on 19 October that 419k USDC of the funds stuck mid-transfer since the August drain has been recovered, and that 24.7k is still held on Hyperlane’s Solana route after three days spent on it with Prime Vault and Hyperlane. Holders are due to get it back as hUSDC.

OTER, which publishes verifiable random numbers for Radix contracts, rotated the signing key of its beacon after its host, Railway, disclosed on 21 October that customer environment variables may have been read. OTER’s post says it found no sign of misuse; random.oter.io now signs a new chain called radixnet, and the old chain stays online until at least 24 October 2028 so that data timelocked to it can still be opened.

Candidacy for the Permanent RAC, the elected council of the Radix DAO, closes on 30 October. An admin of the Radix DLT group counted six candidates by 21 October; the transition council asked for 12 in its RadixTalk call. On Ociswap, seven-day volume doubled to 7.18m $XRD over 2,460 swaps.

The week on the ledger

Read live from the Radix Gateway at epoch 352,576, state version 565,338,197, on 2026-10-25.

The ledger

• Deferred: the Marshall Islands issues a certificate of formation for Radix DAO LLC, due today. radixdao.org/govern still lists the certificate as “Signed · not yet in force”, while its footer calls the LLC registered under Marshall Islands law, and nothing has been announced. Moved once, to 8 November.
• Still open: the five fee rises queued on 17 and 18 October, due 8 November.
• New: six or fewer validators hold more than a third of active stake in any weekly reading by 22 November. The six largest hold 32.6% today. Check: the stake-concentration count on the network page.

Every claim and its evidence: the Week in Review index.

This week on the wiki

Cerberus has a new section on how it commits a transaction across shards, written from sections 4 and 5 of the academic paper: each shard checks its own inputs, tells the others, and commits only if every shard had all of them. The blind signing page now explains why the Badger DAO theft could not work the same way on Radix.

Pages edited in the seven days to 2026-10-25:

Concentration watch

7 validators hold a third of active stake, against the one-third line where a group can stall the network. The six largest hold 32.6%, 26m $XRD short of it.

What to watch

• 30 October: Permanent RAC candidacy closes; the RadixTalk thread shows whether the count reached 12.
• About 1 and 2 November: Sirius, Polaris and RadixTalk reach 100% at epochs 354,383 and 354,385, Cadwynbloc 100% and Radstakes 75% at 354,603. The Gateway’s effective fee settles each.
• Hyperscale Weekly #24, for 19 to 25 October, is not out yet.

Corrections

Nothing this week.

Next week the fee rises land and the council candidacy closes.

Radix Week in Review series, Issue #18: ← Previous: Issue #17, Oct 12–18, 2026 · All recaps · Subscribe

Previously: Hyperscale merged a network layer for home routers, nine accounts unstaked 326m $XRD, and four validators queued fees of 100%.