Radix Week in Review: A Threshold With One Hand Behind It

At 18:25:32 on Saturday evening a transaction committed to a public Radix ledger that nobody sent and nobody paid for. It carries no signature and no intent hash, its fee reads zero, and the single thing it did was create a package called Pool. Forty-five minutes later a second one created AccountLocker. An hour and eight minutes after that, a third rewrote thirty entities at once. Those three commits are Anemone, Bottlenose and Cuttlefish, the protocol updates that Babylon mainnet took four hundred and forty-seven days to walk through, and the freshly reset Stokenet crossed all three between dinner and bedtime.

Each of them had to pass a threshold first. A Radix protocol update is not a date in a calendar or a switch an operator throws. It enacts at the start of an epoch, once validators holding 80% of the stake have signalled readiness for ten consecutive epochs. On Saturday every validator on that network answered to one person, and so the rule was satisfied the way a vote is carried in a room with one chair. The requirement is 80%. So I am pushing it through.

A threshold is a measuring instrument. It reports the weight leaning on it, and it reports nothing at all when every hand belongs to the same body. Three of this week's stories are thresholds, and in each one the interesting question is not where the line sits but how many hands can reach it.

Fifteen months of protocol history in one evening

Radix's public test network went down at 07:00 UTC on Saturday 29 August and came back at 12:52 on a fresh genesis with four validators, the first full reset in three years. Daffy, who has hosted most of the community-run network since the Radix Foundation handed it over, gave a day's notice in the developer chat and eleven days' notice in the main group. Every balance, package and transaction from before is gone. Account addresses, the network ID and the Gateway URL are not.

Keeping the network ID is what made the reset awkward. The genesis for network 2 is compiled into the node binary with no flag to override it, so reusing the name meant changing the binary: a two-line overlay replacing one 538 KB jar, with the rest left as the unmodified official release. The published install instructions no longer work without extra steps.

The deeper consequence surfaced two hours after relaunch. A developer published a package successfully at epoch 28, then found that anything referencing the AccountLocker native package came back ReferencedNodeDoesNotExist, as though the package were not on the ledger. It was not. AccountLocker arrived with Bottlenose, and the reset ledger had restarted at Babylon genesis, behind every protocol update Radix has ever shipped. A chain cannot skip to the head of its own history. It has to walk.

So it walked, and the walk is stamped. Reading the Gateway transaction stream filtered to the entities each update touches, the three flash transactions sit at state version 301,079 in epoch 105 (Anemone, 18:25:32 UTC), state version 337,082 in epoch 117 (Bottlenose, 19:10:30 UTC) and state version 391,138 in epoch 135 (Cuttlefish part 1, 20:18:15 UTC). All three are system transactions, all three carry a fee of zero, and all three land on round 3,000 of their epoch. The same three updates on mainnet sit at state versions 45,746,679, 92,004,925 and 185,552,770, in epochs 70,574, 105,352 and 160,922, dated 7 February 2024, 7 June 2024 and 18 December 2024. Mainnet spent 90,348 epochs and 315 days getting from Anemone to Cuttlefish. Stokenet spent 30 epochs and one hour fifty-two minutes.

The compression is not a performance result and it is not a bug. It is what the 80% rule reduces to when four validator nodes sit behind a single owner badge. The operator said so plainly in the developer chat when a colleague asked whether the validators had to signal at all: he had never handed over the last badge, so he held 100%, the requirement was 80%, and he was pushing it through with or without approval. He stopped at Cuttlefish part 1, which is unshipped on mainnet, on the grounds that enacting it was not his call to make. That last sentence is the whole of the week in one line: one person able to cross any threshold on the network, declining to cross one because nobody had asked them to.

The round numbers carry a second finding. Radix has no block time; it counts rounds, and an epoch finishes on whichever comes first, a five-minute target or a ceiling of 3,000 rounds (the protocol-update mechanism is documented here; the round band lives in the engine's consensus configuration). Between epochs 300 and 370 on Sunday morning, the reset Stokenet averaged 3,004 rounds per epoch and 219.6 seconds. It is hitting the ceiling, not the clock. Mainnet over the same morning averaged 1,601 rounds per epoch and 300.05 seconds, which is the five-minute target to a twentieth of a second. Four validators in close company finish rounds faster than the clock can spend them, and an empty ledger gives them nothing to slow down for. The operator expects epoch times to normalise back to five minutes as activity returns, which is the same statement read from the other end.

The DAO moves its papers out of a personal account

The second threshold of the week has real weight behind it and no way to reach it yet. On Friday the Radix Accountability Council reported that it and counsel had reviewed the MIDAO service agreement, proposed amendments, and had them accepted. Payment follows, then the submission of RADIX DAO LLC to the Marshall Islands government through MIDAO's onboarding platform, starting Monday. The Council was careful about what happens next: the filing has several steps, it is not yet at the point where the clock starts, and the certificate of formation takes the registry four to six weeks once it is.

The same day the governance repository moved off Daffy's personal GitHub account to github.com/RadixDAO, and radixdao.org was switched on as the DAO's official venue, with its deployment automated from that repository. Two notices went up under Notices and Records, both concerning the transition Council's certificate, including its technical details.

The governance index now lists twenty-seven documents. Twenty-six read "Signed, awaiting ratification", one is pending, and the page says in its own words that nothing there is in force. Each signed document is published as a PDF whose SHA-256 digest is recorded in ratification proposal GP-PRE-1. The phase label at the top reads Pre-formation. Ratification begins with a discussion period on the forum, and no date has been set for it.

This is the interesting inversion. Two weeks ago the Radix governance story was about authority that travels with a checkable mark. This week the marks are all in place and the hands are missing: a signed constitution nobody has voted on, a company with no certificate, a council whose permanent members have not been elected. Moving the repository out of one person's account changes nothing a checker can measure today, and it is the most consequential thing that happened to the DAO all week, because a threshold means something only when the thing behind it is not one person's laptop.

One contributor, 349 commits, and a fee table nobody has written

Across the five repositories this series tracks, the seven days to 30 August produced 349 commits, at least 5,886 lines added and 1,483 removed over 385 files. Every one of them is flightofthefox's. hyperscale-vm took 338 of them across five days; hyperscale-rs took 11, mostly bumping the VM it depends on. babylon-node, radixdlt-scrypto and radix-engine-toolkit recorded nothing at all. The network that reached 500,000 transactions a second in public testing is being rebuilt by one person, and the repositories carrying the name of the company that started it are still.

What that person was working through in the open on Thursday was fees, and specifically the thing Radix does not currently charge for. A developer asked why minting a million NFTs on Radix runs to tens of thousands of XRD in batches of twenty, and whether hyperscale-rs could lift the limit. The answer went past the limit to the accounting behind it.

Hyperscale will meet it with state bonding, closest in shape to Sui's among the networks that already do this. The reasoning offered alongside it is the part worth keeping: in a sharded system whose security rests on shuffling validators between shards, state has to move whenever the assignment changes, so bloating state is close to the most expensive thing a user can do to a network that never charged them for it. A transaction size ceiling of about 1 MB was defended in the same thread as an architectural decision rather than a limitation, on the grounds that sizing gossip, serialisation and block space around hundred-megabyte edge cases makes the system worse for everyone else. The Radix Engine's current answer, which is to price execution and storage at the moment of writing and nothing thereafter, is being replaced rather than tuned.

None of that is shipped, and the same developer said on the record this week that the VM is not yet safe to build against. A fee table is the last thing a chain designs and the first thing its users feel.

Fees with a date on them

The third threshold is the one with an epoch number attached. StakeSafe has fee rises queued on three of its validators: Amsterdam and Rotterdam from 15% to 25% at epochs 339,608 and 339,609, holding 76.04 and 81.16 million XRD between them, and its Seed Node from 15% to 100% at 339,609. The validator it Avaunt Staking bought on 21 August, now trading as ShardSpace.app, carries a rise from 2% to 25% at epoch 342,482 on 142.17 million XRD. Leaf Node goes from 1% to 100% at 341,223 and Apollo Pool from 20% to 100% at 342,116. Six pending changes, and the first of them lands today.

A fee change on Radix is announced to the ledger before it takes effect, which is exactly what makes it a threshold rather than an ambush: a delegator who reads the pending field has roughly a fortnight to move. Whether anyone reads it is the open question, and it is the one the numbers below will answer over the next month. Meanwhile 77 validators charge a fee that differs from the one stored in their own substate, 58 of them active and 2.88 billion XRD behind them, because the engine records a change as a request with an effective epoch and never rewrites the stored factor. Any dashboard reading the wrong field shows the old number forever.

CaviarNine's two validators, unmentioned in its departure notice, were still registered and still charging nothing at epoch 339,487 on Sunday, holding 113.72 and 115.61 million XRD.

This week on the wiki

Read Stokenet this week. The article now carries the reset as a dated operational record rather than a description of a service: what went down and when, what came back, what an account holder found in the wallet afterwards, and what a developer has to redo. It is the page that answers the question the developer chat kept asking all weekend, and the only place the reset exists as a single account rather than four hundred messages. The Network, Week by Week carries this week's on-chain reading and the ones before it, and Kangaderoo's page picked up the operator side of the same story.

Pages edited in the seven days to 30 August 2026:

The week on the ledger

Read live from the Radix Gateway at epoch 339,191, state version 556,706,020, on 30 August 2026.

The 349 is the number that carries this week, and Ociswap's 45% fall in seven-day volume is the one that will read differently in a month. Line and file totals are sampled from 71 of the 349 commits, so the lines figure is a floor.

The ledger

Nothing new came due this week. The three open claims were re-read at epoch 339,487 rather than left dated. StakeSafe's three queued fee rises are still queued and none has been withdrawn, with the first landing about ten hours after that reading. CaviarNine's two validators are still registered, still accepting stake and still charging zero. The Radix DAO governance index has gone from twenty-four documents to twenty-seven and none of them is in force, so that claim is untouched by the week's activity.

Two claims are recorded this week. The Marshall Islands issues a certificate of formation for RADIX DAO LLC by 25 October 2026, checkable against the Council's own notices and the registry. And hyperscale-vm records at least 200 commits in the fortnight from 31 August to 13 September, checkable with one GitHub API call against a baseline of 338 in the week just ended. Both, and everything scored before them, are in the running ledger.

Concentration watch

Eight validators hold a third of all staked XRD, unchanged on the week, and the largest ten hold 39.61%, down 0.3%. A third is the line that matters: a colluding group above it can stall the network, whatever anyone else does. The number has not moved in weeks, and the reason it is printed every week regardless is that the fee changes above are the mechanism that moves it. Six validators repricing at once is the largest single push on delegator behaviour the network has seen this year, and by the next reading it will have started.

What to watch

• Today, epochs 339,608 and 339,609. StakeSafe Amsterdam and Rotterdam go to 25% and the Seed Node to 100%. The check is effective_fee_factor.current on the Gateway validators list, not the stored validator_fee_factor, which will still read 0.15 afterwards.
• From Monday 31 August. The RADIX DAO LLC filing goes to the Marshall Islands through MIDAO. The registry takes four to six weeks from the point the submission is complete, so the certificate is an October event. The Council has said it will report advancements rather than steps.
• Whenever the ratification discussion opens. Twenty-seven signed documents and no vote date. The first observable move is a new topic on the forum, not a change on radixdao.org.

Corrections

Nothing this week.

Three flash transactions, twenty-seven unratified documents and one contributor are not obviously the same story, and the ledger says they are. Every rule Radix ran this week worked exactly as written. What none of them could supply is the thing a rule is written to count. The certificate arrives in October, the ratification vote gets a date when the Council sets one, and the first of the fee rises lands tonight. By then the question stops being who can reach the threshold and starts being who bothers to.

Radix Week in Review series, Issue #10: ← Previous: Issue #9, Aug 17–23, 2026 · All recaps · Subscribe

Previously: CaviarNine is leaving Radix, Avaunt sold its validator for 1,456,311 XRD in one atomic transaction, and Stokenet is deleted on Saturday week.