XI'AN · HYPERSCALE FOR RADIX

Hyperscale Weekly

Week #5 · 8–14 Jun 2026 · Milestone 1 · Adaptive Sharding · Week 7 · ~month 2 of 4
Split the work, not the world
MILESTONE 1 PROGRESS41%

Previously: flightofthefox had just closed a fork-safety gap by locking shard consensus into the HotStuff-2 safe-vote rule, and was driving a full epoch transition through the simulator with the new topology schedule.

A wide-ranging week. With the basic building blocks settled, flightofthefox turned to the messy middle: what happens to a cross-shard transaction whose shards are reshaped out from under it? The answer — a fence built on settled waves, a new way for shards to join via snapshot sync, and the first splits and merges actually running end-to-end in the simulator — is the meat of this week's digest.

161
+87% vs last wk
commits
+16k 3.8k
+136% vs last wk
lines changed
276
files touched
27
day commit streak
Commits / day
M
T
W
T
F
S
S
peak 42/day · 7/7 days active
Momentum
Lines changed / week
last 8 weeks
98% Rust2% Config
What flightofthefox built

A fence for stranded transactions

When a shard splits or merges, cross-shard transactions can be left hanging. flightofthefox built a settled-waves fence: the surviving side reconstructs what the disappeared shard had committed, and aborts what can't carry over. Nothing is silently lost.

→ see “The Will” on hyperscale.rs

New shards can snap into place

A new shard no longer has to replay history from scratch. flightofthefox added a snap-sync protocol and a BoundaryStore that pins each shard's state at every epoch crossing. A joining shard downloads a recent checkpoint and catches up fast.

→ see “The Archive” on hyperscale.rs

Splits and merges run for real

The simulator grew end-to-end scenarios: two sibling shards merging into their parent, a shard splitting live with observers tracking it, a validator relocating between shards. Adaptive sharding — the headline promise of M1 — is no longer just theory.

→ see “The Crash Lab” on hyperscale.rs
The bottom line
Standout change
The settled-waves fence is now in place, so cross-shard transactions survive a shard split or merge without getting lost or silently double-spent.
What's next
Likely next: stress-testing the reshape loop in the simulator with rougher scenarios — committee failures, validator drops mid-reshape, rotation stalls — before turning to the lottery-style validator shuffling also promised in M1.
This week's concept
The WillA fence for transactions stranded by a reshape
More than a third of the week's commits are about the settled-waves fence: the machinery that decides, safely and deterministically, what happens to cross-shard transactions when one of their shards disappears at a reshape.
hyperscale.rs
Jargon, decoded
StraddlerA transaction that touches more than one shard, so it needs both shards to commit it together.
Snap-syncA way for a new shard to download a recent checkpoint of state instead of replaying every block from the beginning.
Settled waveA batch of cross-shard transactions that a shard has fully committed and won't reconsider.
Where the work landed
simulation · The Crash Lab
split, merge, and relocate e2e tests
settled-waves fence machinery
per-shard boundary data and QCs
Reference — concept map, roadmap & links
The full picture
The System
The whole sharded design at a glance.
The Journey
One transaction's trip through the network.
The Clock
Consensus-attested time across independent shards.
The Census
The leaderless beacon of validators, stake and shards.
The Overlap
Why two conflicting blocks can never both commit.
The Generals
All-or-nothing cross-shard commits, computed not voted.
The Archive
Every published byte is preserved or provably expired.
The Library
All state in one merkle tree; a shard is a subtree.
The Will
How in-flight transactions settle when a shard dies.
The Lottery
Random, ever-moving committees; proven cheats are jailed.
The Triage
Under overload, urgent traffic never waits for bulk.
The Governor
Validator entry is re-priced every epoch, like a market.
The Crash Lab
The simulator that replays any failure byte-for-byte.
The Proof
Proving safety with maths, not just tests.
The Asterisks
Every design's trade-offs — including Hyperscale's own.
The road to mainnet
M1
Adaptive Sharding
~4 mo
M2
Radix Engine
~5 mo
M3
Gateway & API
~3 mo
M4
Validator GUI
~3 mo
M5
Migration
~3 mo
M6
Live support
12 mo
Now: Milestone 1 · Adaptive Sharding — Week 7 · ~month 2 of 4 · 41%. Since day one: 2,324 commits — one developer, fully in public.
Links
Written by AI from public commits, the community chat and hyperscale.rs — may contain mistakes. · generated 2026-07-09