XI'AN · HYPERSCALE FOR RADIX

Hyperscale Weekly

Week #9 · 6–12 Jul 2026 · Milestone 1 · Adaptive Sharding · Week 11 · ~month 3 of 4

Split the work, not the world

MILESTONE 1 PROGRESS65%

Previously: flightofthefox had just closed the beacon's rare double-commit safety gap and proved the fix in five new formal models covering straddlers, cross-shard commit, and the dual commit itself.

This week the formal-proof work expanded dramatically: flightofthefox built a brand-new Quint model that walks the entire reshape lifecycle from start to finish. Alongside that, the validator shuffling got its first proper adversarial security analysis, and the simulator picked up a fresh batch of nasty network scenarios. Beacon and shard code keep getting tightened, with several pieces of safe-vote and ratification data now durably stored.

-33% vs last wk

commits

+13k −2.3k

-8% vs last wk

lines changed

158

files touched

34+

day commit streak

Commits / day

peak 16/day · 6/7 days active

Momentum

Lines changed / week

last 8 weeks

72% Rust12% Config12% Docs4% Specs

What flightofthefox built

Model F walks the full reshape story

flightofthefox built a new Quint model — 'Model F' — that proves the entire lifecycle of a shard split or merge: the merge path, the ready-signal hand-off, the draw-seed twins, and even the case of an unbound ready signal. This turns the reshape design promises into checkable math.

→ see “The Proof” on hyperscale.rs

The Lottery gets adversarial testing

The validator-shuffling logic was put under a Monte Carlo microscope, with an adaptive worst-case attacker thrown in. Deviations were measured and recorded, so the random committee assignment now has a published security case rather than just hopeful claims.

→ see “The Lottery” on hyperscale.rs

The Crash Lab grows new nightmares

Several fresh partition and Byzantine scenarios landed in the simulator — a beacon pool that stalls during a partition, inter-shard severance, fragment rejoin, exact-quorum heal, a slow proposer, and a stale parent. The Crash Lab now has more ways to break the network on purpose.

→ see “The Crash Lab” on hyperscale.rs

The bottom line

Standout change

A new Quint model now mechanically proves the entire reshape lifecycle — splits, merges, ready signals, and staffing — is safe under formal analysis.

What's next

Likely next: the validator-shuffling logic gets the production-side wiring, now that the security math and simulator scenarios are landing.

Heard in the chat

The mood: DAO and core-team chatter dominates the chat this week — who should run Radix, market-maker contracts expiring with no renewal in sight, and a small side thread about turning these weekly reports into

“finishing the roadmap is only conditional on the requisite fee being paid (and the coins themselves actually having liquid value). if the DAO keeps it's half of the bargain, then i'll keep mine.”

— flightofthefox, in the community Telegram

This week's concept

The Proof— Formal, mathematical verification

Model F is the centerpiece of the week: a single Quint model that mechanizes the whole reshape story — splits, merges, ready signals, staffing — turning the design into provable math.

hyperscale.rs/proof

Jargon, decoded

Monte Carlo — A way of stress-testing something by running huge numbers of random trials and watching the patterns.

Byzantine — Describes a participant in a network that is deliberately misbehaving or sending lies.

Adaptive adversary — A worst-case attacker that watches what's happening and changes its strategy on the fly to do the most damage.

Where the work landed

node · The Census

beacon ratification and safe-vote registers durably stored

shard · The Overlap

fork-safety storage, header scope, ready-signal dwell

simulation · The Crash Lab

new partition and Byzantine scenarios, harness cleanup