Previously: flightofthefox introduced the substate sweep for garbage collection and added signed network headers to transaction intents, with the Radix VM exploit prompting his XRD liquidation and a migration-path sketch
This was the biggest week of Milestone 2 so far — flightofthefox landed a near-complete rewrite of how the VM executes transactions, folding every path through a single checkpoint called the core boundary. Meanwhile, the cross-shard settlement system, the substate sweep, and shard-termination rules all matured in parallel, and the Telegram chat erupted into a heated debate about what Hyperscale means for Radix's future.
230
same as last wk
commits
+19k −16k
+39% vs last wk
lines changed
Commits / day
peak 69/day · 5/7 days active
Momentum
Lines changed / week
last 8 weeks
84% Rust14% Config2% Docs
What flightofthefox built
Execution rewritten around one core
Every smart-contract execution path now runs through a single core boundary where modules are published, judged, and metered. The VM counts its own fuel per memory page, and code can only reach what it was explicitly lent — tightening security and aligning execution with sharding.
→ see “The System” on hyperscale.rsCross-shard ledger takes shape
Abandonment records, counterpart questions, and state probes were unified into one ledger with a single retention rule, replacing scattered ad-hoc checks and making cross-shard commitments cleaner and more provable.
→ see “The Generals” on hyperscale.rsSweep wired into block commits
The substate sweep was folded directly into the block commit path, and every transaction-derived artifact now carries a single expiry grace — so cleanup happens automatically as blocks finalize.
→ see “The Archive” on hyperscale.rsShard death rules hardened
Rules for what happens when a shard terminates mid-transaction were tightened: the settled set and the fence now govern every straddler, with terminal evidence sized to the shard's final span.
→ see “The Will” on hyperscale.rsFork safety tightened
Conflicting commits at the same block height are now refused outright, cross-chain claims must pass a single vote fence, and certificates from discarded rounds are evicted with their ticks — closing several fork-safety gaps.
→ see “The Overlap” on hyperscale.rsSim and CI scaled up
The test suite now runs on arm64 runners with increased parallelism, the simulator gained placement-seated validators bound to their hosts, and both storage backends share one conformance test helper.
→ see “The Crash Lab” on hyperscale.rsAlso this week
- Memory store snapshots now share structure, reducing duplication across nodes.
- The beacon header window is bounded at insert, preventing unbounded growth.
- Merkle proofs carry empty siblings as single bits, saving wire space.
- Block abandonment records are now bounded in bytes rather than by count.
- RocksDB store cloning was simplified by dropping its wrapper type.
- Resharding split logic now leaves pool room the committee shuffle cannot take.
- Four reshape assertions were fixed to actually test what they claimed to test.
- Departed shards' stores are now served and routing is seeded at boot.
The bottom line
Standout change
The VM execution model was rewritten around a single core boundary — the largest architectural change of Milestone 2 so far.
What's next
Likely next: continuing to wire the settlement ledger and the sweep into the execution lifecycle, and pushing the VM core boundary toward feature parity with Babylon's component model.
Heard in the chat
The mood: Heated debate dominated the chat about whether flightofthefox owes Radix a full migration, with frustration boiling over — but he reaffirmed he will help Radix adopt Hyperscale and noted the VM rewrit
“a grant is a beautiful trap: take fifty and wear the whole rap; if the flywheel won't spin they will ask "who cashed in?" - so i'll take the applause, not the clap”
— flightofthefox, in the community Telegram
flightofthefox explained
Why execution took longer
Leg local execution is an almost complete rewrite of how execution works, so the feature has taken longer to land than most.
Substate sweep vs GC
It's cleaning up storage, not memory — reclaiming space from guards that don't need to exist once validity windows for artifacts expire, like subintent nullifiers.
Session mode alternative
Shared-access components combined with subintents probably give as good a UX as sessions, without the potential foot-guns.
Migration harder than expected
He truly believed the existing Radix Engine would be more amenable to sharding — it was a rude awakening, and building a whole new VM was required.
This week's concept
The System— Split the work, not the worldThe core-boundary rewrite is the largest architectural change of Milestone 2 so far — it redefines how every transaction executes, how resources are metered, and how code is contained, all to make execution compatible with sharding.
hyperscale.rs Jargon, decoded
Core boundary — The single checkpoint where all smart-contract code is published, validated, and metered before it can run.
Abandonment record — A note that a transaction left certain state behind and that state is now safe to reclaim.
Meter — A built-in counter inside each module that limits how much computation it can consume, preventing runaway code.
Vote fence — A rule that blocks a block from committing until its claims about other chains have been verified by a vote.
Where the work landed
cross-shard ledger, settlement, and abandonment types
VM core boundary, metering, and reach limits
test harness, arm64 CI, conformance sharing
node-level routing, fetch, and commit paths
kernel imports, core ABI, and fuel wiring
Reference — concept map, roadmap & links▾
The full picture
The ClockConsensus-attested time across independent shards.
The CensusThe leaderless beacon of validators, stake and shards.
The OverlapWhy two conflicting blocks can never both commit.
The GeneralsAll-or-nothing cross-shard commits, computed not voted.
The ArchiveEvery published byte is preserved or provably expired.
The LibraryAll state in one merkle tree; a shard is a subtree.
The WillHow in-flight transactions settle when a shard dies.
The LotteryRandom, ever-moving committees; proven cheats are jailed.
The TriageUnder overload, urgent traffic never waits for bulk.
The GovernorValidator entry is re-priced every epoch, like a market.
The Crash LabThe simulator that replays any failure byte-for-byte.
The ProofProving safety with maths, not just tests.
The AsterisksEvery design's trade-offs — including Hyperscale's own.
The road to mainnet
✓ M1
Adaptive Sharding
~4 mo
Now: Milestone 2 · Radix Engine — Week 6 · since August 2026. Since day one: 4,464 commits — one developer, fully in public.
Links
Questions? flightofthefox is always happy to discuss in the community Telegram.
Written by AI from public commits, the community chat and hyperscale.rs — may contain mistakes. · generated 2026-09-14