For judges · Reactive Omni Buildathon · Every Second Counts
Two chains, one book.
One Reactive contract reads Base locks and Unichain mints together and pauses a bridge seconds after a mint appears with no lock.
Testnet only (Reactive Lasna, Base Sepolia, Unichain Sepolia). No wallet, no login, no clone: every step below is a read-only explorer link. Route A is our own fixture bridge, labelled as one.
Click in this order
The 30-second path
Run bu-021 on the final deployment frozen3 (contract commit e425c78): a forged mint of 500 tUSD on Unichain Sepolia with no lock behind it on Base.
-
The forged mint lands on Unichain Sepolia.
A
Mintedevent with no matchingLockedon Base. A guard on Unichain alone cannot know that.
Unichain0x4bfe94f2…a84e -
One
react()on Lasna sees it, seen 3.175 s after the mint was mined.The same transaction emits two
CallbackRequestlogs, one per gate.
Lasna0xbacbf5a0…9068 -
Both gates are paused: Base at 2.184 s, Unichain at 8.062 s (observer clock).
Block timestamps: mint 1794595469, gate B 1794595470, gate U 1794595471. The Base gate is seen before the Lasna transaction, which we attribute to Lasna RPC visibility lag (not measured separately); block timestamps never invert.
Gate B · Base0x5eacdc82…4303 Gate U · Unichain0x60f4bcba…f3c4 -
The next copycat reverts
GatePaused, and the release on Base reverts too.The mint is in Unichain block 67743041, gate U's
Pausedin 67743043, and the reverted copycat in 67743044. Exposure: 1,000 tUSD (the forged mint plus one copycat that landed before the pause).
Copycat · reverted0x6006802e…2fb6 Release · reverted0x80ff77c4…9bfb -
Open the live status page.
It reads both gates' state straight from the public testnet RPCs and lists every committed run with its hashes.
Status/ledgerlock/status/
Measured live, committed as rows in evidence/
Receipt
p50 from the forged mint being mined to Lasna's react() being seen (p95 4.391 s, n=100)
p95 until both gates are seen shut (p50 5.92 s) over 100 attack runs
evidence/frozen3-bench-runs.csvstrict legitimate transfers paused by mistake
evidence/frozen3-bench-summary.mdbursts stopped at the first reverted copycat; at most 1 copycat landed
evidence/frozen3-burst-verify.csvgate callback legs joined to their Lasna react() across 846 triple rows; 0 CallbackFailure
on-chain cross-check, p50 / p95: Unichain block timestamps from the mint block to the gate's Paused block
| Arm | Copycats landed | tUSD issued per run | Collateral released |
|---|---|---|---|
| Unguarded copy of route A | 6 | 3,500 | 500 (succeeded) |
| Guarded by Ledgerlock | 1 | 1,000 | 0 (reverted) |
The arms use different copycat schedules, so issuance illustrates the effect rather than a like-for-like rate. Tests: 113 contract tests and 29 reactive-kit Solidity tests (100% lines, statements, branches and functions on src/ and the kit), 80 TypeScript unit tests on the reactive-kit helpers at 100%, and one property test that checks 262,144 random queue steps.
Real path first
Reproduce
Re-check the live runs on-chain (public RPC reads, no keys):
git clone --recurse-submodules https://github.com/edycutjong/ledgerlock.git && cd ledgerlock
pnpm install
pnpm burst:verify # re-reads the 25 burst rows (20 headline + 5 paired control) from the chains
pnpm bench:summary # regenerates the latency, false-pause and burst numbers from evidence/frozen3-bench-runs.csv
Re-run the attack bench live on your own deployment (funded testnet keys, about 25 lREACT):
export LEDGERLOCK_ENV=~/.config/<you>/testnet.env # deployer key + RPC URLs, outside the repo
INSTANCE=mine script/core-deploy.sh
script/bench-all.sh mine # 100 legit + 100 attacks, resumable
CI / deterministic replay (Reactive Test Lib, no network):
forge test && FOUNDRY_PROFILE=kit forge test && pnpm test:coverage
What it does not do
Honest limits
- The first unbacked mint is lost. Ledgerlock limits the loss to the mints that land inside the measured pause window; it does not prevent the first one.
- Both routes are ours. Route A is our own fixture bridge with a reproduced Nomad-class bug; route B is our own Hyperlane warp route, delivered by our own
Mailbox.processbecause the public relayer did not deliver within 6 minutes. - If Lasna stops, the guard is blind, and nothing pages the operator: a latched route stays shut until someone looks at the
IncidentLatchedandGateUnreachableevents.
All nine limits are in the README.