Pause first, match later
A Minted with no matching Locked pauses both gates from the same react(). A matching lock inside W reopens them. Matched traffic sends no callback.
Maya runs a lock-and-mint bridge from Base to Unichain, and neither chain can read the other's book. Ledgerlock is one Reactive contract that reads both and shuts both ends when a mint has no lock behind it: both gates shut by 8.245 s (p95, 100 testnet attacks), and 0 of 80 in-order transfers paused.
Route A · run bu-015 · forged mint, nothing locked
Minted, 500 tUSD. Nothing was locked on Base0x3a28…372dreact() finds no lock and requests two pause callbacks0xca7e…28d0Paused: releases frozen0x9994…3f38Paused: mints stop0x1c2f…087fGatePaused, one position after the pause0x0e0c…a6d4evidence/frozen3-bench-runs.csv + frozen3-triples.csv). Observer clock from the mint's receipt, polling public RPCs every 400 ms. bu-015 is one of 4 burst runs where no copycat landed; in the other 16, one did. All testnets; route A is our own fixture bridge.both gates shut, p95 of 100 attack runs
evidence/frozen3-bench-summary.mdin-order legitimate transfers paused
evidence/frozen3-bench-summary.mdcopycat mint landed before the gate shut, in each of 20 burst runs
evidence/frozen3-bench-runs.csvtriple rows joined from chain logs: 636 origin, 210 owner resolve(); 500 with a gate callback
evidence/frozen3-triples.csvScene 01 · What Maya reads at night
The status page reads public testnet RPCs, no wallet. Each route is one track; a lamp turns red only on a pause.
The status page on live testnet RPCs, captured 13 Nov 2026 (UTC); the attack-run strip, rendered from committed rows, re-captured 14 Nov. Its last non-control attack run is ap‑020.
# every bench number on this page, from evidence/frozen3-bench-runs.csv $ pnpm bench:summary {"attacks":100,"detection":{"n":100,"p50":3.873,"p95":4.391,"max":6.545,"min":2.036},"pause_both":{"n":100,"p50":5.92,"p95":8.245,"max":10.97,"min":3.851},"strict_false":0} $ forge test Ran 6 test suites in 4.66s (13.92s CPU time): 113 tests passed, 0 failed, 0 skipped (113 total tests)
Output verbatim, run 14 Nov 2026 (evidence/forge-test-2026-11-14.txt); the # line is ours.
Scene 02 · The bench
Seconds from the forged mint until both gates had shut: 60 forged keys, 20 plausible keys, 20 bursts. Nothing excluded.
detection p50 3.873 s · p95 4.391 sgate B p50 3.378 s · p95 7.079 sgate U p50 5.917 s · p95 8.245 son-chain check (Unichain block timestamps) p50 2 s · p95 3 s
A0, an unguarded copy of route A · 5 runs · per run
A, guarded · bu-021 to bu-025 · per run
Same attack script in both arms (one forged 500 tUSD mint); the unguarded arm fires a fixed 6 copycats, the guarded one up to 30, so issuance is an illustration, not a rate. Rows: evidence/a0-control-runs.csv, evidence/frozen3-bench-runs.csv.
Scene 03 · The interlock
A guard on Base never sees Minted; a guard on Unichain never sees Locked. Reactive Omni lets one contract on Lasna subscribe to both and answer on both.
A Minted with no matching Locked pauses both gates from the same react(). A matching lock inside W reopens them. Matched traffic sends no callback.
Locked on Base Sepolia, Minted on Unichain Sepolia, plus Burned and Released for the way back.
Forged keys pause once, then latch. Only the owner's resolve() reopens the route.
A failed callback is retried once, then reported as GateUnreachable.
Each gate flips the bridge's own pause switch; on route B, a Hyperlane PausableIsm.
Base Sepolia84532 · lock side · testnet
Lasna5318007 · Reactive Omni · testnet
Unichain Sepolia1301 · mint side · testnet
react(). Route A (our fixture) and route B (Hyperlane, self-delivered) each have their own guard and gates. As an image: architecture.png (source: architecture.html).Scene 04 · What it does not do
Mints that land before the pause are lost. The guard does not prevent the first one.README · Honest limits #2
Route A is our own fixture with a deliberately reproduced bug class. Route B is our own warp route andPausableIsm, delivered by our ownMailbox.process, not a third-party relayer.README · Honest limits #5
Nothing pages the operator. A latched route stays shut until someone looks; theIncidentLatchedandGateUnreachableevents are the signals to watch.README · Honest limits #8
All nine limits are in the README.
Questions a judge would ask
No. Every contract and transaction is on testnets: Reactive Lasna (chain 5318007), Base Sepolia (84532) and Unichain Sepolia (1301). "Live" on this page means deployed there and read by the status page. Nothing is deployed to Reactive mainnet or any other mainnet.
No. The forged mint is what trips the guard, so it always lands. The claim: loss is at most the mints inside the measured pause window. In 20 burst runs, at most one copycat landed per run.
A contract on Base never sees Unichain's Minted; one on Unichain never sees Base's Locked. One reactive contract on Lasna reads both and sends both pause callbacks from one react(). No off-chain watcher in the loop.
By our off-chain observer, polling public RPCs every 400 ms, from the forged mint's receipt to each log. On-chain cross-check, Unichain block timestamps from mint to gate U Paused: p50 2 s, p95 3 s.
Every transaction is real, on the three testnets. Route A is our own fixture bridge with a stolen relayer key reproduced on purpose. Route B is a Hyperlane warp route we deployed and deliver ourselves. Reorgs: simulation only (Test Lib).
The gates pause, then reopen by themselves when the matching Locked arrives inside W = 60 s. We forced 20: paused p50 15.2 s, p95 23.1 s (19 rows have a duration). Of 80 in normal order, none paused.
Yes, by design. A forger who picks a plausible nonce holds the route shut until the owner resolves it. Ledgerlock fails closed: legitimate mints wait.