Bridges
For the asset class with the worst historical loss severity, and the most conservative defaults in the product.
onboardBridgeDetectors in play12Boundaries declared3Refusals at onboarding1How this sector loses money.
Cross-chain bridge exploits now total roughly $2.9B, and in Q2 2026 alone they accounted for $351M. For most bridge designs the authority that authorises a mint is an off-chain validator set, so a compromise produces a mint that is correctly signed and completely fraudulent.
- 7Deterministic rules
- 3Invariants
- 2Models
What the preset configures.
One command, and this is its output. The identifiers at the foot are real, so you can diff them against the catalogue rather than trust the list.
onboardBridge(platform, config)
7 configured
- 01Every side of every asset declared, source and destination, across every chain
- 02Message correspondence: every destination mint must match a source lock or burn
- 03Mint caps per window per destination chain, as event invariants
- 04Atomic value conservation over escrow addresses at zero fee tolerance
- 05Supply-versus-backing as a state invariant across bridged assets
- 06Full authority surface on both legs: admin changes, EOA role grants, upgrade diffs, uninitialised proxies
- 07Fund-flow and peel-chain analysis on anything that does leave
- inv.bridge-message-correspondence
- inv.atomic-value-conservation
- invariants.event
- rules.admin-authority-change
- rules.role-granted-to-eoa
- rules.upgrade-implementation-diff
- rules.uninitialized-proxy
- rules.critical-selector
- rules.proportional-drain
- rules.multi-token-sweep
- ml.fundflow-anomaly
- ml.peel-chain-laundering
The perimeter, stated in writing.
Every edge, and the control that closes it. This is the list your auditor and underwriter will ask for, so it sits on the public page rather than an appendix.
An asset declared on a destination chain with no corresponding source side is rejected at onboarding, rather than accepted with a warning.
Left configured, every legitimate mint on that chain has no source leg to match and looks unbacked. The operator then raises the tolerance or mutes the rule, and the bridge ends up with less coverage than if it had never been onboarded. A configuration that predictably ends in a disabled control is broken, not conservative.
How it goes live.
Three rungs, in order, and each one written for this preset rather than for presets in general.
-
Step 1
Scope
Both sides of every asset, on every chain. A destination declared with no source side is rejected rather than accepted with a warning, which makes the declaration the longest part of the call.
-
Step 2
Detect
Every destination mint must match a source lock or burn. Supply against backing is the check that does not care whether the signature was valid, which is why it is mandatory in this preset.
-
Step 3
Prevent
The bounded action is a mint ceiling per window per destination chain. It gates only on chains whose observed contiguity we can vouch for, because reporting our own monitoring gap as a theft switches a control off.
What runs underneath.
Whole-chain verification on export is what makes correspondence defensible, because the claim is worth exactly what its coverage evidence is worth.
Scope Bridges coverage.
Every engagement is scoped by the team, and onboarding is performed by operators. There is no self-serve door, and that is a security decision first.