USDe / evidence method

A claim needs a trail.

Keep the statement, its source and its limits together. A page that sounds certain is not a substitute for a check you can follow.

Different kinds of evidence

A document tells us what its author says. A source file shows behavior in a particular implementation. A deployment match connects source to code at an address. A configuration read records a value at a block. An economic measurement asks whether a position can actually absorb a loss or complete an exit.

These layers can disagree. Their dates can differ. A proposal is not a deployment, an audit is not a warranty, and a global redemption limit is not available inventory.

The control snapshot

The new report uses a supplied Ethereum snapshot from block 25,952,841 on 11 September 2026. Raw integers were decoded against the relevant ABI; source hashes and compiler metadata were checked; supplied explorer runtime fields were compared with supplied RPC code. Three core bundles report exact source matches. This is provider verification plus local cross-consistency, not an independent compiler rebuild or a consensus proof.

The Safe is a proxy. Its signature count does not reveal its implementation, optional authority paths or role in a controller. Unknown links stay visibly unknown in the control map.

Follow the whole operation

Check the caller, permission, asset movement, configurable bounds and failure path. Then inspect the relevant inherited code. A pause on issuer redemption may leave token transfers working; a restriction on staking shares may differ from releasing an already-converted silo claim. The exact action matters more than the button label.

Calculations remain scoped

Raw amount, decimal scale, block and output units travel together. The current control work uses integer decoding and small illustrative limit/nonce models. It does not present those models as executed Solidity tests, historical incidents, live trading simulations or measured loss probabilities.

History and corrections

Each report has its own date and stable route. Later evidence changes the overview and coverage, while the earlier report stays intact. Publication date and observation date are different. The curated control record exposes the current report's measured numbers and provenance boundaries.

What earns a stronger conclusion

Match the actual controller and Safe implementation; reconstruct bounded role history and verify memberships at the same block; identify delay exceptions; measure inventory and executable exits. Event lists alone are insufficient when a material setter lacks a dedicated event. Evidence of a protective mechanism must include the path on which it actually binds.