Smart contract audits ask: did the agent execute correctly?
Reacher asks: which failure paths can the agent actually reach?
Different question. Different methodology. Different artifact.
Reachable failure paths in systems where verification claims are load-bearing. Autonomous agents, DeFi protocols, and the systems that allocate to them.
R/A/U state validation across multiple independent models. ERC-8004 signed attestations.
Verification, not coordination.
R
Reachable
Path exists from current state. Warrant chain complete.
A
Assumption-bounded
Path conditional on unverified premise. Resolves to R or U with direct evidence.
U
Unreachable
Architecturally excluded under protocol constraints.
What we validate
Autonomous agents
ERC-8004 deployments and trader-class agents whose claims of self-verification contain reachable self-licensing failure paths — where the decision logic can certify its own action correct without external check.
DeFi protocols where reachable failure paths hide in component interaction
Systems that pass line-by-line audits yet retain reachable bad states from how components, oracles, and automated decisions interact under stress — interest-rate, peg, and liquidation logic that each verify correctly in isolation but compose into states the audit was not built to surface.
The systems that allocate to them
Where verification is a load-bearing assumption in a thesis — not just in the protocol, but in the decision to back it — the same classification surfaces the failure paths before capital does.
Case study — JELLY
When a verification contract becomes a coordination contract under pressure
Incident: Hyperliquid JELLY (March 2025). Validators redefined an oracle in roughly two minutes.
Reading: A live demonstration of the inevitable self-licensing theorem at protocol scale — the same failure mode as an agent certifying its own action correct, but enforced by a validator set rather than a model.
Class: Reachable governance override of a path the protocol presented as verification-bound.