PolicyProbe
Deterministic onchain policy assertions and repair findings for Hedera Harness.
Demo
About
PolicyProbe extends Hedera Harness with deterministic onchain behavioral verification. Developers can declare what an application is expected to do, execute real actions on Hedera, and compare the observed result against that expectation instead of treating a successful transaction as proof of correctness. For example, if an unverified investor must not receive a tokenized bond, PolicyProbe can execute that action, detect if it unexpectedly succeeds, capture the resulting transaction and state evidence, and emit a structured finding that Harness can feed back into its existing repair loop. The same probe can then be rerun after the fix to prove that the behavior now matches policy. The result is a reusable verification layer for AI-built Hedera applications: expected behavior → real execution → observed state → deterministic PASS/FAIL → repair. Our flagship demo uses a tokenized-asset policy scenario to make that loop concrete and auditable.
How it's made
PolicyProbe is a TypeScript contribution built on Hedera Harness and exercised against an Asset Tokenization Studio bond on Hedera testnet. It extends the Harness recipe schema with optional chain validation actors and assertions. The assertion runner provisions named signers, executes declared actions, resolves EVM transaction hashes or Hedera transaction IDs through Mirror Node, compares confirmed outcomes and optional exact balance deltas, and emits typed policy or infrastructure findings. The implementation stays generic and backward compatible: recipes without chain assertions follow the existing Harness flow. The companion fixture uses the same engine to test transfer restrictions, role boundaries, frozen holders, and pause controls. This repository includes focused tests, a reproducible fixture, architecture documentation, and linked HashScan evidence.