Every result so far comes from one synthetic refund-operations domain: four collections, three task templates. The open question is whether the contract vocabulary and the Contract Agent transfer.
Proposal
Build a second in-memory domain (support ticketing or CRM record management) with its own state schema, tool registry, message policy and task templates, then compile contracts for it with the unchanged Contract Agent v3 prompt and report where the DSL falls short.
Rules
- Synthetic data only, seeded ids and timestamps.
- Every invalid fixture violates exactly one must-pass requirement.
- Gold data stays behind the
@stateproof/benchmark/gold boundary.
Every result so far comes from one synthetic refund-operations domain: four collections, three task templates. The open question is whether the contract vocabulary and the Contract Agent transfer.
Proposal
Build a second in-memory domain (support ticketing or CRM record management) with its own state schema, tool registry, message policy and task templates, then compile contracts for it with the unchanged Contract Agent v3 prompt and report where the DSL falls short.
Rules
@stateproof/benchmark/goldboundary.