Canonical public semantics for dispatching bounded human work from AI agents.
task → claim → human execution → proof → verify → settle
This repository defines the interoperable protocol surface. It does not contain production secrets, operator personal data, or the private deployment configuration of ai2human.io.
- Task contract
- Proof contract
- Lifecycle state machine
- Funding and settlement invariants
- Security model
- AI2Human Improvement Proposals
- Architecture Decision Records
- Creation is not execution.
- Claim is not proof.
- System logs are not human evidence.
- Proof submission is not verification.
- Verification is not settlement.
- A displayed reward is not evidence of funding.
- Paid work must remain closed until funding is confirmed.
- Every terminal payment claim requires a durable receipt or transaction identifier.
preview — the semantics describe the live AI2Human workflow, while field names may still evolve through versioned proposals.
Breaking semantic changes require an AIP, compatibility analysis, migration guidance, and a version boundary.